339 corrections have been published since the log opened on Sep 6, 2026.

Vulcan's Operations block, Key figures, place in the Compare table, the Companies table and the Tiers list, and the research file the keyed API serves, read Not stated for its fleet efficiency and its hashrate, with the note "Not disclosed by the company in the documents read on Sep 6, 2026". They now give a fleet efficiency of 20.8 J/TH as of Jun 30, 2026, and a hashrate of 1 EH/s in the 10-Q for the quarter ended Jun 30, 2026, which gives no day for it. The hashrate is marked not comparable: the 10-Q gives it as the 6,500 miners associated with its cryptocurrency mining, within the combined capacity of its datacenter operations.

Vulcan's 10-Q for the quarter ended Jun 30, 2026 says "As of June 30, 2026, our fleet of miners ranged in efficiency from approximately 15.0 to 34.2 joules per terahash ("J/TH") and had an average efficiency of 20.8 J/TH." It also says "Our datacenter operations consist of approximately 23,500 miners with approximately 2.7 EH/s of combined capacity for both datacenter hosting and cryptocurrency mining, of which 17,000 miners, or 1.7 EH/s, are associated with our datacenter hosting and 6,500 miners, or 1 EH/s, are associated with our cryptocurrency mining."

Bit Digital's Operations block, Key figures, place in the Compare table, the Companies table and the Tiers list read Not stated for its fleet efficiency, with the note "Not disclosed by the company in the documents read on Sep 6, 2026". They now give 22 J/TH as of Dec 31, 2025, which the company words as approximate.

Bit Digital's results release of Mar 31, 2026 says "As of December 31, 2025, the Company's active hash rate was approximately 1.5 EH/s, with an average fleet efficiency of approximately 22 J/Th."

Keel's Operations block, Key figures, place in the Compare table, the Companies table and the research file the keyed API serves read Not stated for its hashrate, with the note "Not disclosed by the company in the documents read on Sep 6, 2026". They now give 14.8 EH/s as of Dec 31, 2025.

Keel's 10-K for 2025 says "As of December 31, 2025, we operated 113,649 ASIC miners with total hashrate of 14.8 EH/s."

The pairs page listed Fleet efficiency as "Not disclosed by the company in the documents read on Sep 6, 2026" for Bit Digital, Keel, DMG Blockchain Solutions, Argo and Vulcan, and Hashrate for Keel, DMG Blockchain Solutions and Vulcan. Bit Digital and Vulcan now have a fleet efficiency and Keel a hashrate, as the rows for their pages say, and the rest read "Not stated in the filings, releases or decks we read". Fleet efficiency counts 10 miners where it counted 8 and realized hashrate 11 where it counted 10 (Vulcan's hashrate is marked not comparable and is not counted). Five tests count one more miner: annualized revenue per MW against fleet efficiency (9 miners, r -0.15, p 0.697; before 8, r 0.13, p 0.760), realized hashrate against annualized revenue (11, r 0.56, p 0.072; before 10, r 0.53, p 0.112), log hashrate against log annualized revenue (11, p 0.033; before 10, p 0.028), hashrate against bitcoin held (11, p 0.067; before 10, p 0.096) and log hashrate against log bitcoin held (10, p 0.090; before 9, p 0.110).

The three figures that joined are the sentences quoted in the rows for Vulcan, Bit Digital and Keel; the counts and the tests are the pairs page's own calculation over the figures it reads, with those added.

The Fleet efficiency list ranked 7 miners and read "Not disclosed by the company in the documents read on Sep 6, 2026" in the cells of Bit Digital, Keel, DMG Blockchain Solutions, Argo and Vulcan. It ranks 9: Vulcan is 8th and Bit Digital 9th, and every other rank is out of 9. The cells of Keel, DMG Blockchain Solutions and Argo read "Not stated in the filings, releases or decks we read".

Vulcan's 10-Q for the quarter ended Jun 30, 2026 gives an average efficiency of 20.8 J/TH as of Jun 30, 2026, and Bit Digital's results release of Mar 31, 2026 an average fleet efficiency of approximately 22 J/Th as of Dec 31, 2025; the rows for their pages quote both sentences.

The notes on Argo's market capitalization and EV per EH/s said "Cash $2.2M, debt $40.2M, and 28.9B shares are on file", with no day for any of the three. They now say only that the research file carries no share price, so neither figure can be formed. The same two notes of the other five companies in the file lost the same list.

Argo's 20-F for 2025 gives cash and cash equivalents of 2,204 ($'000) as at 31 December 2025. Its 20-F for 2024 gives current loans and borrowings of 857 and non-current issued debt of 39,304 ($'000) as at 31 December 2024, which make the $40.2 million: the cash and the debt were of days a year apart.

The Bitcoin held list on the home page (the card behind the tile, the line under it and the card under the globe), the Bitcoin held row of the Compare page, the Bitcoin held tile on the Hut 8 and Cipher Digital pages and the Bitcoin held figures on the Analysis pairs page and the hover of Hut 8's bar of bitcoin held on the Charts page printed Hut 8's 17,316, Cipher Digital's 646 and Digi Power X's 217 as plain numbers. They now read about 17,316, about 646 and about 217, as the Key figures and the table of every miner's metrics already did.

Each 10-Q for the quarter ended Jun 30, 2026 gives the count as approximate. Hut 8: "As of June 30, 2026, we held approximately 17,316 Bitcoin". Cipher Digital: "The Company held approximately 646 and 1,433 bitcoin at June 30, 2026, and December 31, 2025, respectively." Digi Power X: "As of June 30, 2026, the Corporation held a total of approximately 217 Bitcoins".

Bitdeer's Key figures printed "As of Aug 31, 2026" for its hashrate and its power in development, "As of May 31, 2026" for the 50 MW its May update states, "As of Jun 30, 2026" for the 570 MW its June update states, and "Month ended Aug 31, 2026" for its bitcoin mined. They now read "As of August 2026", "As of May 2026", "As of June 2026" and "In August 2026".

Bitdeer's update for August 2026 gives its hashrate under the heading August 2026 ("Self-Mining (Operated in self-owned datacenters) | August 2026 | 79.9") and its bitcoin mined as "BTC Mined | August 2026 | 1,310"; none of its monthly updates gives the last day of the month for a figure. The Key figures already read "As of August 2026" for its bitcoin held.

The note under the Companies comparison and the notes under each company's radar chart said that the readings behind a median were "each as of the day its source gives", and ended their run of dates on "Aug 31, 2026", the day printed for Bitdeer's figures. They now say each reading is as of its own date, and end the run on the month where a source gives a month, "August 2026".

Bitdeer's update for August 2026 gives its hashrate and its pipeline under the heading August 2026 and no day, so the last day of that month is not a day its source gives.

The article printed "financing costs (none this quarter)" beside CleanSpark's cost with depreciation, and said "In its latest quarter CleanSpark mined at a cash margin on its direct cost and at a loss after depreciation". It now says "none in the quarter ended Jun 30, 2026" and that in that quarter CleanSpark's direct cost was below its revenue per bitcoin mined and its cost with depreciation above it, as Riot's was.

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 prints "Average revenue of each bitcoin mined (1) $71,692" against "Cost to mine one bitcoin - direct energy cost - Owned facilities $44,406" and "Direct cost to mine including non-cash depreciation and financing costs - Owned facilities $96,277", of which $51,871 is miner depreciation and none financing cost.

The article printed "financing costs (none this quarter)" for CleanSpark's definition. It now says "none in the quarter ended Jun 30, 2026".

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 gives $44,406 for energy and other direct costs and $51,871 for miner depreciation, which make the $96,277 of its direct cost including non-cash depreciation and financing costs, so no financing cost.

The worked table said CleanSpark's financing costs "were none this quarter". It now says they were none in the quarter ended Jun 30, 2026, the quarter of the table.

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 gives $44,406 for energy and other direct costs and $51,871 for miner depreciation, which make the $96,277 of its direct cost including non-cash depreciation and financing costs, so no financing cost.

The article gave CleanSpark's and Hut 8's revenue per energized megawatt, about $683,000 and about $422,000 a year, "in their latest quarters". It says "in the quarter ended Jun 30, 2026".

CleanSpark's results release of Aug 6, 2026 prints bitcoin mining revenue of $138,006 thousand for its quarter ended Jun 30, 2026, which times four over 808 MW is $683,198; Hut 8's results release of Aug 4, 2026 gives revenue of $74,932 thousand for its quarter ended Jun 30, 2026, which times four over 710 MW is $422,152.

The table and the chart of income tax expense or benefit said "each company's latest fiscal year", and the text said the IRS pages say what they say "today". They name the years (MARA's year to Dec 31, 2025, CleanSpark's year to Sep 30, 2025), and the sentence on the IRS pages and the sentence on when a deferred tax is recorded no longer say "today".

MARA's 10-K filed Mar 2, 2026 is for the year ended Dec 31, 2025 and CleanSpark's 10-K filed Nov 25, 2025 for the year ended Sep 30, 2025, as the rows of the table say.

The line under the chart of income tax expense or benefit named the two year ends itself: "MARA's year to Dec 31, 2025 and CleanSpark's year to Sep 30, 2025, as filed". It now reads "MARA and CleanSpark, the fiscal year each 10-K covers, as filed", since the line under a chart carries no date of its own. The chart's two rows and its note still give the year ends.

MARA's 10-K, filed Mar 2, 2026, says "For the fiscal year ended December 31, 2025" and CleanSpark's 10-K, filed Nov 25, 2025, says "For the fiscal year ended September 30, 2025", the two years the rows of the chart name.

The table row "latest 10-Q cover" now reads "10-Q cover page", with the two dates it already gave; the chart's line and period no longer say "the latest quarterly cover page" (its period gives the counts of Aug 4, 2026 and Jul 30, 2026), and the reading "MARA's rose every year and is still rising" says it rose every year to its 10-Q cover page of Jul 30, 2026.

CleanSpark's count on its 10-Q cover page of Aug 4, 2026 is 256,817,073 and MARA's on its cover page of Jul 30, 2026 is 386,299,297; MARA's year-end counts rise from 81,974,619 at the end of 2020 to 379,464,892 at the end of 2025.

The worked example said "using the latest quarter on file" and "The latest quarter's revenue". It names the quarter, 2026Q2.

MARA's 10-Q for the quarter ended Jun 30, 2026 gives revenue of $174,881 thousand and Riot's gives $174,235 thousand, the two figures the example divides.

Eight other articles named a period or a day by its distance from the day they are read: the article on a date, a source and a boundary ("filed last week"), on raising money and on dilution ("the next quarter's count", "the newest 424B5 or 10-Q", "today's price", "no dilution today"), on Form 4 ("the latest trade", "the company's next quarter"), on promised and delivered hashrate ("the current quarter's numbers come with the next quarter's expectations"), on the last halving ("a machine bought today"), on rate cuts ("worth 9.09 million today") and on why miners move more than bitcoin ("the newest 10-Q", "next quarter"). Each now names the period by what it is to the figures in the sentence (the quarter you are reading, the quarter that follows a sale, the date you are measuring to) or says it in general terms.

A period named as the latest, the newest or the next is the one on the day a page is read, not the one the filings date; the articles' figures are the filings' for the quarters and years they name.

Bitdeer's Key figures listed Cash and net debt under "Not stated in the filings, releases or decks we read", and so did its cell in the Companies table, while the note on its months of liquidity said "cash $456.8M" with no day. The tile and the cell now read $456.8M as of Jun 30, 2026, and the note reads "cash $456.8M at Jun 30, 2026".

Bitdeer's results release for the second quarter of 2026 gives, in its unaudited condensed consolidated balance sheets in US$ thousands, "Cash and cash equivalents 456,838" at Jun 30, 2026 against 149,352 at Dec 31, 2025.

Keel's Key figures listed Cash and net debt under "Not stated in the filings, releases or decks we read", and so did its cell in the Companies table, while the note on its months of liquidity said "Cash $768.9M is on file", with no day. That amount is its cash together with its restricted cash. The tile and the cell now read $715.5M as of Jun 30, 2026, its cash alone, and the note reads "Cash $715.5M at Jun 30, 2026".

Keel's 10-Q for the quarter ended Jun 30, 2026 gives cash of 715,516 in its balance sheet in US$ thousands, and in the table that reconciles the balance sheet to its statement of cash flows "Cash 715,516", "Restricted cash 53,377" and "Total cash and restricted cash 768,893" for the six months ended Jun 30, 2026.

The notes beside the Months of liquidity figures named a cash with no day: "cash $831.8M" for Cipher Digital, "cash $1.77B" for Core Scientific, "cash $2.62B" for TeraWulf, and "Cash $208M", "Cash $83.6M", "Cash $3.2M" and "Cash $2.2M" for HIVE, Bit Digital, Vulcan and Argo. Each now gives the day of the balance sheet it is from: Jun 30, 2026, except Argo's, which is Dec 31, 2025.

Each amount is the company's cash and cash equivalents in the balance sheet of its filing: the 10-Q for the quarter ended Jun 30, 2026 of each company but Argo, and Argo's 20-F for 2025, which gives "Cash and cash equivalents 2,204" in $'000 as at 31 December 2025.

The clause after the formula of each proxy for months of liquidity gave the quarter of the cost and not the day of the cash. For CleanSpark, Riot, MARA and IREN it now adds "the cash is as of Jun 30, 2026". IREN's formula printed its cash as 5,895,600,000; it now prints 5,895,591,000.

Each cash is the company's cash and cash equivalents in the balance sheet its filing gives for Jun 30, 2026: the 10-Q for that quarter of CleanSpark, Riot and MARA, and IREN's 10-K for the year ended Jun 30, 2026, which gives 5,895,591,000. IREN's cost in the formula is of the quarter ended Dec 31, 2025.

The keyed API's production figures for Keel dated 67 figures Dec 31, 2025. They came from three restated reports, all filed Mar 31, 2026: 26 from the report for the three months ended Mar 31, 2025, 24 from the one for the three and six months ended Jun 30, 2025 and 17 from the one for the three and nine months ended Sep 30, 2025. Of the 26, 25 were then dated to the three months ended Mar 31, 2025. The other 42 carried no day: the one figure a comparison introduces with "from", and every figure of the two longer reports, whose headings name a quarter and its year to date. That reading did not take the day a sentence gives with "at" either, as in "17.7 EH/s at June 30, 2025". 26 more figures, from Keel's filings of Mar 31, 2026, Nov 13, 2025 and Oct 10, 2025, lost a period their sentences do not state. The 62 December 2025 rows of the monthly figures, which came from the 67, are gone. A later row of Oct 10, 2026 for this page records that most of the 25 lost the first quarter of 2025 as well, and supersedes the counts given here.

The report for the three months ended Mar 31, 2025 opens with a notice whose first sentence reads "As of December 31, 2025, Bitfarms Ltd. ("Company") determined that it would prepare its financial statements in accordance with U.S. generally accepted accounting standards ("U.S. GAAP")." The notices of the other two reports begin the same way and name "accounting principles generally accepted in the United States of America". That is the day of the company's accounting decision, and the reading took it for the day of the figures that follow. The heading of each report gives its period, for example "BITFARMS LTD.'S RESTATED MANAGEMENT'S DISCUSSION AND ANALYSIS FOR THE THREE AND SIX MONTHS ENDED JUNE 30, 2025 AND JUNE 30, 2024". That report says in two paragraphs, in the same words, that revenue rose because of an average hashrate increase "of 5.8 EH/s, 2.2 EH/s and 1.6 EH/s" in Q2 2025 and "of 4.4 EH/s and 2.2 EH/s" in YTD Q2 2025, so a sentence of it that names no period cannot be put in either.

Keel's Key figures gave 170 MW under Power online and 210 MW under Power under contract, each as of Dec 31, 2025, and its row of the Companies table gave the same two. Both now read Not stated in the filings, releases or decks we read. The chart against the sector on Keel's page, which needs three figures, is gone: two remain.

The 170 MW is a sentence of Keel's restated report for the three and nine months ended Sep 30, 2025: "The Company operates a total of eight data centers with an aggregate energized power capacity of 170 MW located in the Province of Quebec, Canada." It counts the Canadian sites alone and gives no day. The 210 MW is a sentence of the restated report for the three months ended Mar 31, 2025: "The Company retains the option, subject to the receipt of additional government approvals, to expand the Rio Cuarto operation to the full contracted amount of 210 MW." It is an option at one site. Both were dated Dec 31, 2025, the day of the accounting decision that a notice before each report gives, and neither report gives that day to either figure.

The chart against the sector on a company page prints the sector median of each figure beside it, and the medians of Power online and Power under contract were taken over Keel's two figures. With them gone, the Power online median is 76 MW over 8 miners, where it was 102 MW over 9, and the Power under contract median is 700 MW over 5 miners, where it was 635 MW over 6. Keel's cells in the Companies table read Not stated.

Keel's 170 MW and 210 MW were sentences of its restated reports of 2025 that were dated Dec 31, 2025 by mistake; the row for Keel's page quotes both. A median is the middle of the figures the table prints, so it follows the table.

CleanSpark's Key figures read Not stated for its hashrate and its fleet efficiency, while its Operations block gave a hashrate of 38.6 EH/s for July 2026 and an efficiency of 15.8 J/TH. The Key figures, the Operations block and the Companies table now give the 10-Q's figures: 42.6 EH/s for June 2026 and 15.8 J/TH as of Jun 30, 2026. EV per EH/s, which divides by the hashrate, moved from $125.4 million to $113.6 million.

The 10-Q for the quarter ended Jun 30, 2026 says "During June 2026, our operating mining units produced an average computing power of 42.6 exahash per second ("EH/s")" and "The miners in service as of June 30, 2026 had a range of energy efficiency (joules per terahash - "J/TH") of 13.5 to 29.5 J/TH with an average operating energy efficiency of 15.8 J/TH." The 38.6 EH/s was the average for July from a newswire release that we cannot check against a filing.

Hut 8's Key figures read Not stated for its hashrate and its fleet efficiency, while its Operations block gave an efficiency of 14.1 J/TH. The Key figures, the Operations block and the Companies table now give a hashrate of 28.1 EH/s and an efficiency of 16.0 J/TH, both from its 10-Q for the quarter ended Jun 30, 2026.

The 10-Q says the miners delivered in April 2026 increased "American Bitcoin's total owned fleet capacity from approximately 25.1 to approximately 28.1 EH/s while improving overall portfolio efficiency from approximately 16.3 to approximately 16.0 J/TH." The 14.1 J/TH was the efficiency of American Bitcoin's operational fleet in a release of its own, a different company's filing.

MARA's Key figures read Not stated for its fleet efficiency, while its Operations block gave 18.6 J/TH as of Dec 31, 2025. The Key figures, the Operations block and the Companies table now give 17.3 J/TH as of Jun 30, 2026.

The 18.6 J/TH was the efficiency of about 35,000 new miners, not of the fleet: its letter for the fourth quarter of 2025 says "Deployed approximately 35,000 new miners with current energy efficiency of 18.6 joules per terahash ("J/TH") as of December 31, 2025." Its 10-Q for the quarter ended Jun 30, 2026 gives, as of June 30, 2026, "Miner efficiency (in joules per terahash) 17.3", which it defines as "The average number of joules of energy required to produce one terahash of computing power."

Riot's Operations block gave a fleet efficiency of 20.2 J/TH as of Mar 31, 2026. It now reads Not stated in the filings, releases or decks we read, as its Key figures do.

The 20.2 J/TH was read from a release on Riot's own website, which we cannot check against a filing, and none of its filings of the last year states a fleet efficiency: its second quarter 2025 deck says "Fleet Efficiency 21.2 J/TH", and no later filing gives one.

IREN's Operations block gave a fleet efficiency of 15 J/TH as of June 2025. It now reads Not stated in the filings, releases or decks we read, as its Key figures do. Its EV per EH/s formula printed its cash as 5895600000.0, beside the 5,895,591,000 of its months of liquidity; it prints the stored 5,895,591,000 now, with its debt, each as of Jun 30, 2026, and the figure moved from $832,066,396 to $832,066,784 per EH/s.

The 15 J/TH is the phrase "best-in-class efficiency (15 J/TH)" of its update for June 2025, over a year old, and none of its filings since gives a fleet efficiency. Its 10-K for the year ended Jun 30, 2026 gives cash and cash equivalents of 5,895,591,000.

Cipher's Operations block gave a fleet efficiency of 17.2 J/TH as of Aug 4, 2026. It now reads Not stated in the filings, releases or decks we read, as its Key figures do.

The ~17.2 J/TH stands on one slide of its second quarter 2026 deck, under the heading "Odessa Site Snapshot" beside "207 MW Total Capacity": it is the figure of one site, and the deck states none for the fleet.

Core Scientific's Key figures read Not stated for its hashrate, while its Operations block gave 15.7 EH/s. The Key figures and the Companies table now give 15.7 EH/s as of Dec 31, 2025. The note on its fleet efficiency said it was not disclosed in the FY2025 10-K or the 2026 10-Qs; it now adds that its 2025 10-Qs give 24.4, 24.7 and 24.6 J/TH for the quarters ended Mar 31, Jun 30 and Sep 30, 2025.

Its 10-K for 2025 says "As of December 31, 2025, we deployed approximately 151,400 bitcoin miners, which number consists of approximately 135,500 self-miners and approximately 15,900 hosted miners, which represented 15.7 exahash per second ("EH/s") and 2.2 EH/s for self-miners and hosted miners, respectively." Its 10-Q for the first quarter of 2025 says "Our average self-mining fleet energy efficiency for the three months ended March 31, 2025 was 24.4 joules per terahash"; its 10-Qs for the second and third quarters give 24.7 and 24.6.

TeraWulf's Key figures read Not stated for its hashrate and its fleet efficiency, while its Operations block gave 9.3 EH/s and 17.5 J/TH. The Key figures and the Companies table now give both as of Dec 31, 2025.

Its 10-K for 2025 gives, as of December 31, 2025, "Miner efficiency (w/th) 17.5" and "TeraWulf operational hashrate (EH/s) 9.3". The efficiency is in watts per terahash, which is the same number as joules per terahash.

Bitdeer's Key figures read Not stated for its fleet efficiency, while its Operations block gave 15.8 J/TH. The Key figures and the Companies table now give 15.8 J/TH, the average of the quarter ended Jun 30, 2026.

Its results release for the second quarter of 2026 gives, in its operational summary for the three months ended June 30, "Average miner efficiency (J/TH) 15.8", against 25.7 for 2025.

HIVE's Key figures read Not stated for its fleet efficiency, while its Operations block gave 16.5 J/TH. The Key figures and the Companies table now give 16.5 J/TH as of Jul 31, 2026, marked not comparable since the company gives it as implied by the nameplate.

Its 10-Q for the quarter ended Jun 30, 2026 says "As of July 31, 2026, the Company's total installed hashrate was approximately 25.2 EH/s with an implied efficiency of 16.5 J/TH, based on the nameplate hashrate and power consumption of the installed miners."

Argo's Key figures read Not stated for its hashrate, while its Operations block gave 2.4 EH/s. The Key figures and the Companies table now give 2.4 EH/s as of Dec 31, 2025, marked not comparable since it is what the fleet can generate, not hashrate realized.

Its 20-F for 2025 says "As of December 31, 2025, we had a fleet of approximately 22,600 machines mining Bitcoin that can generate up to 2.4 exahash per second of computing power."

The Fleet efficiency list covered 9 miners and ranked 7. It covers 7 and ranks 6: CleanSpark and Bitdeer tie for 1st at 15.8 J/TH, MARA is 3rd at 17.3 (its 18.6 had no rank), TeraWulf 4th, Vulcan 5th and Bit Digital 6th. Hut 8's figure is 16.0 J/TH (it was 14.1) and still has no rank. Riot, IREN and Cipher read Not stated in the filings, releases or decks we read: Riot's 20.2 and Cipher's 17.2 J/TH had ranked, and IREN's 15 J/TH of June 2025 had read as not current. The note that MARA's figure was the efficiency of about 35,000 new miners goes with that figure, and Hut 8's note now says its 16.0 is the portfolio efficiency of its subsidiary American Bitcoin.

The 14.1, 18.6, 20.2, 15 and 17.2 J/TH the list gave for Hut 8, MARA, Riot, IREN and Cipher were not each company's fleet figure in a filing of the last year; the rows for their pages quote the sentences that replace the first two and say why the other three go.

The Hashrate column of the Companies table gave figures for 10 miners and gives 15: CleanSpark, Core Scientific, Hut 8, TeraWulf and Argo joined it. Its Fleet efficiency column gained CleanSpark, MARA, Hut 8, TeraWulf, Bitdeer and HIVE. The sector median of hashrate that the charts on the company pages print is 15.7 EH/s over 15 miners, where it was 19.4 EH/s over 10.

Each figure is the sentence quoted in the row for its company's page.

The tests on the pairs page that use hashrate, EV per EH/s or fleet efficiency moved with the figures above. Revenue against hashrate (11 miners): r 0.58, p 0.063 where it read r 0.56, p 0.072, and in logs r 0.65, p 0.032 where r 0.64, p 0.033. Market cap against EV per EH/s (6 miners): p 0.187 where p 0.190, and in logs r 0.81, p 0.052 where r 0.80, p 0.058. Market cap against hashrate (9 miners): r -0.49, p 0.184 where r -0.47, p 0.203, and in logs r -0.50, p 0.172 where r -0.48, p 0.189. Bitcoin held against hashrate (11 miners): r 0.58, p 0.064 where r 0.57, p 0.067, and in logs (10 miners) r 0.57, p 0.087 where r 0.56, p 0.090. Fleet efficiency against revenue per MW counts 6 miners where it counted 9: r -0.21, p 0.685 where r -0.15, p 0.697.

The pairs are the page's own calculation over the figures it reads; the figures that moved are the ones in the rows for CleanSpark, Hut 8, MARA, Riot, IREN and Cipher.

The keyed API's production figures for Keel gave 25 figures of the report and the statements for the three months ended Mar 31, 2025 that quarter as their period, among them a sale of up to 1,000 Bitcoin authorised in 2022, a target of 21 EH/s for Jun 30, 2025, a plan to raise the hashrate under management up to 1.6 EH/s and what a site could support, over 7.0 EH/s. They now carry no day. A level that its sentence dates with "at" keeps that day: 12.8 EH/s at Dec 31, 2024, 19.5 EH/s at Mar 31, 2025 and 17.7 EH/s at Jun 30, 2025, which carried none or the quarter. The counts of bitcoin mined, 693 and 718, carry the first and the second quarter of 2025, the quarters of the headings of their sections. The monthly figures gain four rows from those levels, for Dec 2024, Mar 2025 (two) and Jun 2025. This supersedes the counts of the first row of the day for this page.

The report for the three months ended Mar 31, 2025 says "On August 1, 2022, Management received approval from the Board to sell daily production, in addition to any sale of up to 1,000 Bitcoin from treasury, should market conditions and the Company's projected financing requirements justify such sales in Management's discretion", "is executing towards its target of 21 EH/s by June 30, 2025", "the Mining operations are expected to increase the online Hashrate under management up to 1.6 EH/s" and "the site could support over 7.0 EH/s with current generation Miners". A report that names its span also states plans, authorisations and capacities of other days, so a sentence of it that names no period is not put in the span. The same report says "Increased Hashrate under Management from 12.8 EH/s at December 31, 2024 to 19.5 EH/s at March 31, 2025" and, under the heading "4. FIRST QUARTER 2025 FINANCIAL RESULTS AND OPERATIONAL HIGHLIGHTS", "Earned 693 Bitcoin at an average direct cost of $47,800". The report for the three and six months ended Jun 30, 2025 says "Decreased Hashrate under Management from 19.5 EH/s at March 31, 2025 to 17.7 EH/s at June 30, 2025" and, under "4. SECOND QUARTER 2025 FINANCIAL RESULTS AND OPERATIONAL HIGHLIGHTS", "Earned 718 Bitcoin at an average direct cost of $48,200".

The keyed API's production figures for HIVE gave 78 figures of two management's discussion reports a span they do not state: 40 figures of the report filed Nov 17, 2025 read as the six months ended Sep 30, 2025 and 38 figures of the report filed Feb 17, 2026 read as the nine months ended Dec 31, 2025. Both reports cover a quarter and its year to date, so a sentence that names no period is not put in either: 74 of the figures now carry no period. The four counts of bitcoin mined carry the span of the heading of their section: 718 for the three months ended Sep 30, 2025 and 884 for the three months ended Dec 31, 2025, which had been read as the longer spans, and 1,125 for the six months ended Sep 30, 2025 and 2,009 for the nine months ended Dec 31, 2025.

The first report says it is "on a consolidated basis for the three and six months ended September 30, 2025", and under the heading "RESULTS FOR THE THREE MONTHS ENDED SEPTEMBER 30, 2025 COMPARED TO THE THREE MONTHS ENDED SEPTEMBER 30, 2024" it says "The Company mined 718 BTC compared to 340 BTC in the comparative prior period." Under "RESULTS FOR THE SIX MONTHS ENDED SEPTEMBER 30, 2025 COMPARED TO THE SIX MONTHS ENDED SEPTEMBER 30, 2024" it says "The Company mined 1,125 BTC compared to 789 BTC in the comparative prior period." The second report is "for the three and nine months ended December 31, 2025" and says "The Company mined 884 BTC compared to 322 BTC in the comparative prior period." under the heading for the three months and "The Company mined 2,009 BTC compared to 1,111 BTC in the comparative prior period." under the heading for the nine months.

The keyed API's production figures for Digi Power X gave the nine months ended Sep 30, 2025 to 13 figures of its filing of Nov 14, 2025, among them the 10 bitcoin it mined in the three months and a target of 195 MW for 2027. They now carry no period. Four figures of two releases of Oct 21 and Dec 3, 2025 read as a month or a quarter their filing days had not reached, March 2026 and the first quarter of 2026, and now carry none; the monthly figures lose their row for Mar 2026.

The management's discussion for the three and nine months ended Sep 30, 2025 says "For the three-month period ended September 30, 2025, Digi Power mined a total of 10 Bitcoins compared to 9 for the three-month period ended September 30, 2024": the count is the quarter's, and it was read as the nine months. The material change report of the filing attaches a release of "unaudited financial results for the three and nine months ended September 30, 2025" and states "2027 Operational Target: 195 MW total, including 140 MW critical IT load." A target for 2027 is no figure of either span. The release of Oct 21, 2025 says five new modules are "scheduled to come online at its Columbiana, Alabama facility in March 2026", and the release of Dec 3, 2025 is headed "Digi Power X Implementing First B200 GPU Cluster in Alabama with Plans to Begin Data Processing in Q1 2026": a plan is no period of the release.

CleanSpark's page printed its cost per bitcoin of $96,277 as "All-in cost per bitcoin" and "All-in cost per bitcoin mined", and the Key figures tile gave the period only as "As of Jun 30, 2026". It now says "Cost per bitcoin with depreciation" and "Cost per bitcoin mined, as the company states it", the tile reads "Quarter ended Jun 30, 2026", and the hover text gives CleanSpark's own definition and what the quarter's figure is made of. The number did not change.

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 gives $96,277 as the direct cost to mine one bitcoin including non-cash depreciation and financing costs at its owned facilities. The table adds energy and other direct costs of $44,406 to depreciation of $51,871, with no financing costs in the quarter. Payroll, professional fees and general and administrative expenses are separate lines of the 10-Q and are not in it, and the site's all-in cost counts them, so "all-in" said more than the figure counts.

Riot Platforms' page printed its cost per bitcoin of $90,631 as "All-in cost per bitcoin" and "All-in cost per bitcoin mined"; the Key figures tile said "Includes depreciation, excludes SG&A and interest" and gave the period only as "As of Jun 30, 2026". It now says "Cost per bitcoin with depreciation"; the tile reads "Includes depreciation, excludes SG&A" and "Quarter ended Jun 30, 2026"; and the hover text says the figure is net of $10.1 million of power curtailment credits. The number did not change.

Riot's results release for the second quarter of 2026 builds $90,631 from the cost of power and other direct costs of self-mining, less $10.1 million of power curtailment credits, plus $64.6 million of miner depreciation, over 1,587 bitcoin. Selling, general and administrative expense is a separate line and is not in it, and the site's all-in cost counts it, so "all-in" said more than the figure counts. The release does not say the figure leaves out interest; it says only that costs to finance the purchase of miners were zero, so "and interest" claimed more than the filing says. No earlier word said the figure is net of credits.

TeraWulf's page printed its cost per bitcoin of $95,611 as "All-in cost per bitcoin" and "All-in cost per bitcoin mined", and its Key figures tile and its Companies comparison cell gave "as of Dec 31, 2025", the same words as a quarter's figure. It now says "Cost per bitcoin with depreciation"; the tile says "12 months ended Dec 31, 2025", and the comparison cell shows "12 months" beside the figure and gives "12 months ended Dec 31, 2025" in its hover text; and the hover text says the 10-K's own cost to mine one bitcoin is $53,681, a cash cost, and that $95,611 is that cost plus $41,930 of depreciation. The number did not change. A later row of Oct 9, 2026 for this page records the move to TeraWulf's figure for the quarter ended Jun 30, 2026, and supersedes the figure and the period given here.

$95,611 is the total for the year 2025 that a footnote of TeraWulf's 10-K gives for what its cost to mine one bitcoin would be if depreciation of the miner fleet were added to the $53,681 the table states. Our data filed it as a quarter, so no cell gave its period, and "all-in" said more than it counts: the table has no general and administrative expense, and the filing says costs related to raising capital are not included.

MARA's page said in the hover text on its cost per bitcoin ($38,690) that hosted sites "pay all-in hosting fees", gave the period as "As of Jun 30, 2026", and described its direct cost card as "Method A: power and hosting costs divided by bitcoin mined". The hover text now gives MARA's own definition, the amounts paid to power providers for power consumed at its owned sites, and says that third-party hosting and other energy costs are a separate line and not in it. The tile reads "Quarter ended Jun 30, 2026", and the card no longer calls the figure power and hosting. The number did not change.

MARA's 10-Q for the second quarter of 2026 defines purchased energy costs per BTC as the amounts paid to power providers for power consumed, divided by the bitcoin produced at its owned mining operations, which were 1,260 of the 2,422 bitcoin it produced in the quarter. The $69.2 million line is called third-party hosting and other energy costs and includes energy for mining other digital assets; it is not a set of "all-in hosting fees", and hosting is not in the figure.

IREN's page said in the hover text on its cost per bitcoin ($34,400) "Company-reported purchased energy cost per bitcoin: power paid to providers only", which is MARA's definition, and described its direct cost card as "Method A: power and hosting costs divided by bitcoin mined". The hover text now gives IREN's own: net electricity costs per Bitcoin mined, a non-GAAP measure, after demand response program income, plus the program's fees, and after taking out the electricity of its AI Cloud Services, and without renewable energy certificates. The tile reads "Quarter ended Dec 31, 2025". The number did not change.

IREN's 10-Q for the quarter ended Dec 31, 2025 defines net electricity costs as electricity charges, less demand response program income, plus its fees, adjusted for realized gains or losses on electricity financial assets, less the electricity of AI Cloud Services, and excluding renewable energy certificates: $57.2 million over 1,664 Bitcoin mined, printed as 34.4 in thousands of dollars. That is not the amount paid to power providers, it nets $1.8 million of demand response income, and it is not power and hosting.

Cipher Digital's page described its cost per bitcoin mined ($34,189 for the third quarter of 2025) as "All-in electricity cost per bitcoin: power only, not all operating costs", and its research row said Cipher "reports no all-in cost metric". The card now gives Cipher's own definition: the combined electricity cost of its mining sites, including transmission, settlement and other charges, net of revenue from opportunistic power sales. The research row is named "Cost per bitcoin with depreciation" and says Cipher states no cost per bitcoin that counts depreciation. The number did not change.

Footnote 3 of Cipher's third-quarter 2025 business update gives the figure as the reconciled combined electricity cost of Odessa, Black Pearl, Alborz, Bear and Chief, including TDU charges, taxes, settlement charges, TDSP charges, customer charges, contract charges and a 2021 storm surcharge, and net of revenue generated from opportunistic power sales. The old words left out that it is net of power-sale revenue, and the old research row used the site's name for a cost that counts depreciation and general and administrative expense.

Bitdeer's research row for cost per bitcoin said "Derived all-in cost of revenue" for an arithmetic figure of about $74,200. It now says "Derived total cost of revenue", under the row name "Cost per bitcoin with depreciation". The figure is our own arithmetic, not a company figure, and did not change.

"All-in" is the name of the site's own cost, which counts general and administrative expense. The derived figure is cost of revenue including depreciation and stock compensation, and leaves that expense out, so the word said more than the arithmetic counts.

On the Companies page the cost chart and the cost table called CleanSpark's, Riot's and TeraWulf's cost per bitcoin "All-in" and the direct figures "Direct: power and hosting", and the comparison table gave TeraWulf's 12 months and the others' quarters no period. They now say "With depreciation" and "Direct, before depreciation"; every cost figure in the comparison prints 3 months or 12 months beside the number; a new note under the table says the column covers 3 months or 12 months; the note above it lists five sets of words, since Riot's no longer says it leaves out interest; and the reading under the cost table says each figure with depreciation leaves out SG&A instead of saying each states it. The numbers did not change.

None of the three figures called all-in counts general and administrative expense, which the site's all-in cost counts. The direct figures are not power and hosting: MARA's and IREN's are electricity only, and the others count the other direct costs their tables list. TeraWulf's is the year 2025, CleanSpark's, Riot's and MARA's are the quarter ended Jun 30, 2026 and IREN's is the quarter ended Dec 31, 2025. And no filing says in so many words that it excludes SG&A: the tables simply do not include it.

The article called CleanSpark's, Riot's and TeraWulf's costs "all-in", said Riot's figure leaves out interest, described MARA's hosting as fees "not expressed per bitcoin", and said TeraWulf's figure was "not updated in the 2026 10-Qs". It now says "cost with depreciation", no longer says Riot's leaves out interest, describes MARA's and IREN's figures in their filings' words, and says TeraWulf's 10-Q for the quarter ended Jun 30, 2026 prints newer figures. The numbers in the tables and the chart did not change; the chart's series is named "With depreciation" and its line says each company states its own.

The three figures count depreciation and not general and administrative expense, so "all-in", which the site keeps for its own cost that counts it, said more than they count. Riot's release does not say it leaves out interest. TeraWulf's 10-Q for the quarter ended Jun 30, 2026 prints a cost to mine of $44,547 and $97,486 with depreciation for that quarter, so "not updated" was wrong.

The article's section, chart and steps called CleanSpark's $96,277 and Riot's $90,631 "all-in", and said the filing attributes CleanSpark's rise from $86,428 to $96,277 to depreciation and financing. They now say "with depreciation", and the article says depreciation per bitcoin rose from $41,601 to $51,871 while financing costs fell from $21 to none. The numbers did not change.

CleanSpark's 10-Q gives both quarters' figures in one table: energy and other direct costs of $44,806 and $44,406, depreciation of $41,601 and $51,871, financing costs of $21 and none. The rise is depreciation, not financing. And the figures count depreciation but not general and administrative expense, so "all-in" said more than they count.

The article called Riot's $90,631 an "all-in" cost and said it leaves out selling, general and administrative costs and interest. It now says Riot's cost including depreciation leaves out selling, general and administrative costs. The number did not change.

Riot's results release for the second quarter of 2026 builds the figure from cost of revenue lines and miner depreciation. Selling, general and administrative expense is not in it, but the release does not say interest is left out, and "all-in" is the site's name for a cost that counts that expense.

The article's sources and its real case said three miners publish an "all-in" cost per bitcoin and three publish none. They now say a cost per bitcoin with depreciation. The count did not change.

The three figures count depreciation and not general and administrative expense, so "all-in" said more than they count.

The card for the cost article and its chart line said "direct and all-in cost" and "the three all-in figures sit near 91,000 to 96,000". They now say "direct cost and cost with depreciation" and "the three figures with depreciation". The numbers did not change.

CleanSpark's, Riot's and TeraWulf's figures count the wear on the miners, but none counts general and administrative expense, which the site's own all-in cost counts, so "all-in" said more than they count. Each is the company's own figure, and the names now say what the figures include.

The cost column of the Operations table on the home page was headed "All-in $/BTC". It is now headed "With depreciation $/BTC", and its hover text says what the company's figure counts. The numbers did not change.

CleanSpark's, Riot's and TeraWulf's figures count the wear on the miners, but none counts general and administrative expense, which the site's own all-in cost counts, so "all-in" said more than they count. Each is the company's own figure, and the names now say what the figures include.

The stated mining cost chart wrote "All-in, company's definition" under CleanSpark, Riot and TeraWulf and described a direct figure as "Direct: power and hosting". It now writes "With depreciation" and "Direct, before depreciation", and the chart's definition of the measure says the same. The numbers did not change.

CleanSpark's, Riot's and TeraWulf's figures count the wear on the miners, but none counts general and administrative expense, which the site's own all-in cost counts, so "all-in" said more than they count. Each is the company's own figure, and the names now say what the figures include. A direct figure is the company's cost before depreciation, which is electricity alone for MARA and IREN.

The pair of gross margin against the company-stated cost was named "Margin against all-in cost" and the cost "All-in cost per bitcoin". They are now "Margin against cost with depreciation" and "Cost per bitcoin with depreciation". The figures did not change.

CleanSpark's, Riot's and TeraWulf's figures count the wear on the miners, but none counts general and administrative expense, which the site's own all-in cost counts, so "all-in" said more than they count. Each is the company's own figure, and the names now say what the figures include.

The tier list named the company-stated cost "All-in cost per bitcoin, as the company reports it" and described the direct cost as power and hosting. It now says "Cost per bitcoin with depreciation, as the company reports it" and "the cost of each bitcoin mined before depreciation", and the note on MARA no longer says its hosted sites "pay all-in fees". The figures did not change.

CleanSpark's, Riot's and TeraWulf's figures count the wear on the miners, but none counts general and administrative expense, which the site's own all-in cost counts, so "all-in" said more than they count. Each is the company's own figure, and the names now say what the figures include.

The compare tool named the company-stated cost row "All-in cost per bitcoin". It is now "Cost per bitcoin with depreciation", and its definition says the figure is the company's own. The figures did not change.

CleanSpark's, Riot's and TeraWulf's figures count the wear on the miners, but none counts general and administrative expense, which the site's own all-in cost counts, so "all-in" said more than they count. Each is the company's own figure, and the names now say what the figures include.

Beside the workbench's own all-in cost (Method B), the line for a company that states no figure with depreciation said "The company reports no all-in cost per bitcoin in the files", and the company's own figure was shown as "The company's own". They now say "no cost per bitcoin with depreciation" and "The company's own, with depreciation". The workbench's own all-in cost, which counts SG&A, keeps its name.

The company's figure with depreciation does not count general and administrative expense, so it is not the all-in cost the workbench works out, and setting the two side by side under one name could be read as one measure.

The methodology's list of operational metrics named "All-in cost per bitcoin" and defined it as Method B, and described the direct cost as "Method A: power and hosting costs divided by bitcoin mined", although the figures on the site are the companies' own. The entries are now "Cost per bitcoin with depreciation" and "Direct cost per bitcoin", each saying the figure is the company's own and for the period it states, and that Methods A and B are the site's own and are worked out in the workbench.

Method B counts payroll and general and administrative expense, and no figure in the entry does. Method A counts power and hosting, and the direct figures are the companies' own definitions: electricity alone for MARA and IREN, and the other direct costs their tables list for the rest.

Cipher Digital's Key figures listed Cost per bitcoin among the figures "not stated in the filings, releases or decks we read". They now say, on a line of their own, that the latest figure on file, $34,189 per BTC for the quarter ended Sep 30, 2025, is over a year old, so the list leaves it out. The figure is not in the list, as before, and the rule that keeps a figure of more than a year ago out did not change.

Cipher's business update for the third quarter of 2025 gives an all-in electricity cost per BTC of $34,189 for that quarter and $27,324 for the one before. Its business updates for the fourth quarter of 2025 and the first and second quarters of 2026 give none. The quarter ended Sep 30, 2025, more than a year before Oct 9, 2026, so the figure is on file and old, which is not the same as no source stating it.

In the Companies comparison, Cipher's cost per bitcoin cell said "not stated" and its hover text said "No filing, release or deck we read states it"; the cell now says "over a year old" and the hover text gives the figure, $34,189 per BTC for the quarter ended Sep 30, 2025, and says the table leaves it out. CleanSpark's cost per bitcoin was described as "includes depreciation and financing costs, excludes SG&A, owned sites only"; it now says "includes depreciation and financing costs, excludes SG&A, covers all of its mining", and the note above the table lists the new words. The hover text on each cost figure now opens with "What this figure counts:" instead of "As the company defines it:". The numbers did not change.

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 says the company "conducts mining exclusively at owned and leased facilities" and that its disclosures "reflect only the cost structure and operating statistics of owned and leased mining locations". Its table is headed owned facilities, but the 1,925 bitcoin it counts are before pool fees and the company mined 1,920 after them, so the table is all of its mining, unlike MARA's, which covers 1,260 of the 2,422 bitcoin MARA produced. Cipher's figure is on file and more than a year old. And not every cost is a figure the company defines as one: TeraWulf's $95,611 is the total a footnote of its 10-K gives.

CleanSpark's page said "owned sites only" under its cost per bitcoin ($96,277), titled its direct cost ($44,406) "Electricity cost per bitcoin" with the note "not hosting or its other costs", and showed a rank chip of "3rd of 3" on the cost with depreciation. The words now say the figure covers all of its mining; the direct cost is titled "Direct cost per bitcoin", and its card says CleanSpark's table calls the row its direct energy cost and that it is not electricity alone; the chip is gone; and each cost says its period in view, "Quarter ended Jun 30, 2026". The numbers did not change.

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 builds the $44,406 row from $44,317 of cost of energy and $89 of other direct costs of mining, non-energy utilities, so it is not the power bill alone. The same 10-Q says the company conducts mining exclusively at owned and leased facilities, so its table, headed owned facilities, covers all of its mining. The chip set CleanSpark's quarter against TeraWulf's year, an order of figures for different periods, which the tier list already declines to rank.

Riot Platforms' page said in the hover text on its cost per bitcoin ($90,631) that the $10.1 million of power curtailment credits were $7.4 million of manual curtailment credits and $2.7 million of demand response credits, under a link to its results release. The hover text now says the release gives the total and that the 10-Q's table of the credits gives the split. The rank chip ("1st of 3") on the cost with depreciation is gone, and the cost says "Quarter ended Jun 30, 2026" in view. The number did not change.

Riot's results release for the second quarter of 2026 shows "Less: power curtailment credits" of $10,054 thousand and says only that the credits come from participating in ERCOT's Demand Response Service Programs. The split is in the 10-Q: manual curtailment power credits of $7,361 thousand and demand response power credits of $2,693 thousand, the latter from ERCOT and MISO and received whether or not the company is called on to power down. The chip set Riot's quarter against TeraWulf's year, an order of figures for different periods.

TeraWulf's page showed a rank chip of "2nd of 3" on its cost with depreciation ($95,611), and the hover text on its cost per bitcoin opened with "As the company defines it:". The chip is gone, the hover text opens with "What this figure counts:", and the cost says "12 months ended Dec 31, 2025" in view. The number did not change. A later row of Oct 9, 2026 for this page records the move to TeraWulf's figure for the quarter ended Jun 30, 2026, and supersedes the figure and the period given here.

The chip set TeraWulf's figure for the year 2025 beside CleanSpark's and Riot's for the quarter ended Jun 30, 2026, an order of figures for different periods. And $95,611 is a total that a footnote of TeraWulf's 10-K gives, not the cost the 10-K tabulates: the table's own cost to mine one bitcoin is $53,681, a cash cost, and the footnote says that factoring in depreciation of $41,930 would bring the total "cost to mine one bitcoin" to $95,611.

Bitdeer's research row gave its derived figures for cost per bitcoin, about $74,200 with depreciation and stock compensation in cost of revenue and about $36,400 for electricity alone, with no period. The notes now say they are for the quarter ended Jun 30, 2026. The figures are our own arithmetic, not company figures, and did not change.

The inputs are from Bitdeer's results release for the second quarter of 2026: cost of revenue of $171.0 million for self-mining and $28.8 million for co-mining, electricity of $84.7 million and $13.3 million, and 2,694 bitcoin mined, all for the three months ended Jun 30, 2026.

The tier list said of the direct cost to mine that MARA's and CleanSpark's figures "cover their own sites only". It now says that MARA's covers its owned sites only. The Bitdeer rows now say their derived figures are for the quarter ended Jun 30, 2026. The figures did not change.

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 says the company conducts mining exclusively at owned and leased facilities, and its table counts 1,925 bitcoin before pool fees against the 1,920 it mined after them, so its figure covers all of its mining. MARA's covers 1,260 of the 2,422 bitcoin it produced in the quarter. Bitdeer's derived figures use the three months ended Jun 30, 2026 of its results release.

The article's list of sources gave MARA's 1,260 bitcoin produced at owned facilities as read from its shareholder letter, and listed neither MARA's 10-Q nor TeraWulf's 10-Q for the quarter ended Jun 30, 2026. Both are now listed, and the 1,260 is given under MARA's 10-Q. The article's figures did not change.

MARA's shareholder letter gives the purchased energy cost per bitcoin and the 2,422 bitcoin produced; the 1,260 produced at owned facilities is in the owned facilities statistics of its 10-Q. The $44,547 and $97,486 in the note on TeraWulf's 10-K entry are printed in TeraWulf's 10-Q for the quarter ended Jun 30, 2026.

Vulcan's Key figures listed Cost per bitcoin among the figures "not stated in the filings, releases or decks we read", the cost line on the page said "Not disclosed: no cost to mine one bitcoin is stated in a filing or release on file", and the Companies comparison said "not stated". They now show $114,083 per BTC for the quarter ended Jun 30, 2026, before depreciation, and the hover text gives the company's own definition. The block of research figures further down the page still reads "not disclosed", as of the research pass of Sep 6, 2026.

Vulcan's 10-Q for the quarter ended Jun 30, 2026, filed Aug 14, 2026, prints "Cost to mine one bitcoin(1) $114,083" for the three months and defines it: "Computed as cost of revenue of cryptocurrency mining divided by number of bitcoins produced from cryptocurrency mining." Its table of cost of revenue, which is stated exclusive of depreciation and amortization, gives $2,738 thousand for cryptocurrency mining, and its table of bitcoin produced gives 24 from cryptocurrency mining and 5 under datacenter hosting, so the divisor is the 24. The figure was in the filing while the page said no filing states one.

Argo's Key figures listed Cost per bitcoin among the figures "not stated in the filings, releases or decks we read", the cost line on the page said "Not disclosed: no cost to mine one bitcoin is stated in a filing or release on file", and the Companies comparison said "not stated". They now show $101,560 per BTC for the 12 months ended Dec 31, 2025, including depreciation and leaving out SG&A, covering all of its mining, and the hover text gives the company's definition and its direct cost of $84,193. The block of research figures further down the page still reads "not disclosed", as of the research pass of Sep 6, 2026.

Argo's 20-F for 2025, filed Apr 30, 2026, reconciles its two measures in a table over 150 bitcoin mined: "Total Cost of Bitcoin Mined ($000s) 15,234", "Average Total Cost Per Bitcoin Mined ($) 101,560", "Depreciation of mining equipment ($000s) (2,605)", "Direct Cost of Bitcoin Mined ($000s) 12,629" and "Average Direct Cost Per Bitcoin Mined ($) 84,193". The total is the direct cost plus the depreciation of mining equipment (12,629 plus 2,605 is 15,234). The 20-F's text above the table words the total differently, as "direct costs divided by total bitcoin mined"; the figures here follow the table, which closes.

Keel's Key figures listed Cost per bitcoin among the figures "not stated in the filings, releases or decks we read", the cost line on the page said "Not disclosed: no cost to mine one bitcoin is stated in a filing or release on file", and the Companies comparison said "not stated". They now say that the latest figure on file, $44,100 per BTC for the quarter ended Sep 30, 2025, is over a year old, so the list and the table leave it out, and the cost line gives the figure with its period and the company's definition. The rule that keeps a figure of more than a year ago out did not change.

The restated management's discussion for the third quarter of 2025, which Keel filed on Form 6-K on Mar 31, 2026 under its former name Bitfarms, prints "Direct Cost per Bitcoin (in U.S. dollars) 44,100" for the three months ended Sep 30, 2025, from continuing operations, over 520 bitcoin earned. Keel's 10-K for 2025 says this about a cost per bitcoin and nothing more: total cash cost per bitcoin "would be reduced to $75,400 after considering the realized and unrealized gain on Bitcoin option contracts", a conditional figure with no base figure beside it, which the site does not show as Keel's cost.

Core Scientific's Key figures listed Cost per bitcoin among the figures "not stated in the filings, releases or decks we read", the cost line on the page said "Not disclosed: no cost to mine one bitcoin is stated in a filing or release on file", and the Companies comparison said "not stated". They now say that the latest figure on file, $79,002 per BTC for the quarter ended Sep 30, 2025, is over a year old, so the list and the table leave it out, and the cost line gives the figure with its period and the company's definition. The rule that keeps a figure of more than a year ago out did not change.

Core Scientific's amended 10-Q for the quarter ended Sep 30, 2025, filed Mar 2, 2026, prints a "Total cost to self-mine one bitcoin" of $79,002 for the three months (power cost of $64,783 and operational costs of $14,219) and defines it: "Represents our direct cash costs of power and operational costs based on our self-mining/hosting mix divided by total bitcoin self-mined during the periods presented." The original 10-Q of Oct 24, 2025 prints the same figure. Its 10-K for 2025 and its 10-Qs for 2026 print no such figure.

In the Companies comparison, the cost per bitcoin cells of Vulcan and Argo said "not stated" and their hover text said "No filing, release or deck we read states it"; they now show $114,083 per BTC for the quarter ended Jun 30, 2026 and $101,560 per BTC for the 12 months ended Dec 31, 2025, each with the company's own definition in the hover text. The cells of Keel and Core Scientific said "not stated"; they now say "over a year old", and the hover text gives the figure and its period ($44,100 and $79,002, both for the quarter ended Sep 30, 2025). The Costs disclosed tile said that 5 of 18 companies state a cost to mine a bitcoin "in their latest reports"; it now says the count is of the reports the research pass read on Sep 6, 2026, which do not include Vulcan's and Argo's. The period line under each cost chart said "each company's latest stated period", which is not true of IREN, whose latest figure is for the quarter ended Dec 31, 2025 and whose later reports print none; the lines now say what each chart shows.

Vulcan's 10-Q for the quarter ended Jun 30, 2026 prints "Cost to mine one bitcoin(1) $114,083" and Argo's 20-F for 2025 prints "Average Total Cost Per Bitcoin Mined ($) 101,560"; Keel's restated discussion for the third quarter of 2025 prints "Direct Cost per Bitcoin (in U.S. dollars) 44,100" and Core Scientific's amended 10-Q for that quarter a "Total cost to self-mine one bitcoin" of $79,002. The rows for each company's page give the sentence and the table. IREN's 10-Qs and 10-K after the quarter ended Dec 31, 2025 print no net electricity cost per Bitcoin mined.

The chart Stated mining cost against bitcoin price named Argo and Core Scientific under "No figure stated", and left the cost bar of Vulcan and of Keel empty with the words "No cost to mine stated". It now draws a bar for each of the four: $114,083 for Vulcan (quarter ended Jun 30, 2026), $101,560 for Argo (12 months ended Dec 31, 2025), and $44,100 for Keel and $79,002 for Core Scientific, both for the quarter ended Sep 30, 2025, which the period and the source under the chart say. TeraWulf's bar is now its $97,486 for the quarter ended Jun 30, 2026.

The same filings: Vulcan's 10-Q prints "Cost to mine one bitcoin(1) $114,083", Argo's 20-F "Average Total Cost Per Bitcoin Mined ($) 101,560", Keel's restated discussion "Direct Cost per Bitcoin (in U.S. dollars) 44,100", Core Scientific's amended 10-Q a "Total cost to self-mine one bitcoin" of $79,002, and TeraWulf's 10-Q for the quarter ended Jun 30, 2026 "Cost to mine one bitcoin(1) $44,547" with $52,939 of depreciation "bringing the total" to $97,486. The chart draws the company's own number with its basis written under the name and does not adjust any of them to one definition.

CleanSpark's cost per bitcoin with depreciation ($96,277) was described as including "financing costs", and the card of its direct cost ($44,406) in the block of research figures named the owned facilities table as its method. The words now say "financing costs (none this quarter)", and the cards say the direct cost is the cost of energy, $44,317, plus $89 of other direct costs of mining, and that it covers all of its mining. The numbers did not change.

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 prints "Financing costs per bitcoin mined" as a dash for the quarter, so the financing cost that its definition adds was nil in that quarter. The same table builds the $44,406 row from $44,317 of cost of energy and $89 of other direct costs of mining (non-energy utilities), and the 10-Q says the company mines exclusively at owned and leased facilities, so the table covers all of its mining.

TeraWulf's block of research figures read the research pass of Sep 6, 2026: a cost with depreciation of $95,611 and a direct cost of $53,681, both for the year 2025, beside the quarter ended Jun 30, 2026 that the rest of the page shows. It now shows the figures of that quarter, $97,486 with depreciation and $44,547 direct, for the Lake Mariner Data Campus alone, each with its period beside it.

TeraWulf's 10-Q for the quarter ended Jun 30, 2026, filed Aug 5, 2026, prints "Cost to mine one bitcoin(1) $44,547" in its table for the Lake Mariner Data Campus and says that factoring in the depreciation of its miner fleet would add $52,939 for the quarter, bringing the total "cost to mine one bitcoin" to $97,486. Its 10-K for 2025 gave $53,681 for a table that also included the company's net share of the Nautilus Cryptomine Facility.

Vulcan's block of research figures read "not disclosed" for the cost per bitcoin, as of the research pass of Sep 6, 2026, beside Key figures that show $114,083. It now shows $114,083 per BTC for the quarter ended Jun 30, 2026, before depreciation, with its period beside it, as the rest of the page does.

Vulcan's 10-Q for the quarter ended Jun 30, 2026, filed Aug 14, 2026, prints "Cost to mine one bitcoin(1) $114,083" for the three months, and defines it as the cost of revenue of cryptocurrency mining, stated exclusive of depreciation and amortization, divided by the bitcoin produced from cryptocurrency mining.

Argo's block of research figures read "not disclosed" for the cost per bitcoin, as of the research pass of Sep 6, 2026, beside Key figures that show $101,560. It now shows $101,560 per BTC with depreciation and its direct cost of $84,193, each for the 12 months ended Dec 31, 2025, with the period beside it.

Argo's 20-F for 2025, filed Apr 30, 2026, reconciles "Average Total Cost Per Bitcoin Mined ($) 101,560" and "Average Direct Cost Per Bitcoin Mined ($) 84,193" over 150 bitcoin mined in a table that closes: the total is the direct cost of $12,629 thousand plus $2,605 thousand of depreciation of mining equipment.

IREN's block of research figures said under its cost with depreciation "Not disclosed. Derived FY26 cost of revenue per BTC is about $33,400 (see direct cost note)", although the direct cost note holds IREN's net electricity cost of $34,400 for the quarter ended Dec 31, 2025, not that figure. The note now gives its own inputs and period: for the year ended Jun 30, 2026, derived here and not a company figure, the Bitcoin mining cost of revenue, exclusive of depreciation and amortization, of $202.8 million over 6,075 Bitcoin mined is about $33,400 per Bitcoin, which leaves depreciation out.

IREN's 10-K for the year ended Jun 30, 2026 shows a cost of revenue for Bitcoin mining, exclusive of depreciation and amortization, of $202,774 thousand and says the revenue was generated from the mining and sale of 6,075 Bitcoin. $202,774 thousand over 6,075 is $33,378. The 10-K states no cost per Bitcoin, so the figure is our arithmetic.

Core Scientific's block of research figures gave two derived figures after the words "Not disclosed", one with no period and one for "H1 2026", and its direct cost read "not disclosed". The notes now open with their period, "For the six months ended Jun 30, 2026, derived here, not a company figure", and the direct cost says that the latest figure on file, $79,002 per BTC for the quarter ended Sep 30, 2025, is over a year old, so the row leaves it out. The derived figures, about $116,400 and $64,900 per bitcoin, did not change.

Core Scientific's 10-Q for the quarter ended Jun 30, 2026 shows a cost of digital asset self-mining of $80,889 thousand for the six months and about 695 bitcoin mined; $80.9 million over 695 bitcoin is about $116,400 per bitcoin, our arithmetic. Its amended 10-Q for the quarter ended Sep 30, 2025 prints the total cost to self-mine one bitcoin of $79,002.

Cipher's block of research figures read "not disclosed" for its direct cost per bitcoin and did not say that a figure is on file. It now says that the latest figure on file, $34,189 per BTC for the quarter ended Sep 30, 2025, power cost only, is over a year old, so the row leaves it out. The month-of-liquidity note beside it says that no direct cost per bitcoin that is current is on file, in place of "the company does not state the direct cost per bitcoin".

Cipher's business update for the third quarter of 2025 gives an all-in electricity cost per BTC of $34,189 for that quarter, which counts electricity only and not other operating costs. Its later business updates give none, so the newest figure is more than a year before Oct 9, 2026.

Keel's block of research figures read "not disclosed" for its direct cost per bitcoin and did not say that a figure is on file. It now says that the latest figure on file, $44,100 per BTC for the quarter ended Sep 30, 2025, before depreciation, is over a year old, so the row leaves it out.

The restated management's discussion for the third quarter of 2025, filed on Form 6-K on Mar 31, 2026, prints "Direct Cost per Bitcoin (in U.S. dollars) 44,100" for the three months ended Sep 30, 2025, which is more than a year before Oct 9, 2026.

Hut 8's cost line and its block of research figures said "not disclosed" for its cost per bitcoin. They now say "not stated" and that no filing, release or deck we read states one, and that the only cost to mine one bitcoin on file is American Bitcoin's, a Hut 8 subsidiary, from its own results release. No figure was added or removed.

Hut 8's filings and releases that we read state no cost to mine a bitcoin of the parent's own; American Bitcoin, a listed subsidiary, states its own in its results release, and a subsidiary's figure is not the parent's. "Not disclosed" says the company withholds a figure, which we do not know; "not stated" says what we read.

The hover text on MARA's cost per bitcoin ($38,690) quoted that 1,260 of the 2,422 bitcoin it produced were at its owned facilities, a fact that is in MARA's 10-Q and not in the shareholder letter the figure links. The card now links both documents. The number did not change.

MARA's shareholder letter for the second quarter of 2026 gives the purchased energy cost per BTC of $38,690 and the 2,422 bitcoin produced; the 1,260 produced at owned facilities is in the owned facilities statistics of its 10-Q.

The card of Riot's cost per bitcoin ($90,631) quoted the split of its power curtailment credits, a fact that is in Riot's 10-Q and not in the results release the figure links. The card now links both documents. The number did not change.

Riot's results release shows "Less: power curtailment credits" of $10,054 thousand; the split into manual curtailment power credits of $7,361 thousand and demand response power credits of $2,693 thousand is in the table of credits of its 10-Q.

The Companies page called its cost bars "the ranked bar", ordered Riot, TeraWulf and CleanSpark by figure, and said under the chart that each company's "latest stated period" was drawn, although each figure is the company's own definition for its own period and the tier list says these figures are not ranked. The chart now says it orders the companies by size, with a figure with depreciation first and the lowest first, that this is an order and no ranking, and that the length of each company's period is under its name. The "Costs disclosed" tile counted five of 18 from the research pass of Sep 6, 2026; it now counts the stored rows, the companies with a cost per bitcoin for a period that ended within the last year (seven of 18 on Oct 9, 2026, with Vulcan and Argo added), and says what it counts. The cost table, the bars and the comparison print the length of each period beside the figure (3 months, 12 months). The numbers did not change except for those of TeraWulf, whose year 2025 gave way to the quarter ended Jun 30, 2026 in the table and the bars.

Each figure is a company's own cost per bitcoin on its own definition: TeraWulf's 10-Q for the quarter ended Jun 30, 2026 prints "Cost to mine one bitcoin(1) $44,547" with $52,939 of depreciation "bringing the total" to $97,486, Vulcan's 10-Q prints $114,083 for the quarter and Argo's 20-F $101,560 for 2025. The research pass of Sep 6, 2026 had read fewer documents than the stored rows hold, so it counted five, and it held TeraWulf's year 2025.

The cost with depreciation row of the comparison drew bars from zero across figures of different periods (TeraWulf's year beside two quarters) and printed only a date chip; a company the research pass had not covered read "Not disclosed" in that row. The row now prints the length of each period beside its figure, draws no bar for a cost, and says under the table why; a company with no figure reads "not stated".

Each cost per bitcoin is the company's own definition for its own period, and the figures with different words are not comparable, so a bar from zero sets them in an order of size that the words do not support.

The chart Stated mining cost against bitcoin price said it covered "each company's latest stated period" and drew bars for Cipher ($34,189), Core Scientific ($79,002) and Keel ($44,100), all for the quarter ended Sep 30, 2025, with nothing to say they were more than a year old. It now says it covers the latest period read for each company, with the length of its own under its name. The three figures are not drawn, since they are more than a year old: they are named under the chart with their period and what they count, and Keel's empty bar says it is over a year old.

The pipeline's freshness rule, which did not change, leaves a figure out when its period ended more than a year before the day the file is written: Cipher's business update, Keel's restated discussion and Core Scientific's amended 10-Q are all for the quarter ended Sep 30, 2025.

The cost categories of the tier list read the research pass of Sep 6, 2026: TeraWulf's year 2025, "not disclosed" for Vulcan and Argo, and a flag on MARA. They now read the stored rows, with each figure's period and what it counts in its sentence, Vulcan's and Argo's figures, and Cipher, Core Scientific and Keel as "over a year old" with the figure and its period. The categories stay unranked.

The stored rows hold the cost per bitcoin of each company's latest filing or release as the company states it: TeraWulf's 10-Q for the quarter ended Jun 30, 2026 ($44,547 direct, $97,486 with depreciation), Vulcan's 10-Q ($114,083) and Argo's 20-F ($84,193 direct, $101,560 with depreciation).

The home Operations table printed the cost with depreciation with no period, and the sector average cost per bitcoin held gave its date in the hover text only. The cost cells now print the length of their period under the figure, the sector average line says "newest as of Jun 30, 2026", and a company the research pass has not covered reads "Not stated" for the cost in place of "Not disclosed". The numbers did not change.

The sector average is the total cost basis over the total bitcoin held of the miners whose filings give both; the newest of their dates is Jun 30, 2026. A cost per bitcoin is a flow over a period, so a quarter and a year in one column need their lengths.

The cost per bitcoin figures of the analysis pages were the research pass's of Sep 6, 2026: TeraWulf's year 2025 ($95,611 with depreciation, $53,681 direct), and none for Vulcan and Argo. They are now the stored rows': TeraWulf's quarter ended Jun 30, 2026 ($97,486 and $44,547), Vulcan's $114,083 and Argo's $101,560 and $84,193, each with its period. The three pairs that use a cost (cost with depreciation against margin, direct cost against margin, and efficiency against direct cost) are recomputed on them, with the same number of companies and different statistics.

TeraWulf's 10-Q for the quarter ended Jun 30, 2026 prints "Cost to mine one bitcoin(1) $44,547" and a total with depreciation of $97,486. Vulcan and Argo have no margin or fleet efficiency on file, so they join no pair.

The article's table gave TeraWulf's costs for the year 2025 beside the quarter ended Jun 30, 2026 of the two other companies, said that TeraWulf's 10-K states no value per bitcoin mined to set a cost against, and described the figures "on energy alone". It now gives TeraWulf's quarter ended Jun 30, 2026 ($44,547 direct, $97,486 with depreciation) against the $71,704 its 10-Q prints as the value of each bitcoin mined, 62.1 percent for the direct cost and 136.0 percent with depreciation (the second is this page's arithmetic), and speaks of the direct cost. The periods of the tables and the chart are those of the stored rows.

TeraWulf's 10-Q for the quarter ended Jun 30, 2026 prints "Value of each bitcoin mined" of $71,704 and the cost to mine one bitcoin as 62.1 percent of it. Its 10-K for 2025 also prints a value of each bitcoin mined, $101,307, with the cost as 53.0 percent of it, so the sentence that the 10-K states none was wrong.

The article said that CleanSpark's $44,406 per bitcoin was the "energy at owned facilities" and that its $96,277 added "energy, machine wear and financing costs". It now says that the $44,406 is the cost of energy, $44,317, plus $89 of other direct costs of mining, that the figure with depreciation adds miner depreciation of $51,871, and that financing costs were none in the quarter. The numbers did not change.

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 builds the direct row from $44,317 of cost of energy and $89 of other direct costs of mining (non-energy utilities), adds $51,871 of miner depreciation per bitcoin for the figure with depreciation, and prints financing costs per bitcoin mined as a dash.

The article said that CleanSpark's quarter "cost 44,406 dollars of energy" and named a range of direct costs by company without their periods. It now says the $44,406 is $44,317 of energy and $89 of other direct costs, says the quarter ended Jun 30, 2026, and gives each end of the range of direct costs with its company and its period. The numbers did not change.

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 builds the $44,406 from $44,317 of cost of energy and $89 of other direct costs of mining (non-energy utilities).

The keyed API's research datasets, hero_metrics and tier2, gave the research pass of Sep 6, 2026 for the two cost_to_mine fields of every company. TeraWulf's read $95,611 with depreciation and $53,681 before, both for "FY2025", and Argo, Keel and Vulcan read "Not disclosed by the company in the documents read on Sep 6, 2026". Each field now holds the stored row's figure, the day its period ended, its period in words, what it counts, its sources and the company's own definition. TeraWulf reads $97,486 with depreciation and $44,547 before, for the quarter ended Jun 30, 2026, at its Lake Mariner Data Campus alone; Argo $101,560 with depreciation and $84,193 before, for the 12 months ended Dec 31, 2025; Vulcan $114,083 before depreciation, for the quarter ended Jun 30, 2026. Where the latest figure on file is over a year old (Cipher's $34,189, Core Scientific's $79,002 and Keel's $44,100, each for the quarter ended Sep 30, 2025), the field has no value, its note gives the figure with its period and says it is over a year old, and a new field called stale holds it. Where no filing, release or deck we read states a cost, the note says so. The figures of CleanSpark, Riot, MARA and IREN did not change; their periods now read as the pages read them, for example "quarter ended Jun 30, 2026" in place of "Q3 FY2026". No field was removed or renamed. Added: period_kind, basis, stale, research_note and research_as_of (the Sep 6 notes that stay, dated), and a cost_to_mine block giving the day the cost fields are read as of.

TeraWulf's 10-Q for the quarter ended Jun 30, 2026 prints "Cost to mine one bitcoin(1) $44,547" for its Lake Mariner Data Campus and a footnote that adds $52,939 of depreciation "bringing the total 'cost to mine one bitcoin' to $97,486"; the year 2025 the dataset held is the 10-K's. Argo's 20-F for 2025 prints "Average Total Cost Per Bitcoin Mined ($) 101,560" and "Average Direct Cost Per Bitcoin Mined ($) 84,193". Vulcan's 10-Q prints "Cost to mine one bitcoin(1) $114,083" for the three months ended Jun 30, 2026. Keel's restated discussion for the third quarter of 2025 prints "Direct Cost per Bitcoin (in U.S. dollars) 44,100". The research pass had read fewer documents than the pipeline holds, and the pages already showed these figures, so the API and the pages disagreed; both now read the same stored rows.

TeraWulf's months of liquidity note said "Not disclosed: the company does not state the bitcoin held" and listed "a direct cost of $53,681 per bitcoin for 2025 (its 10-K) and 179 bitcoin mined last quarter", a year's cost beside a quarter's bitcoin. It now says months of liquidity is not formed, and lists cash of $2.62B, 2 bitcoin held at Jun 30, 2026, a direct cost of $44,547 per bitcoin for the quarter ended Jun 30, 2026 (its 10-Q) and 179 bitcoin mined in that quarter; the reason it is not comparable is unchanged. The same note is in the comparison, the tier list and the keyed API.

TeraWulf's 10-Q for the quarter ended Jun 30, 2026 prints "Cost to mine one bitcoin(1) $44,547" and "Total bitcoin mined 179" for the three months, and says "As of June 30, 2026, our digital asset balance was comprised of 2 bitcoin recorded at its fair value of $133,000", so the company does state its bitcoin held. $53,681 is the cost its 10-K gives for the year 2025.

MARA's months of liquidity read 104.6. Its method divided cash of $421.3 million plus 35,577 bitcoin at $80,038 by a monthly cost of $31.2 million: its purchased energy cost per bitcoin, $38,690, times all 2,422 bitcoin it produced in the quarter, over 3. It now reads 201.2 months, from a monthly cost of $16.25 million, a third of the $48.75 million of purchased energy costs for the quarter ended Jun 30, 2026, and its method says that this is the run-rate of its owned mining operations only and leaves out third-party hosting and other energy costs. The 10-Q is cited for the $48.75 million and the 1,260 bitcoin.

MARA's 10-Q for the quarter ended Jun 30, 2026 prints "Purchased energy costs 48,750" (thousands) in its statement of operations and, in its owned facility statistics, "Purchased energy costs per BTC $38,690" with "Total BTC produced during the period, in whole BTC at owned facilities 1,260"; it defines the per-bitcoin figure as the amounts paid to power providers for power consumed "divided by the quantity of bitcoin produced during the period related to our owned mining operations". Only 1,260 of the 2,422 bitcoin it produced in the quarter were produced at owned facilities, so 38,690 times 2,422, $93.7 million, is almost twice the $48.75 million of purchased energy the 10-Q prints.

IREN's months of liquidity read 730.5. Its method divided cash of $5,895.6 million by "direct cost 26,259 x BTC mined last quarter 922 / 3". The 26,259 is one month: it is the "Net electricity cost per Bitcoin mined ($26,259)" of IREN's monthly update for June 2025, set against a quarter's bitcoin. The method now divides by IREN's net electricity costs of its Bitcoin mining for the quarter ended Dec 31, 2025, $57.2 million, the latest quarter it states them, over 3, and the months read 309.1. The with-depreciation row of its research figures no longer says that its derived $33,400 "leaves depreciation out" under a label that says with depreciation: the note says first that the number is before depreciation. The same words are in the comparison and the keyed API.

IREN's 10-Q for the quarter ended Dec 31, 2025 prints "Net electricity costs - Bitcoin Mining (57,225)" and "Bitcoin mined 1,664" (thousands of dollars) and "Net electricity costs per Bitcoin mined (34.4)". Its monthly update for June 2025 prints "Bitcoin mined 620 BTC" and "Net electricity cost per Bitcoin mined ($26,259)" for that month alone. Its 10-K for the year ended Jun 30, 2026 states no cost per Bitcoin, so the $33,400 derived from its cost of revenue, exclusive of depreciation and amortization, is before depreciation.

Five notes of the months of liquidity category began "Not disclosed:". Bitdeer's ("the company does not state the direct cost per bitcoin") and Hut 8's now read "Not stated in the filings, releases or decks we read", which says what was read and not that the company withholds a figure. Cipher's and Core Scientific's, which said that no current direct cost per bitcoin is on file and that the latest is over a year old, now give that latest figure with its period, $34,189 and $79,002 per BTC for the quarter ended Sep 30, 2025, and begin "Months of liquidity cannot be formed". TeraWulf's, MARA's and IREN's notes are the rows above. The word a research tile prints when it has no value, on the company pages, the comparison and the home table, reads "Not stated" in place of "Not disclosed", to match its note. The category stays unranked.

"Not disclosed" says a company withholds a figure, which we do not know; "not stated" says what we read (the Hut 8 row of this log). Cipher's business update for the third quarter of 2025 gives an all-in electricity cost per BTC of $34,189 and Core Scientific's amended 10-Q for that quarter a total cost to self-mine one bitcoin of $79,002, so each company has a figure, over a year old, and the note now gives it.

In the cost table on the Companies page, TeraWulf's and Vulcan's revenue per bitcoin mined read "No revenue on file", and so did the share of each cost against it. They now show $71,704 and $70,208, each from the company's 10-Q table: Vulcan's direct cost is 162.5 percent of its value of each bitcoin mined and TeraWulf's 62.1 percent, as the tables state, and TeraWulf's cost with depreciation, 136.0 percent, is worked out on the page from its two figures. The hover text of the cost cells of Cipher, Core Scientific and Keel, which say the latest figure on file is over a year old, now also gives what the figure counts, after the words "What this figure counts:".

TeraWulf's 10-Q for the quarter ended Jun 30, 2026 prints "Value of each bitcoin mined(2) $71,704" and "Cost to mine one bitcoin as % of value of bitcoin mined 62.1"; Vulcan's prints "Value of each bitcoin mined(2) $70,208" and "Cost to mine one bitcoin as % of value of bitcoin mined 162.5". The page read a share of revenue only from the research pass's notes of Sep 6, which were written about TeraWulf's year 2025 and read nothing for Vulcan. 97,486 over 71,704 is 135.95 percent.

The article's table said that American Bitcoin's $36,500 per bitcoin "Includes colocation and managed-services fees paid to Hut 8", and its list of sources said the same of the results release. The release does not say it. The table and the sources now say that the release does not define the figure, and cite American Bitcoin's 10-Q: cost of revenue before depreciation and amortization of $34.0 million over about 932 bitcoin mined, which is about the same figure, with all its miners at Hut 8's facilities, where Hut 8 billed it $31.7 million under the colocation agreement and $3.4 million under the management agreement for the quarter. The number did not change.

American Bitcoin's results release of Aug 3, 2026 says only "Cost to mine was ~$36,500 per Bitcoin in Q2 2026 and held roughly flat compared to ~$36,200 per Bitcoin in Q1 2026, driven by marginally higher energy costs at selective sites". Its 10-Q for the quarter ended Jun 30, 2026 prints cost of revenue, exclusive of depreciation and amortization, of $34,009 thousand and says production rose "to approximately 932 Bitcoin mined"; it says "all of the Company's Bitcoin miners are located at Hut 8's facilities" and that "fees and expenses billed by Hut 8 to the Company under the MCSA were $31.7 million" and under the management agreement $3.4 million for the three months. 34,009 over 932 is $36,491. Neither document says the cost to mine includes those fees.

The Methodology page said, of the pair of cost per bitcoin against revenue per bitcoin mined, that "no company has both halves on file for the same quarter (probe run Sep 10, 2026)", while CleanSpark, Riot, TeraWulf and Vulcan each print a cost to mine and a revenue per bitcoin mined for the quarter ended Jun 30, 2026. It now says what the halves are (a quarter's revenue as tagged in XBRL over the bitcoin mined in its three monthly releases, and a cost for the same quarter), that no company had all three on file in the probe of Sep 10, 2026 or in its run of Oct 9, 2026, and that the four companies' own ratios are on the Companies page and are not this pair.

The check that was run builds revenue per bitcoin from a quarter's tagged revenue and the bitcoin mined in its three monthly releases, and sets it against a cost for the same quarter. Run again on Oct 9, 2026 over the four quarters ending Sep 30, 2025 to Jun 30, 2026, it finds no company with the quarter's tagged revenue, the bitcoin mined in all three monthly releases and a cost for the same quarter all on file, so no complete pair. The four companies print their own revenue per bitcoin mined: CleanSpark's 10-Q an average revenue per bitcoin mined of $71,692, Riot's release a production value of $71,667 per bitcoin mined, TeraWulf's 10-Q "Value of each bitcoin mined" of $71,704 and Vulcan's $70,208.

The changelog's list of datasets dated the operating figures "hashrate, efficiency, cost to mine" as of Sep 6, 2026, the day of the research pass, though the cost to mine is read from the stored rows, which the pipeline rewrites each night. It now has two lines: the operating figures (hashrate, efficiency, months of liquidity) as of Sep 6, 2026, and the cost to mine a bitcoin as of the day of the stored rows. The entry of Oct 9, 2026 that said every cost per bitcoin on the site takes its period, scope and age from the one stored row, and that a company no source states a cost for reads "not stated", was true of the pages and not of the data the keyed API serves, the months of liquidity notes or American Bitcoin's figure; it now says what is true.

The cost figures come from the canonical metrics file, whose day is the day the pipeline last ran, and the research pass's two files are dated Sep 6, 2026, so a line that named the cost to mine under the second date made the figures look a month older than they are. The entry's sentence was written when the pages had changed and the data files they sit beside had not.

The Across companies tab of the analysis tested the cost to mine a bitcoin in three pairs (the cost with depreciation against gross margin, the direct cost against gross margin, and fleet efficiency against the direct cost) and drew each cost as a ranked spread with a mean, a standard deviation and quartiles, reading for example "ARBK is highest, at $101,560 per BTC, and RIOT lowest, at $90,631 per BTC. The median of the 4 is $96,882" and "VIP is highest, at $114,083 per BTC, and IREN lowest, at $34,400 per BTC. The median of the 7 is $44,547". Each of those figures is a company's own definition for its own period, so a highest, a lowest, a median or a fitted line across them ranked and averaged what is not one measure. The tab no longer tests or draws a cost per bitcoin to mine, and says why in one sentence. Gross margin, which only those two pairs used, is no longer drawn as a spread either; the other seven pairs and their figures are unchanged, and the Findings list loses the tests of the three removed pairs.

The stored cost rows give each company its own period and definition. Argo's $101,560 with depreciation is for the 12 months ended Dec 31, 2025 (its 20-F: "Average Total Cost Per Bitcoin Mined ($) 101,560"); CleanSpark's $96,277 ("Direct cost to mine including non-cash depreciation and financing costs"), Riot's $90,631 ("Cost to mine one bitcoin, including bitcoin miner depreciation") and TeraWulf's $97,486 (its 10-Q: depreciation "would add $52,939", "bringing the total ... to $97,486", for the Lake Mariner Data Campus alone) are for the quarter ended Jun 30, 2026. Of the direct figures, MARA's $38,690 ("Purchased energy cost per BTC was $38,690 in Q2 2026 for our owned sites") is the power bill of its owned sites, IREN's $34,400 ("Net electricity costs per Bitcoin mined") is the power bill for the quarter ended Dec 31, 2025, Vulcan's $114,083 leaves out the bitcoin produced under datacenter hosting, and Argo's $84,193 is for 12 months. On Oct 9, 2026 no two of the eleven figures shared a basis, a period and a scope.

The cost row of the Compare page showed the cost with depreciation alone, so the direct costs of CleanSpark ($44,406), Riot ($49,912), TeraWulf ($44,547), Argo ($84,193), MARA ($38,690), IREN ($34,400) and Vulcan ($114,083) were on no row of it, and MARA, IREN and Vulcan, which state only a direct figure, read "not stated" there. The page now has a row for the direct cost under the cost with depreciation, each cost figure names its length and the day its period ends, and the line under each cost row names the period its figures share or says that each is for its own. The line used to say "as of" a day, which does not say whether a figure is for a quarter or a year.

The stored cost rows hold each direct figure for the period its document states. For the quarter ended Jun 30, 2026: CleanSpark's 10-Q table "Cost to mine one bitcoin - direct energy cost - Owned facilities $44,406", Riot's release "Cost to mine one bitcoin, excluding bitcoin miner depreciation" $49,912, TeraWulf's 10-Q "Cost to mine one bitcoin" $44,547, MARA's "Purchased energy cost per BTC was $38,690 in Q2 2026 for our owned sites" and Vulcan's 10-Q "Cost to mine one bitcoin" $114,083. For the 12 months ended Dec 31, 2025: Argo's 20-F "Average Direct Cost per Bitcoin Mined (excludes depreciation of mining machines)" 84,193. For the quarter ended Dec 31, 2025: IREN's 10-Q "Net electricity costs per Bitcoin mined" 34.4 thousand dollars. Argo's year and IREN's quarter both end on Dec 31, 2025.

The chart Cost to mine a bitcoin drew Argo's figures for 12 months and IREN's for the quarter ended Dec 31, 2025 from zero among the other companies' figures for the quarter ended Jun 30, 2026, with only a length ("3 months", "12 months") under each name, and its line ended with "As of Sep 6, 2026", the day of the spot price. The bars now stand in groups by the period their figures are for, the latest first, each under a heading that names the period and the day it ends, and the chart's line carries no day of its own. The cost table, the table of every miner's metrics and the Operations table on the home page now print the day each period ends under the figure, beside the length.

The stored cost rows give Argo's figures for the 12 months ended Dec 31, 2025 (its 20-F: "Average Total Cost Per Bitcoin Mined ($) 101,560"), IREN's for the quarter ended Dec 31, 2025 (its 10-Q: "Net electricity costs per Bitcoin mined" 34.4 thousand dollars) and CleanSpark's, Riot's, TeraWulf's, MARA's and Vulcan's for the quarter ended Jun 30, 2026. Bars from zero for all of them in one list read as one comparison of one period; a group per period says which is which.

The chart Stated mining cost against bitcoin price drew the same bars from zero side by side across periods (Argo's 12 months, IREN's quarter ended Dec 31, 2025 and the other miners' quarter ended Jun 30, 2026), with only a length under each name, and its line ended with an as-of day that was not any bar's period. The bars now stand in groups by the period of the cost to mine, the latest first, each under a heading that names the period and the day it ends, with the miners whose newest cost to mine is over a year old in a group of their own beside their other bar, and the line carries no day of its own.

The stored cost rows give each miner's period: Argo's 12 months ended Dec 31, 2025, IREN's quarter ended Dec 31, 2025 and, for CleanSpark, MARA, Riot, TeraWulf and Vulcan, the quarter ended Jun 30, 2026, each from the filing or release the bar's hover names.

The article's first chart, Cost per bitcoin by company, drew IREN's direct figure for the quarter ended Dec 31, 2025 from zero beside the other companies' figures for the quarter ended Jun 30, 2026, under "As of Sep 12, 2026", the day of the article, and its reading gave a range across companies that each count their own way: "Six companies' direct cost sits between 34,000 and 50,000 dollars a bitcoin. The three figures with depreciation sit near 91,000 to 97,000." The chart now draws only the figures for the quarter ended Jun 30, 2026, which its key names (CleanSpark, Riot, TeraWulf, MARA and American Bitcoin), says in its note that IREN's is for another quarter and is in the table above, and reads that each of the three companies that state both is higher with depreciation. It carries no day of its own and gives no range. The same chart is drawn on the card for the article on the Learn page.

TeraWulf's 10-Q states $44,547 before depreciation and $97,486 with it for the quarter ended Jun 30, 2026, CleanSpark's $44,406 and $96,277 and Riot's $49,912 and $90,631 for the same quarter; MARA's $38,690 ("Purchased energy cost per BTC was $38,690 in Q2 2026 for our owned sites") is for that quarter too, and American Bitcoin's results release gives about $36,500 for it. IREN's 10-Q gives its net electricity costs per bitcoin mined, 34.4 thousand dollars, for the quarter ended Dec 31, 2025.

The changelog entry on the held bitcoin cost lines said that TeraWulf and Bitdeer "are named with their reasons in a line of their own instead of under No figure stated". Only Bitdeer had been under No figure stated: TeraWulf states a cost to mine and has a bar for it. The entry now says that TeraWulf and Bitdeer are named with their reasons in a line of their own, and Bitdeer is no longer under No figure stated.

The line headed No figure stated under the chart Stated mining cost against bitcoin price lists the miners whose filings state neither a cost to mine nor a cost basis for the bitcoin held; TeraWulf's 10-Q states $44,547 before depreciation and $97,486 with it for the quarter ended Jun 30, 2026, so it was never among them.

Bitdeer's Bitcoin held in the operating figures read 257 as of Jul 31, 2026, the count in its July update, while its August update gives 61 at Aug 31, 2026. It now reads 61 for the month ended Aug 31, 2026, with the August update as its source, and the months of liquidity note that lists the bitcoin on file says the same. The home page's headline for Bitcoin held falls from 82,133 to 81,937 bitcoin from 13 companies and is now as of Aug 31, 2026, and the keyed API's figure for Bitdeer follows. A later row of Oct 9, 2026 for this page words that count by its month, August 2026, and supersedes the day given here.

Bitdeer's August 2026 production and operations update, filed Sep 16, 2026, gives "BTC Held (footnote 7): August 2026 61", with 257 for July 2026 and 1,934 for August 2025; its footnote 7 says "Bitcoins held does not include Bitcoins from customer deposits but does include Bitcoins that are pledged as collateral by us". The tile had been written from the July update, filed Aug 26, 2026, which the research pass of Sep 6, 2026 had read.

The cost line on CleanSpark's page said "Not disclosed: the filings carry bitcoin at fair value and tag no cost basis." for the average cost per bitcoin held and "Not disclosed" for the bitcoin per share and the cost per bitcoin per share. It now shows $90,188 per BTC for the 12,205 bitcoin held at Jun 30, 2026, 47.5 BTC per million shares and a cost basis per share of $4.29, and the Key figures, which listed the bitcoin held among the figures no source states, show the 12,205.

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 prints the cost basis in its note on bitcoin and tags none of it: "Number of bitcoin held 12,205 ... Cost basis - per bitcoin $90,188 ... Cost basis of bitcoin (in '000s) $1,100,759". It says the cost basis "represents the valuation of bitcoin at the time of initial recognition", and the 1,719 further bitcoin posted as collateral and carried as a receivable are not in the 12,205. The bitcoin per share divides the 12,205 by the 256,796,280 shares outstanding on the balance sheet, and the cost basis per share divides $1,100,759,000 by the same shares. The July operating update's 13,931 bitcoin is a different count for a later day.

The cost line on Core Scientific's page said "Not disclosed: the filings carry bitcoin at fair value and tag no cost basis." for the average cost per bitcoin held and "Not disclosed" for the bitcoin per share and the cost per bitcoin per share, beside 848 bitcoin held. It now shows $77,186 per BTC for the 848 bitcoin held at Jun 30, 2026, 2.65 BTC per million shares and a cost basis per share of $0.20, and the Key figures, which listed the bitcoin held among the figures no source states, show the 848.

Core Scientific's 10-Q for the quarter ended Jun 30, 2026 prints the cost basis in its note on digital assets and tags none of it: "The following table presents the Company's bitcoin holdings (in thousands, except for quantity): Quantity Cost Basis Fair Value June 30, 2026 848 $ 65,454 $ 49,675". The shares are the 319,587 thousand its balance sheet gives as issued and outstanding.

The cost line on Hut 8's page said "Not disclosed: the filings carry bitcoin at fair value and tag no cost basis." for the average cost per bitcoin held and "Not disclosed" for the bitcoin per share and the cost per bitcoin per share. It now shows $65,921 per BTC for the 17,316 bitcoin held at Jun 30, 2026, 141 BTC per million shares and a cost basis per share of $9.27, says the count is consolidated with American Bitcoin, and the Key figures, which listed the bitcoin held among the figures no source states, show the 17,316.

Hut 8's 10-Q for the quarter ended Jun 30, 2026 prints, in USD thousands: "Number of Bitcoin held as of June 30, 2026 17,316 ... Cost basis of Bitcoin held as of June 30, 2026 $ 1,141,496". It consolidates American Bitcoin: "we held approximately 17,316 Bitcoin, comprising approximately 9,314 Bitcoin held by Hut 8 and approximately 8,002 Bitcoin held by American Bitcoin", and the non-controlling interest "consists of 46.37% ownership by third parties in American Bitcoin". The figure is Hut 8's own statement of its consolidated balance sheet, not a subsidiary's release.

The cost line on Cipher's page said "Not disclosed: the filings carry bitcoin at fair value and tag no cost basis." for the average cost per bitcoin held and "Not disclosed" for the bitcoin per share and the cost per bitcoin per share, beside 646 bitcoin held. It now shows about $71,100 per BTC for approximately 646 bitcoin held at Jun 30, 2026, 1.56 BTC per million shares and a cost basis per share of $0.11, and the Key figures, which listed the bitcoin held among the figures no source states, show the 646.

Cipher's 10-Q for the quarter ended Jun 30, 2026 says: "The Company held approximately 646 and 1,433 bitcoin at June 30, 2026, and December 31, 2025, respectively. The associated fair values and cost bases of bitcoin held were $37.8 million, and $45.9 million, respectively, at June 30, 2026". Both inputs are rounded, so the quotient, $71,053, is read to the nearest $100 and the page says it is approximate.

The cost line on Bit Digital's page said "holds ether as well; the cost basis tag does not split it from bitcoin" for the average cost per bitcoin held, and the Key figures and the keyed API said its bitcoin held was not stated or "Not disclosed". It now shows $73,814 per BTC for the 21.5 bitcoin held at Jun 30, 2026, 0.0606 BTC per million shares and a cost basis per share of $0.0045.

Bit Digital's 10-Q for the quarter ended Jun 30, 2026 splits its bitcoin from its ether in a table in thousands of dollars: "Quantity Cost Basis Fair Value BTC 21.5 $ 1,587 $ 1,250 ETH 75,757.5 267,526 118,899". The figure uses the BTC row alone.

The cost line on Digi Power X's page said the CryptoAssetCost tag "holds the 10% price-sensitivity amount, not a cost basis" and that the company "states no bitcoin cost basis". The tag is still that amount, but the 10-Q states the cost basis in a note. The page now shows about $94,000 per BTC for approximately 217 bitcoin held at Jun 30, 2026, 2.20 BTC per million shares and a cost basis per share of $0.21, and the Key figures show 217 bitcoin held at Jun 30, 2026 in place of 133 at Dec 31, 2025.

Digi Power X's 10-Q for the quarter ended Jun 30, 2026 says: "The cost bases of the Corporation's holdings of digital currencies consist of: As at June 30, 2026 ... Bitcoin $ 20,387,947 ... Ethereum 3,020,828", and "the Corporation held a total of approximately 217 Bitcoins". The quotient, $93,954, is read to the nearest $100 because the filing says approximately and counts whole bitcoin.

The cost line on DarkHorse Technologies' page said "no bitcoin count on the same date as the cost basis" for the average cost per bitcoin held, and the Key figures said no source states its bitcoin held. It now shows $71,220 per BTC for the 20.5 bitcoin held at Jun 30, 2026, 2.68 BTC per million shares and a cost basis per share of $0.19.

DarkHorse Technologies' 10-Q for the quarter ended Jun 30, 2026 prints, in thousands except the number of Bitcoin: "June 30, 2026 December 31, 2025 Number of Bitcoin held 20.5 37.3 Carrying basis of Bitcoin $ 1,460 $ 3,835". It measures a sale's gain or loss against "the carrying basis of the Bitcoin as determined on a first in-first out (FIFO) basis", so the carrying basis is the amount the page calls a cost basis.

MARA's page showed $54,907 per BTC for the average cost per bitcoin held and a cost per bitcoin per share of $5.12. It now shows $54,662 per BTC and a cost basis per share of $5.09. The last of the three lines under the cost to mine, once called "Cost per bitcoin per share" on every company page, is now called "Cost basis per share", and its hover text says it is a cost basis and not the cost to mine.

MARA's 10-Q for the quarter ended Jun 30, 2026 gives the cost basis of its bitcoin on its own line: "Total bitcoin holdings 35,577 1,944,717 2,083,091" (in thousands except quantity), beside "Other digital assets 8,711" and "Total digital assets held as of June 30, 2026 $ 1,953,428". The old figure divided the cost of all digital assets, $1,953,428 thousand, by the bitcoin; the bitcoin alone is $1,944,717 thousand, which is $54,662 a bitcoin and $5.09 over its 381,888,004 shares. The label was wrong because the line is the cost basis of the bitcoin the company holds, and the cost to mine a bitcoin is a different figure the page gives beside it.

The cost line on Vulcan's page showed $100,439 per BTC for the average cost per bitcoin held, from 98 bitcoin at Mar 31, 2026, 6.04 BTC per million shares and a cost per bitcoin per share of $0.61, and the keyed API said its bitcoin held was "Not disclosed". It now shows $93,165 per BTC for the 103 bitcoin held at Jun 30, 2026, 5.72 BTC per million shares and a cost basis per share of $0.53, and the API serves the 103.

Vulcan's 10-Q for the quarter ended Jun 30, 2026 prints, in thousands except bitcoin held: "Number of bitcoin held 103.0 74.0 Carrying basis of bitcoin $ 9,596 $ 8,042", and says "The carrying basis represents the valuation of bitcoin at the time the Company earns the bitcoin through mining activities". The earlier figure was the Mar 31, 2026 quarter's, from a release's count of 98 bitcoin.

The cost line on TeraWulf's page showed $67,500 per BTC for the average cost per bitcoin held, 0.00401 BTC per million shares and a cost per bitcoin per share of $0.00, and its Operations row and the keyed API said its bitcoin count was "not disclosed". The three held figures are withdrawn, and the page says why. The Operations row and the API now give the 10-Q's count of 2 bitcoin held at Jun 30, 2026.

TeraWulf's 10-Q for the quarter ended Jun 30, 2026 states: "our digital asset balance was comprised of 2 bitcoin recorded at its fair value of $133,000", and its note on digital assets prints, in thousands except quantity of bitcoin, "Number of bitcoin held 2 3 Carrying basis of bitcoin $ 135 $ 269 Fair value of bitcoin $ 133 $ 270". A fair value of $133 thousand over 2 bitcoin is $66,500 a bitcoin, where CleanSpark's 10-Q for the same day gives $58,524, so the 2 is a count rounded to the whole bitcoin from about 2.3. A count that may be 1.5 to 2.5 puts a cost per bitcoin anywhere from $54,000 to $90,000, so none is published.

The cost line on IREN's page said "Not disclosed: the filings carry bitcoin at fair value and tag no cost basis." for the average cost per bitcoin held and "Not disclosed" for the bitcoin per share and the cost per bitcoin per share. It now says on each line that IREN holds no bitcoin, and the Key figures show 0 bitcoin held at Jun 30, 2026.

IREN's 10-K for the year ended Jun 30, 2026 says: "We typically liquidate all the Bitcoin we mine daily and therefore did not have any Bitcoin held on our balance sheet as of June 30, 2026." A company that holds no bitcoin has no cost basis for it, and "not disclosed" said it withheld one.

The cost line on Bitdeer's page said "Not disclosed: the filings tag no cost basis for bitcoin." It now says no cost per bitcoin held is given, and why: Bitdeer's statements for Jun 30, 2026 print a cost basis for 406 bitcoin but do not say whether those include the bitcoin it borrowed.

Bitdeer's 6-K of Sep 29, 2026, with its unaudited statements for Jun 30, 2026, prints in its note on digital assets "BTC (Bitcoin) 406 33,850 24,455" (units, cost basis and fair value, in thousands of USD except units) and, in its note on related parties, "an outstanding balance of 1,953.84 Bitcoin (US$117.5 million fair value equivalent)" borrowed from BIT Group. Its own production report counts 150 bitcoin held at Jun 30, 2026. The filing does not say how the 406, the 1,953.84 and the 150 relate, so a cost per bitcoin held from the 406 could be the cost of coins that are not Bitdeer's.

The cost line on HIVE's page said "Not disclosed: the filings carry bitcoin at fair value and tag no cost basis." for the average cost per bitcoin held, and its Key figures and the keyed API said its bitcoin held was not stated or "Not disclosed". The line now reads "Not stated: the 10-Q for the quarter ended Jun 30, 2026 carries its 190 bitcoin at fair value and prints no cost basis for them", and the Key figures and the API show the 190 bitcoin held.

HIVE's 10-Q for the quarter ended Jun 30, 2026 says "As of June 30, 2026, we held 190 Bitcoin" and carries its digital currencies at fair value: "Digital currencies, June 30, 2026 $ 11,065 190". It prints no cost basis for them. Its share count is not stored, so the bitcoin per share reads "Not computed".

The cost line on Argo's page said "Not disclosed: the filings tag no cost basis for bitcoin." It now reads "Not stated: the 20-F for 2025 gives the fair value of its bitcoin ($42 thousand at Dec 31, 2025), but no count and no cost basis."

"Not disclosed" says a company withholds a figure, which we do not know. Argo's 20-F for the year ended Dec 31, 2025 says "At the period end, the Group held Bitcoin representing a fair value of $42k" and prints neither a count nor a cost basis for it, so "not stated" says what was read.

The cost line on Big Digital Energy's page said "Not disclosed: the filings carry bitcoin at fair value and tag no cost basis." It now reads "Not stated: the 10-Q for the quarter ended Jun 30, 2026 shows no bitcoin on its balance sheet and prints no cost basis for bitcoin."

Big Digital Energy's 10-Q for the quarter ended Jun 30, 2026 carries no bitcoin on its balance sheet: its only cryptocurrency line, "Cryptocurrencies held for customers", is nil against $903,784 at Dec 31, 2025. "Not disclosed" said the company withheld a figure, and the 10-Q simply holds none.

The cost line on DMG Blockchain's page said "Not disclosed: the filings tag no cost basis for bitcoin." It now reads "Not stated: DMG Blockchain files with Canadian regulators and has no SEC filing on file to read a cost basis from."

Our company list records DMG Blockchain as filing in Canada only (SEDAR), with no EDGAR filer number, so there is no 10-Q or 20-F to read. "Not disclosed" said it had published none, and "tag" said a filing had been read.

The home page's sector average cost per bitcoin held read $64,876 per BTC across 5 of 18 miners. It now reads $69,080 per BTC across 11 of 18 miners, newest as of Jun 30, 2026, and its hover text names the 11.

The figure is still the counted miners' total cost basis over their total bitcoin held, now $5,567,524,947 over 80,595 bitcoin. The five were the miners whose filings tag a cost basis; the 10-Qs of CleanSpark, Core Scientific, Hut 8, Cipher, Bit Digital, Digi Power X and DarkHorse print one in a note. MARA's input is now its bitcoin alone, Vulcan's is its Jun 30, 2026 quarter, and TeraWulf (a count of 2 bitcoin rounded to the whole bitcoin) and Bitdeer (a count its 6-K cannot reconcile) are left out.

The home page's Bitcoin held headline read 81,816 BTC, from the 9 companies whose counts the research pass found. It now reads 82,133 BTC from 13, because the research files now carry TeraWulf's 2, Vulcan's 103, Bit Digital's 21.5 and HIVE's 190 bitcoin held at Jun 30, 2026, and its hover text says it is the sum of 13 of 18 companies' latest reported figures, where it said 9 of 18.

The four counts are those of the 10-Qs for the quarter ended Jun 30, 2026: TeraWulf "our digital asset balance was comprised of 2 bitcoin", Vulcan "Number of bitcoin held 103.0", Bit Digital "BTC 21.5" and HIVE "we held 190 Bitcoin". 81,816 + 2 + 103 + 21.5 + 190 is 82,132.5, which the tile prints as 82,133. The headline still adds the counts the research files give, as of the day each states, and leaves out a company with none.

The keyed API served TeraWulf's bitcoin held as "BTC count not disclosed" and Vulcan's, Bit Digital's and HIVE's as "Not disclosed by the company in the documents read on Sep 6, 2026", each with no value. It now serves 2, 103, 21.5 and 190 bitcoin held at Jun 30, 2026, each with the filing sentence it was read from and its document, and marks them as read from the stored row. The months of liquidity notes of Vulcan, Bit Digital and HIVE name the same counts. The fields of the response did not change.

The four counts are those of the 10-Qs for the quarter ended Jun 30, 2026: TeraWulf "our digital asset balance was comprised of 2 bitcoin", Vulcan "Number of bitcoin held 103.0", Bit Digital "BTC 21.5" and HIVE "we held 190 Bitcoin". The research pass of Sep 6, 2026 had read none of them.

The chart Stated mining cost against bitcoin price drew the average cost of the bitcoin held for 5 miners and said "No cost basis tagged" under the empty bar of the others, though the 10-Qs of CleanSpark, Core Scientific, Hut 8, Cipher, Bit Digital, Digi Power X and DarkHorse print one. It now draws that bar for 11 miners, from $54,662 for MARA to about $94,000 for Digi Power X, with CleanSpark's solid because its 10-Q prints the $90,188 itself and the others hatched. TeraWulf's $67,500 bar is withdrawn: its empty bar says "Not computed", IREN's says "Holds no bitcoin" and the rest say "No cost basis figure", each with its reason in the hover. TeraWulf and Bitdeer are named, with the reason from the filing, in a line of their own, "Average cost per bitcoin held not computed", and Bitdeer is no longer under "No figure stated". The method note under the chart now says the cost basis is the one a 10-Q prints in a table, not one it tags.

Each bar is the filing's cost basis over the bitcoin its own table counts. TeraWulf's 10-Q for the quarter ended Jun 30, 2026 prints "Number of bitcoin held 2 3 Carrying basis of bitcoin $ 135 $ 269", a count of 2 that is rounded to the whole bitcoin, so no cost per bitcoin is worked out from it. Bitdeer's 6-K of Sep 29, 2026 prints "BTC (Bitcoin) 406 33,850 24,455" and also "an outstanding balance of 1,953.84 Bitcoin" borrowed from a related party, without saying whether the 406 include any of it. IREN's 10-K says it "did not have any Bitcoin held on our balance sheet as of June 30, 2026". CleanSpark's 10-Q prints "Cost basis - per bitcoin $90,188".

CleanSpark's Key figures listed Bitcoin mined under "Not stated in the filings, releases or decks we read", beside a cost card that printed 1,920 bitcoin. The tile now reads 1,920 BTC for the quarter ended Jun 30, 2026, net of mining pool fees, and the table of every miner's metrics and the comparison chart carry it.

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 says: "During the three months ended June 30, 2026, we mined 1,920 bitcoin, net of mining pool fees, at an average bitcoin price of $71,881". The 1,925 in its cost table is the count before pool fees.

TeraWulf's Key figures listed Bitcoin mined under "Not stated in the filings, releases or decks we read", beside a cost card that printed 179 bitcoin. The tile now reads 179 BTC for the quarter ended Jun 30, 2026.

TeraWulf's 10-Q for the quarter ended Jun 30, 2026 says the company "significantly curtailed its bitcoin operations resulting in total bitcoin mined of 179 bitcoin during the three months ended June 30, 2026 as compared to 485 bitcoin mined during the same period in the prior year".

Vulcan's Key figures listed Bitcoin mined under "Not stated in the filings, releases or decks we read", beside a cost card that printed 24 bitcoin. The tile now reads 24 BTC for the quarter ended Jun 30, 2026, its own cryptocurrency mining, and says that a further 5 bitcoin were produced under its datacenter hosting and are counted apart.

Vulcan's 10-Q for the quarter ended Jun 30, 2026 prints, under Cryptocurrency Mining Metrics, "Bitcoins produced: Datacenter hosting 5 ... Cryptocurrency mining 24 ... Total bitcoins produced 29". Its cost to mine one bitcoin is computed "as cost of revenue of cryptocurrency mining divided by number of bitcoins produced from cryptocurrency mining", so the 24 is the count the cost uses.

Hut 8's Key figures listed Bitcoin mined under "Not stated in the filings, releases or decks we read", though its 10-Q states a count. The tile now reads about 935 BTC for the quarter ended Jun 30, 2026 and says it is counted for its ASIC Compute operations, which run through American Bitcoin, a subsidiary that outside holders own 46.37% of. The Key figures also say "about" before the 17,316 bitcoin held, which the 10-Q words as approximate.

Hut 8's 10-Q for the quarter ended Jun 30, 2026 says its ASIC Compute operations "operate generally through our majority-owned subsidiary, American Bitcoin", that its Compute revenue rose on "an increase in Bitcoin mined from approximately 308 to approximately 935", and that the non-controlling interest "consists of 46.37% ownership by third parties in American Bitcoin". It holds "approximately 17,316 Bitcoin".

Core Scientific's Key figures listed Bitcoin mined under "Not stated in the filings, releases or decks we read", beside a liquidity note that printed about 695 bitcoin. The tile now reads about 695 BTC for the 6 months ended Jun 30, 2026, the span its 10-Q states. The months of liquidity note in the comparison, the tier list and the keyed API said the bitcoin mined last quarter was not stated; it now says the quarter alone is not stated and that the 10-Q states about 695 for the six months.

Core Scientific's 10-Q for the quarter ended Jun 30, 2026 says: "During the six months ended June 30, 2026, we mined approximately 695 bitcoin, sold 2,385 bitcoin for aggregate proceeds of $208.2 million". It gives no count for the quarter alone.

IREN's Key figures listed Bitcoin mined under "Not stated in the filings, releases or decks we read", beside a cost card that printed 1,664 bitcoin. The tile now reads 6,075 BTC for the 12 months ended Jun 30, 2026, the newest count it states; the 1,664 for the quarter ended Dec 31, 2025, which its cost per bitcoin is for, stays in the row beside it.

IREN's 10-K for the year ended Jun 30, 2026 says its Bitcoin mining revenue "was generated from the mining and sale of 6,075 and 5,499 Bitcoin during the years ended June 30, 2026 and 2025, respectively". Its 10-Q for the quarter ended Dec 31, 2025 says "the mining and sale of 1,664 and 1,347 Bitcoin during the three months ended December 31, 2025 and 2024".

Argo's Key figures listed Bitcoin mined under "Not stated in the filings, releases or decks we read", beside a cost card that printed 150 bitcoin. The tile now reads 150 BTC for the 12 months ended Dec 31, 2025, and says the count includes rewards and transaction fees for mining other cryptocurrencies, which Argo generally exchanges for bitcoin.

Argo's 20-F for 2025 says: "In 2025 we mined 150 bitcoin and realized $103,473 per bitcoin", and defines Bitcoin Mined as including "rewards and transaction fees for mining Bitcoin and other cryptocurrencies", which it generally exchanges "into Bitcoin in the period in which such currency is earned".

Cipher's Key figures listed Bitcoin mined under "Not stated in the filings, releases or decks we read", beside a liquidity note that printed 346 bitcoin mined last quarter. The tile now reads about 346 BTC for the quarter ended Jun 30, 2026, from its business update deck, and says it was mined at its Odessa site. The Key figures also say "about" before the 646 bitcoin held and the 11.6 EH/s of hashrate, which its filing and deck word as approximate, and the months of liquidity note names the quarter instead of "last quarter".

Cipher's business update deck of Aug 4, 2026 says "~346 BTC Mined at Odessa in Q2 2026", and its 10-Q for the quarter ended Jun 30, 2026 says the quarter's revenue "was generated from bitcoin mining operations at the Odessa Facility", where the six months' sentence names the Odessa and Black Pearl Facilities. The deck of May 5, 2026 gave the same "~346 BTC Mined at Odessa" for Q1 2026, when Black Pearl mined as well, so that count is not used.

Digi Power X's Operations block said its bitcoin held had not been read from its reports, beside a Key figures tile that printed 217 BTC. It now lists the six figures it has not read and says its bitcoin held is stated above, the Key figures say "about 217 BTC" as the 10-Q words it, and the comparison and the home page's Bitcoin held headline count the 217.

Digi Power X's 10-Q for the quarter ended Jun 30, 2026 says: "As of June 30, 2026, the Corporation held a total of approximately 217 Bitcoins with an inventory value of $12,705,648".

DarkHorse Technologies' Operations block said its bitcoin held had not been read from its reports, beside a Key figures tile that printed 20.5 BTC, and the comparison said the same. It now lists the six figures it has not read and says its bitcoin held is stated above, and the comparison and the home page's Bitcoin held headline count the 20.5.

DarkHorse Technologies' 10-Q for the quarter ended Jun 30, 2026 prints, under Bitcoin holdings, "Number of Bitcoin held 20.5" for Jun 30, 2026.

The comparison's Bitcoin held row said "Not stated" for Digi Power X and DarkHorse Technologies, which the research pass did not cover, while their pages and the Companies page printed 217 and 20.5 bitcoin at Jun 30, 2026. The row now reads the same stored count as their pages, for the quarter ended Jun 30, 2026, with the 10-Q as its source.

Digi Power X's 10-Q for the quarter ended Jun 30, 2026 says it held "approximately 217 Bitcoins"; DarkHorse Technologies' prints "Number of Bitcoin held 20.5".

The home page's Bitcoin held headline counted 13 of 18 companies and left out Digi Power X (217 BTC) and DarkHorse Technologies (20.5 BTC), which the sector average beneath it counted. It now counts the 15 companies whose filings state a count, 82,174 BTC, and the line under the sector average says that the 11 miners it counts are the ones of those 15 that also print a cost basis. The hover text of that average named DarkHorse Technologies by its old symbol ANY and now says DRK, as every other page does. The table of every company showed Digi Power X's units held as 9 BTC at Dec 31, 2025, a tag its 10-K contradicts in its own text; the figure is held for review and the hover text gives the 10-K's words and the count the company states now. A later row of Oct 9, 2026 for this page words the line under the sector average, and a later row for the page of Digi Power X words the note on its tag of 9; each supersedes the sentences given here on those two.

Digi Power X's 10-K says: "As of December 31, 2025, the Corporation held a total of approximately 133 bitcoins with an inventory value of $11,812,321", and the 10-Q's table of digital currencies gives 133 bitcoin at Dec 31, 2025 and 217 at Jun 30, 2026. The tagged count of 9 is an error.

The asset list of the return tables named DarkHorse Technologies by its old symbol, ANY, beside every other asset's symbol. It reads DRK now, and the table's name for screen readers does too; the choice itself is still stored under the old key.

DarkHorse Technologies' report of Sep 17, 2026 says its common shares "began trading on the Nasdaq Capital Market under the new ticker symbol DRK at the market open on 17 September 2026"; no share consolidation came with it.

The months of liquidity lines of the tier list gave Bitdeer's bitcoin held as 257, which was its July count, though its August update gives 61, and said Bit Digital's, HIVE's and Vulcan's bitcoin held was not stated or was Vulcan's March count of 98. They read 61 for August 2026, 21.5, 190 and 103 at Jun 30, 2026, as the pages and the keyed API do, and Cipher's and Core Scientific's lines carry the counts of bitcoin mined with their periods. The analysis files these lines are copied into were regenerated.

Bitdeer's production update for August 2026 prints "BTC Held (footnote 7): August 2026 61"; Bit Digital's, HIVE's and Vulcan's 10-Qs for the quarter ended Jun 30, 2026 print 21.5, 190 and 103.0 bitcoin held.

On 15 company pages the Figure labels box under Operations said "Not disclosed means the company does not publish it." It now says "Not stated means no filing, release or deck we read gives it."

A company can publish a figure that none of the filings, releases and decks we read gives, so "not disclosed" said more than we know. The cost lines, the Companies table and the Compare page already said "not stated" for that reason; this box kept the old word.

The note under the cost table on the Companies page said "Above 100% is a loss per coin on that basis in that period", beneath columns that divide each cost by the bitcoin price of Sep 6, 2026, although Argo's 127% and 105% are for 2025 and IREN's 43% is for the quarter ended Dec 31, 2025. The note now says that against revenue per bitcoin mined, above 100% is a loss in that period, and that against spot it compares that period's cost with the Sep 6, 2026 price. The definition of the cost to mine on each company page, which said to set it against "the bitcoin price of the same period", now says the same. No number changed. A later row of Oct 9, 2026 for this page words the note again, and supersedes the sentence about a loss in that period given here.

The spot columns of the cost table divide each cost by one price, the spot price of Sep 6, 2026, which is later than the period of any cost they are set against, so a share of spot says how large a past period's cost is beside a later price and nothing of that period's profit. A loss in the period is the cost against the company's own revenue per bitcoin mined for the same period, as TeraWulf's 10-Q for the quarter ended Jun 30, 2026 prints it: "Value of each bitcoin mined(2) $71,704", with its cost to mine as 62.1 percent of it.

The chart Stated mining cost against bitcoin price said "Each bar is the number the company states." Of its 11 held bars, ten are drawn hatched because they are this site's division of the filed cost basis by the bitcoin held. It now says that each cost-to-mine bar is the number the company states, and that a held bar is the filed cost basis over the bitcoin held, hatched where it is worked out here and solid where the filing prints the figure itself.

CleanSpark's 10-Q prints the figure itself: "Cost basis - per bitcoin $90,188". The other ten 10-Qs print a cost basis and a count of bitcoin held, and the bar is the first divided by the second.

The two cost cards of the tier list printed "As of Jun 30, 2026" over IREN's quarter ended Dec 31, 2025 and Argo's 12 months ended Dec 31, 2025, and the cards for the change in share count and for fleet efficiency printed one day over figures of other days. A card whose figures are not all for one day and one length of period now says "Dates" and that each figure is for its own period, from the earliest day to the latest (the cost cards: "Each figure for its own period (12 months and 3 months), from Dec 31, 2025 to Jun 30, 2026"), and its chart prints no day of its own; each row gives its own length and day. A card whose figures share a day and a length still says "As of".

A bare "As of" day says that every figure under it is as of that day. IREN's 10-Q for the quarter ended Dec 31, 2025 prints "Net electricity costs per Bitcoin mined (3) (34.4)" for the three months ended Dec 31, 2025, and Argo's 20-F says: "In 2025 we mined 150 bitcoin and realized $103,473 per bitcoin".

The methodology page said the seven operating metrics a profile opens with "come from the research pass and are not on the compare tool". The Compare tool opens with them. The page now says they come from the research pass, except the cost per bitcoin, which is read from the stored cost rows, and a bitcoin held the pass has no count for, which is read from the stored bitcoin held row; that the Compare tool opens with them; and that it says in a company's column when the pass did not cover it. In its account of a metric the pass has no figure for, the page reads "not stated in what we read" where it read "not disclosed".

The Compare page opens with a block of operating figures for the companies chosen, and reads the bitcoin held of Digi Power X, 217, and of DarkHorse Technologies, 20.5, from the stored rows: Digi Power X's 10-Q says it held "approximately 217 Bitcoins" and DarkHorse Technologies' prints "Number of Bitcoin held 20.5".

The note under the Compare table said that operating figures come from the research pass. It now says that they do, except the cost per bitcoin, which is read from the stored cost rows, and a bitcoin held the pass has no count for, which is read from the stored bitcoin held row.

The Compare table reads Digi Power X's bitcoin held, 217, and DarkHorse Technologies', 20.5, from the stored rows, which the research pass of Sep 6, 2026 did not cover: Digi Power X's 10-Q says it held "approximately 217 Bitcoins" and DarkHorse Technologies' prints "Number of Bitcoin held 20.5".

The article said the cost article's chart "leaves a gap where those three bars would be", and that TeraWulf's revenue per bitcoin "prints not stated". The chart has two gaps for the cost with depreciation, MARA's and American Bitcoin's, and does not draw IREN's direct figure, which is for the quarter ended Dec 31, 2025 and not for the quarter of the other bars; and the Companies table prints TeraWulf's revenue per bitcoin, $71,704. The article now names the gaps the chart has and the figure it does not draw, and gives IREN, for which the table prints "No revenue on file", as its example of a missing figure. Its source for IREN's cost, the monthly update of June 2025, is now the 10-Q for the quarter ended Dec 31, 2025 that the stored figure comes from.

TeraWulf's 10-Q for the quarter ended Jun 30, 2026 prints "Value of each bitcoin mined(2) $71,704". IREN's 10-Q for the quarter ended Dec 31, 2025 prints "Net electricity costs per Bitcoin mined (3) (34.4)" and no revenue per bitcoin mined, so the table has nothing to set its cost against.

The chart Cost and revenue per bitcoin said "As of Sep 12, 2026", the day the article was written, over CleanSpark's and Riot's figures for the quarter ended Jun 30, 2026. It no longer prints that day, and its line says the figures are for one quarter, as filed; the note under the chart gives the quarter.

A day printed as a chart's is true of every figure under it, and the day of the article is the date of none of them. CleanSpark's 10-Q gives its figures for the quarter ended Jun 30, 2026: "Cost to mine one bitcoin - direct energy cost - Owned facilities $ 44,406" and "Average revenue of each bitcoin mined (1) $ 71,692".

The article ran words together in three places: "against71,692", "44,406in" and "IREN,quarter ended". They read "against 71,692", "44,406 in" and "IREN, quarter ended". Its two margin charts printed "As of Sep 12, 2026", the day the article was written, over figures for the quarter ended Jun 30, 2026 and cases worked out on the page; they print no day now, and the note under each gives its period.

The figures are CleanSpark's, from its 10-Q for the quarter ended Jun 30, 2026: "Average revenue of each bitcoin mined (1) $ 71,692" and "Cost to mine one bitcoin - direct energy cost - Owned facilities $ 44,406".

The Learn page's card for the article on why miners move more than bitcoin drew its chart under "As of Sep 12, 2026", the day the article was written, over figures for the quarter ended Jun 30, 2026 and two cases worked out on the page. The card's chart prints no day now.

The card's figures are CleanSpark's and Riot's for the quarter ended Jun 30, 2026, as filed, and two cases worked out from them; none is as of the day of the article.

The chart of the interest each financing route costs a year printed "As of Sep 12, 2026", the day the article was written, over rates and amounts from filings of 2025 and 2026, and so did the article's chart on the Learn page. They print no day now, and the note under each says its period.

The routes' sources are dated apart: the chart's own notes give IREN's 8-K of Aug 27, 2026 for the loan against its machines, MARA's prospectus supplement of Mar 28, 2025 for its at-the-market ceiling and TeraWulf's 8-K of Oct 30, 2025 for its convertible note.

The hover text on the average cost per bitcoin held and on the cost basis per share of the 11 companies that have one said "A cost basis is the amount the filing records the bitcoin at when the company got it." Bit Digital's 10-Q says otherwise, and seven of the eleven 10-Qs say nothing of how the amount is set. The hover text now says a cost basis is the amount a company's filing records its bitcoin at, and the note beside it gives the filing's own words for how it is set, or says the filing gives none. Bit Digital's: the post-impairment value of what it held at the adoption of ASU 2023-08 on Jan 1, 2024, and the valuation at the time it determined for revenue recognition for what it earned after. CleanSpark's: the valuation at initial recognition. Riot's: the initial value set for revenue recognition. Vulcan's: the valuation when it earns the bitcoin. MARA, Core Scientific, Hut 8, Cipher, Keel, Digi Power X and DarkHorse Technologies are said to give none. No figure changed.

Bit Digital's 10-Q says: "The cost basis is equal to the post-impairment value of all BTC and ETH held as of the adoption of ASU 2023-08 on January 1, 2024. For BTC and ETH earned subsequent to the adoption of ASU 2023-08, the cost basis of the BTC and ETH represents the valuation at the time the Company determined for revenue recognition purposes."

Under Method A and Method B on the workbench, the line that gives a company's own cost per bitcoin said only "Set the power price until yours matches, if you want the model to start from it." That figure may count the power bill alone, or whatever its own definition says, and is for the period named above it, not for the model's 12 months. The line now says so, and for Method B that Method B also counts SG&A, other costs and depreciation, which the company's figure may leave out.

IREN's 10-Q for the quarter ended Dec 31, 2025 prints "Net electricity costs per Bitcoin mined (3) (34.4)", electricity charges alone for one quarter; TeraWulf's 10-Q for the quarter ended Jun 30, 2026 prints "Cost to mine one bitcoin(1) $44,547" for one site. The model's cost is for 12 months and, under Method B, counts SG&A, other costs and depreciation.

TeraWulf's Key figures tile, its cost line and its cell in the Companies comparison gave $95,611 per BTC for the 12 months ended Dec 31, 2025, as the earlier rows of this page record. They now give $97,486 per BTC for the quarter ended Jun 30, 2026, which includes depreciation, leaves out SG&A and interest and covers its Lake Mariner Data Campus alone, and the hover text on the figure gives the company's direct cost for that quarter, $44,547. The figure for the quarter replaces the one for the year on the page, the comparison, the bars, the tier list, the charts and the keyed API, and the earlier rows describe the page as it stood when they were written.

TeraWulf's 10-Q for the quarter ended Jun 30, 2026, filed Aug 5, 2026, prints "Cost to mine one bitcoin(1) $44,547" in its table for the Lake Mariner Data Campus and says that factoring in the depreciation of its miner fleet would add $52,939 for the quarter, bringing the total "cost to mine one bitcoin" to $97,486. The newest period a company states is the one the page shows, and the 10-K's figure for 2025 is the older.

Argo's Key figures listed Cash and Fully diluted shares under "Not stated in the filings, releases or decks we read", while its Operations notes gave cash of $2.2M and 28.9B shares as on file. The tiles now read $2.2M of cash at Dec 31, 2025 and 2.2B shares, the weighted average for 2025, and the note on the shares says that its cover page counts 28,857,673,160 ordinary shares outstanding at Dec 31, 2025, after it cancelled and equitized its 8.75% Senior Notes.

Argo's 20-F for 2025 prints "Cash and cash equivalents 2,204" in its statement of financial position, in thousands of dollars, as at 31 December 2025, and "Weighted average number of ordinary shares in issue ('000) 2,188,075" for 2025, with its note that the potential dilutive ordinary shares "would be anti-dilutive" and so are not in the diluted count. Its cover page says: "28,857,673,160 Ordinary Shares" ... "were outstanding as of December 31, 2025".

The findings log of the analysis named DarkHorse Technologies by its old symbol, ANY, in the descriptions of its two price tests ("Over the last 365 sessions, ANY closed up on ...") and in the names of their rows, beside every other page's DRK. It reads DRK now. The log is written by the nightly analysis, which still keys its rows by the old symbol; the page shows the symbol the company trades under. A later row of Oct 9, 2026 for the Learn page on pooled quarter-to-quarter tests records a page that still named the company by the old symbol, and supersedes the words "beside every other page's DRK" given here.

DarkHorse Technologies' report of Sep 17, 2026 says its common shares "began trading on the Nasdaq Capital Market under the new ticker symbol DRK at the market open on 17 September 2026"; no share consolidation came with it.

The line under the sector average cost per bitcoin held said that the 4 miners the Bitcoin held headline counts and the average does not "state a count and print no cost basis for it", and its hover text said the 11 miners in the average are those "whose latest balance sheet gives both a cost basis for bitcoin and the bitcoin held". The line now says that the 4 state a count, but no cost per bitcoin held is formed for them, and the hover text no longer says what the other miners' balance sheets give.

TeraWulf's 10-Q for the quarter ended Jun 30, 2026 prints, in its table of bitcoin holdings (in thousands, except for quantity of bitcoin), "Number of bitcoin held 2" and "Carrying basis of bitcoin $ 135" for Jun 30, 2026. Bitdeer's unaudited financial statements for the six months ended Jun 30, 2026, filed on Form 6-K on Sep 29, 2026, print "BTC (Bitcoin) 406 33,850 24,455" under Units, Cost Basis and Fair Value at Jun 30, 2026. A cost per bitcoin held is not formed for either.

The note on Digi Power X's units held of 9 BTC at Dec 31, 2025, which are held for review, said that the 10-K's own text gives another count, "so the tag of 9 is an error", and its short line said the filing's own text gives another count. The note now says that footnote 1 to the 10-K's table of pay for key management says the company paid 9 BTC to officers and directors as compensation in the year ended Dec 31, 2025, and that the 10-K's text says it held approximately 133 bitcoins at that date; the short line says the tagged count is not shown until checked. The count stays held for review.

Footnote 1 to the table of remuneration of key management personnel in Digi Power X's 10-K for 2025 says: "During the year ended December 31, 2025 the Corporation paid 9 BTC (year ended December 31, 2024 - 26 BTC) as compensation for the services provided in by officers and directors". The 9 is tagged in the unit Bitcoin as of Dec 31, 2025 and the 26 as of Dec 31, 2024. The 10-K's text says: "As of December 31, 2025, the Corporation held a total of approximately 133 bitcoins with an inventory value of $11,812,321".

The comparison's Units held row for Digi Power X carried the same note as its page, that the 10-K's own text gives another count "so the tag of 9 is an error". It now says that footnote 1 to the 10-K's table of pay for key management says the company paid 9 BTC to officers and directors as compensation in the year ended Dec 31, 2025, and that the 10-K's text says it held approximately 133 bitcoins at that date. The count stays held for review.

Footnote 1 to the table of remuneration of key management personnel in Digi Power X's 10-K for 2025 says: "During the year ended December 31, 2025 the Corporation paid 9 BTC (year ended December 31, 2024 - 26 BTC) as compensation for the services provided in by officers and directors". The 9 is tagged in the unit Bitcoin as of Dec 31, 2025 and the 26 as of Dec 31, 2024. The 10-K's text says: "As of December 31, 2025, the Corporation held a total of approximately 133 bitcoins with an inventory value of $11,812,321".

The note under the comparison said of the companies the research pass does not cover that "every operational cell there says so instead of reading empty". It now names the columns the pass does not cover and says no more of their cells.

The comparison prints, in the Bitcoin held row, the counts of the two companies the pass does not cover: Digi Power X's 10-Q for the quarter ended Jun 30, 2026 says it held "approximately 217 Bitcoins" and DarkHorse Technologies' 10-Q prints "Number of Bitcoin held 20.5".

The months of liquidity lines of the tier list said "2,694 bitcoin mined last quarter" for Bitdeer and "BTC mined last quarter" in the formulas for CleanSpark and Riot, the list's definition of months of liquidity said "the bitcoin it mined last quarter", and Bitdeer's line gave its 61 bitcoin held "at Aug 31, 2026". They now read "2,694 bitcoin mined in the quarter ended Jun 30, 2026", "61 bitcoin held (August 2026)", "BTC mined" in the formulas, with the quarter in the clause after each, and "one quarter's direct cost per bitcoin times the bitcoin mined in that quarter, or that quarter's cost in dollars where the filing gives it that way, over three" in the definition.

Bitdeer's report for the second quarter of 2026, filed on Form 6-K on Aug 10, 2026, prints under Three Months Ended June 30 "BTC Mined (footnote 8)" as 2,694 for 2026 and 565 for 2025. Its production update for August 2026, filed on Form 6-K on Sep 16, 2026, prints "BTC Held (footnote 7): August 2026 61", and footnote 7 says "Bitcoins held does not include Bitcoins from customer deposits but does include Bitcoins that are pledged as collateral by us", with no date.

Bitdeer's Bitcoin held tile read "month ended Aug 31, 2026" and "As of Aug 31, 2026", and its months of liquidity note read "61 bitcoin held at Aug 31, 2026 and 2,694 bitcoin mined last quarter". They now read "August 2026", "As of August 2026" and "61 bitcoin held (August 2026) and 2,694 bitcoin mined in the quarter ended Jun 30, 2026", and the home page's headline for Bitcoin held names its newest day as August 2026.

Bitdeer's report for the second quarter of 2026, filed on Form 6-K on Aug 10, 2026, prints under Three Months Ended June 30 "BTC Mined (footnote 8)" as 2,694 for 2026 and 565 for 2025. Its production update for August 2026, filed on Form 6-K on Sep 16, 2026, prints "BTC Held (footnote 7): August 2026 61", and footnote 7 says "Bitcoins held does not include Bitcoins from customer deposits but does include Bitcoins that are pledged as collateral by us", with no date.

The note on CleanSpark's cost per bitcoin with depreciation said "No financing costs this quarter", and the formula of its months of liquidity said "BTC mined last quarter". The note now says "No financing costs in the quarter ended Jun 30, 2026", and the formula says "BTC mined" with the quarter in the clause after it.

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 prints, in its table of the cost to mine one bitcoin, "Financing costs per bitcoin mined" with a dash for the three months ended June 30, 2026 and 21 for the same months of 2025.

Digi Power X's Key figures read 23 BTC mined for the month ended Nov 30, 2025. They now read about 3 BTC for the quarter ended Jun 30, 2026.

Digi Power X's 10-Q for the quarter ended Jun 30, 2026 says: "For the three-month period ended June 30, 2026, Digi Power mined a total of approximately 3 Bitcoins as compared to a total of approximately 14 Bitcoins for the three-month period ended June 30, 2025".

DarkHorse Technologies' Key figures read 111.6 BTC mined for 2025. They now read 29 BTC for the quarter ended Jun 30, 2026.

DarkHorse Technologies' 10-Q for the quarter ended Jun 30, 2026 says: "During the second quarter of 2026 and 2025, we mined 29.0 and 30.9 Bitcoin, respectively" and "During the six months ended June 30, 2026 and 2025, we mined 54.3 and 61.3 Bitcoin, respectively".

The spread of Bitcoin held counted 13 of 18 miners and listed Digi Power X and DarkHorse Technologies under "Not on file" with the reason "no research pack carries this company". It now counts 15 of 18, with their 217 and 20.5 bitcoin from their 10-Qs for the quarter ended Jun 30, 2026, and the three it does not count (DMGGF, ARBK and BGDE) carry the reason their stored rows give, "No deck, release or filing on file states it in a form the pipeline reads.", in place of "Not disclosed by the company in the documents read on Sep 6, 2026." for two of them.

Digi Power X's 10-Q for the quarter ended Jun 30, 2026 says it held "approximately 217 Bitcoins", and DarkHorse Technologies' 10-Q prints "Number of Bitcoin held 20.5" for Jun 30, 2026.

The spread "Bitcoin held, as tagged in XBRL" listed 8 miners, Digi Power X's 9 BTC at Dec 31, 2025 among them. It lists 7: Digi Power X's tagged count is held for review, as its page says, and the pair of the XBRL count against Bitcoin held has no point for it.

Footnote 1 to the table of remuneration of key management personnel in Digi Power X's 10-K for 2025 says: "During the year ended December 31, 2025 the Corporation paid 9 BTC (year ended December 31, 2024 - 26 BTC) as compensation for the services provided in by officers and directors". The 9 is tagged in the unit Bitcoin as of Dec 31, 2025 and the 26 as of Dec 31, 2024. The 10-K's text says: "As of December 31, 2025, the Corporation held a total of approximately 133 bitcoins with an inventory value of $11,812,321".

The cost table's three cells for Argo (its revenue per bitcoin mined, and its cost with depreciation and its direct cost as shares of it) said "No revenue on file". They now read $103,473, 98% and 81%: $101,560 and $84,193 over $103,473, each computed here, for 2025.

Argo's 20-F for 2025 says: "In 2025 we mined 150 bitcoin and realized $103,473 per bitcoin as compared to 755 and $62,303 in 2024, respectively." The same figures over the spot price of Sep 6, 2026 are 127% and 105%, which the cost table's spot columns give.

The note under the cost table said that against revenue per bitcoin mined "above 100% is a loss in that period" and named Riot Platforms, CleanSpark and TeraWulf as having "mined at a loss in the period stated once depreciation is counted". It now says that against spot each cost is compared with the Sep 6, 2026 price, and that those three had a cost with depreciation above their revenue per bitcoin mined in the period and a direct cost below it. The definition of the cost to mine on each company page says that a share of the revenue per bitcoin mined is that cost as a share of what each bitcoin mined brought in.

CleanSpark's 10-Q for the quarter ended Jun 30, 2026 prints, for the three months ended June 30, 2026, "Average revenue of each bitcoin mined (1) $71,692" against "Direct cost to mine including non-cash depreciation and financing costs - Owned facilities $96,277" and "Cost to mine one bitcoin - direct energy cost - Owned facilities $44,406". TeraWulf's 10-Q for the same quarter prints "Value of each bitcoin mined(2) $71,704" against costs of $97,486 and $44,547. Both costs leave out SG&A, so neither is a profit or a loss of the company.

Argo's Key figures tile for its share count read "Fully diluted shares", 2.2B shares, "As of Dec 31, 2025", under the hover text "Shares outstanding plus those that warrants, notes and unvested awards could add". It now reads "Average shares", 2.2B shares, "12 months ended Dec 31, 2025", and its hover text starts "Average shares for the 12 months ended Dec 31, 2025". Its note still gives the 28,857,673,160 ordinary shares its cover page counts at Dec 31, 2025.

Argo's 20-F for 2025 prints "Weighted average number of ordinary shares in issue ('000) 2,188,075" for the year ended 31 December 2025 in its earnings per share note, and its cover page says "28,857,673,160 Ordinary Shares" were outstanding as of December 31, 2025.

The Key figures tile and the Companies table column for the share count of every company whose row is a weighted average (16 of 18, every row the file holds) read "Fully diluted shares" and "As of <day>". They now read "Average shares" with the quarter or the 12 months each average is for, and the note under the table says where the periods differ.

Each row is the filing's weighted average number of diluted shares (the tag for the weighted average number of diluted shares outstanding, or Argo's weighted average number of ordinary shares in issue) for the span of days its filing states, and none is a count of shares outstanding on a day or a pro forma fully diluted count.

The chart of income tax expense or benefit printed "As of Sep 11, 2026" over MARA's year to Dec 31, 2025 and CleanSpark's year to Sep 30, 2025. It prints no "As of" day, and its rows name their own year ends.

MARA's 10-K filed Mar 2, 2026 is for the year ended Dec 31, 2025 and CleanSpark's 10-K filed Nov 25, 2025 for the year ended Sep 30, 2025, as the two rows of the chart say; Sep 11, 2026 is the year end of neither.

The chart of hashrate targets and filed results printed "As of Oct 2, 2026" over targets stated in 2025 and results filed in January and February 2026. It prints no "As of" day, and its period line gives the two spans.

MARA's 75 EH/s target is from its June 2025 production update, a press release dated Jul 1, 2025, and its 66.4 EH/s result from an 8-K filed Feb 26, 2026; Bitdeer's 50 EH/s target is from a 6-K filed Dec 15, 2025 and its 55.2 EH/s result from a 6-K filed Jan 12, 2026.

The chart of IREN's installed hashrate printed "As of Oct 2, 2026" over its figure for June 2025 and its figure for June 2026. It prints no "As of" day, and its rows are named June 2025 and June 2026.

The 50 EH/s is from IREN's June 2025 monthly update, an 8-K filed Jul 7, 2025, and the 23.2 EH/s from its 10-K for fiscal 2026, filed Aug 27, 2026.

The two charts of the dilution article printed "As of Sep 6, 2026": one over share counts for the year ends 2020 to 2025 and the latest 10-Q cover pages, the other over a 424B5 of Mar 28, 2025 and a count at Dec 31, 2024. They print no "As of" day, and their period lines give the spans.

CleanSpark's latest count in the first chart is from its 10-Q cover page of Aug 4, 2026 and MARA's from its cover page of Jul 30, 2026; the second chart rests on MARA's 424B5 of Mar 28, 2025 and its 340,258,453 shares at Dec 31, 2024. Sep 6, 2026 is the date of none of them.

The two charts of the article on promised and delivered hashrate printed "As of Sep 6, 2026" over promises stated from December 2024 to December 2025 and outcomes filed from August 2025 to April 2026. They print no "As of" day, and their period lines give the spans.

The rows are dated by their own filings, from a 10-K filed Dec 3, 2024 to a 20-F filed Apr 30, 2026, as the notes on the chart's rows say; Sep 6, 2026 is the date of none of them.

The chart of the contract value of twelve datacenter leases printed "As of Sep 6, 2026" over leases announced from 2024 to August 2026. It prints no "As of" day, and its period line gives the span.

The notes on the chart's rows date the leases by their own filings, from an 8-K of Sep 25, 2025 to an 8-K of Aug 10, 2026, and one by a deck of Jul 28, 2026 that says it was contracted in 2024; Sep 6, 2026 is the date of none of them.

The chart of pooled revenue and cost changes named DarkHorse Technologies by its old symbol ANY, not its symbol DRK, on 17 of its rows (for example ANY 2019Q2). It names the company DRK on those rows.

DarkHorse Technologies' report of Sep 17, 2026 says its common shares "began trading on the Nasdaq Capital Market under the new ticker symbol DRK at the market open on 17 September 2026"; no share consolidation came with it.

The answer on what reported, computed and the hatched mark mean said that "a figure that is not disclosed is shown as words" and that a computed proxy stands in "where a company does not disclose the inputs". It now says that a computed proxy stands in where no filing, release or deck we read gives the inputs a figure needs, and that Not stated means no filing, release or deck we read gives it.

A company can publish a figure that none of the filings, releases and decks we read gives, so "not disclosed" said more than we know; the cost lines, the Companies table and the Compare page already said "not stated" for that reason.

The definition of months of liquidity on the tier list said that "each figure's note names the quarter and the filing". The sentence is gone; the note and the formula of each proxy give the quarter, and the definition still says that the monthly cost is one quarter's cost over three.

The notes of the proxies for CleanSpark, Riot and MARA name the quarter ended Jun 30, 2026, and IREN's the quarter ended Dec 31, 2025, in the clause after each formula; the definition says no more of them.

Cipher Digital's Power in development read 3,300 MW as "Guidance for Sep 30, 2026, as of Jun 8, 2026", with a guidance mark, on its company page and in the Companies comparison. It now reads 3,300 MW as of Jun 8, 2026, the day the presentation was filed, with no target day and no guidance mark. The number did not change.

The presentation filed Jun 8, 2026 says Cipher is well positioned to capitalize on AI demand with about 3.3 GW of pipeline sites, and gives no day for it. The September 2026 the page used is in the next bullet, which says the Black Pearl and Barber Lake construction projects are on schedule and tracking towards delivery of their initial capacity in September 2026. That date belongs to those two projects, not to the pipeline, so the pipeline figure is Cipher's statement as of the day it filed, not a target for September 2026.

Vulcan Infrastructure & Power's Power in development read 654 MW as "Guidance for Sep 30, 2027, as of Sep 10, 2026" on its company page and in the Companies comparison. It now reads "Guidance, as of Sep 10, 2026": the target day is gone and the figure is dated by the day the release was filed. The number did not change.

The release filed Sep 10, 2026 says Vulcan "has identified a 654 MW combined development pipeline across its owned sites" and gives no day for it. The third quarter of 2027 the page used is in the sentence before it, which says 40 MW at the Columbus, Mississippi site "is expected to become available by the third quarter of 2027". That date belongs to the 40 MW, not to the 654 MW.

Vulcan Infrastructure & Power's Power in development of 654 MW read "Guidance, as of Sep 10, 2026" on its company page and in the Companies comparison. It now reads "As of Sep 10, 2026", with no Guidance label. The number and the day did not change.

The release filed Sep 10, 2026 says Vulcan "has identified a 654 MW combined development pipeline across its owned sites". That sentence reports a pipeline Vulcan has identified and states no target or expectation for it. The Guidance label came from the sentence before it, which says 40 MW at the Columbus, Mississippi site "is expected to become available by the third quarter of 2027". The expectation belongs to those 40 MW, not to the 654 MW.

CleanSpark's Key figures tile was labelled "Cash cost per bitcoin" and showed $96,277 per BTC. It is now labelled "Cost per bitcoin" and says "All-in, includes depreciation". The number did not change.

The 10-Q for the quarter ended Jun 30, 2026 gives $96,277 as the direct cost to mine one bitcoin including non-cash depreciation and financing costs at the owned facilities, and $51,871 per bitcoin of it is miner depreciation. Depreciation is not a cash cost, so the figure is an all-in cost and "cash cost" was the wrong name for it.

MARA's Key figures tile was labelled "Cash cost per bitcoin" and showed $38,690 per BTC. It is now labelled "Cost per bitcoin" and says "Power cost only". The number did not change.

The Q2 2026 shareholder letter gives $38,690 as the purchased energy cost per bitcoin at MARA's owned sites: power paid to providers and nothing else. Hosting fees at its other sites, depreciation and every other cost are not in it, so "cash cost" overstated what the number covers.

Riot Platforms' Key figures tile was labelled "Cash cost per bitcoin" and showed $90,631 per BTC. It is now labelled "Cost per bitcoin" and says "All-in, includes depreciation". The number did not change.

The results release for the second quarter of 2026 gives $90,631 as the cost to mine one bitcoin including bitcoin miner depreciation. Depreciation is not a cash cost, so the figure is an all-in cost and "cash cost" was the wrong name for it.

TeraWulf's Key figures tile was labelled "Cash cost per bitcoin" and showed $95,611 per BTC. It is now labelled "Cost per bitcoin" and says "All-in, includes depreciation". The number did not change. A later row of Oct 9, 2026 for this page records the move to TeraWulf's figure for the quarter ended Jun 30, 2026, and supersedes the figure and the period given here.

The 10-K for 2025 gives $95,611 as the total cost to mine one bitcoin once $41,930 of depreciation per bitcoin is added to the direct cost. Depreciation is not a cash cost, so the figure is an all-in cost and "cash cost" was the wrong name for it.

IREN's Key figures tile was labelled "Cash cost per bitcoin" and showed $34,400 per BTC. It is now labelled "Cost per bitcoin" and says "Power cost only". The number did not change.

The 10-Q for the quarter ended Dec 31, 2025 prints $34.4 thousand as the net electricity cost per bitcoin mined: the cost of electricity for bitcoin mining divided by the bitcoin mined, with no hosting, depreciation or other cost in it. "Cash cost" overstated what the number covers.

The comparison of every miner on the canonical metrics named a column "Cash cost per bitcoin (USD per BTC)" and set CleanSpark's $96,277, Riot's $90,631 and TeraWulf's $95,611, which include machine depreciation, beside MARA's $38,690 and IREN's $34,400, which are the power bill alone. The column is now named "Cost per bitcoin (USD per BTC)", each figure says in words which kind it is, and a note above the table says the column mixes the two. The numbers did not change. A later row of Oct 9, 2026 for the page of TeraWulf records its move to the figure for the quarter ended Jun 30, 2026, and supersedes TeraWulf's figure given here.

A cost per bitcoin counts whatever each company chooses to count. The first three figures include depreciation, which is not a cash cost, and the last two count electricity only. Called cash costs and sorted in one column, they read as the same measure, which they are not.

The Hive page showed 2,885 bitcoin mined for the 12 months ended Mar 31, 2026 as its latest bitcoin mined figure. It now shows 1,004 bitcoin mined in the quarter ended Jun 30, 2026, and the Companies table prints 3 months beside the number.

Hive's results release for the first quarter of its fiscal 2027 says 1,004 bitcoin were received as rewards during the period, the quarter ended Jun 30, 2026. Hive's fiscal year ends on Mar 31, so that quarter runs April to June, but our reading took Q1 2027 for January to March 2027 and never read the sentence. The same reading dated each Hive fiscal quarter nine months too late.

CleanSpark's Key figures tile said "All-in, includes depreciation" under Cost per bitcoin ($96,277 per BTC). It now says "Includes depreciation and financing costs, excludes SG&A, owned sites only", and the hover text on the figure gives the company's own definition. The number did not change.

"All-in" said more than the figure counts. The 10-Q for the quarter ended Jun 30, 2026 gives $96,277 as the direct cost to mine one bitcoin including non-cash depreciation and financing costs, for the owned facilities, and the note on file says it leaves out SG&A, which is selling, general and administrative expense, where officer pay and overhead are booked. On this site an all-in cost counts SG&A, so this figure is not one. The earlier correction for this page, which called the figure an all-in cost, was wrong for the same reason.

MARA's Key figures tile said "Power cost only" under Cost per bitcoin ($38,690 per BTC). It now says "Power cost only, owned sites only", and the hover text on the figure gives the company's own definition. The number did not change.

The Q2 2026 shareholder letter gives $38,690 as the purchased energy cost per bitcoin "for our owned sites". The same letter says MARA produced 2,422 bitcoin in the quarter; its note on file puts 1,260 of them at owned facilities, about 52 percent, and says the hosting fees its other sites pay ($69.2 million in the quarter) are not in the figure. Words that said only "power cost only" left out that the figure covers about half of what MARA mined.

Riot Platforms' Key figures tile said "All-in, includes depreciation" under Cost per bitcoin ($90,631 per BTC). It now says "Includes depreciation, excludes SG&A and interest", and the hover text on the figure gives the company's own definition. The number did not change.

The results release for the second quarter of 2026 gives $90,631 as the cost to mine one bitcoin including bitcoin miner depreciation, and the note on file says the figure excludes SG&A and interest. "All-in" said more than that: on this site an all-in cost counts SG&A, where officer pay and overhead are booked, and this figure leaves it out. The earlier correction for this page, which called the figure an all-in cost, was wrong for the same reason.

TeraWulf's Key figures tile said "All-in, includes depreciation" under Cost per bitcoin ($95,611 per BTC). It now says "Includes depreciation, excludes SG&A and interest", and the hover text on the figure gives the company's own definition. The number did not change. A later row of Oct 9, 2026 for this page records the move to TeraWulf's figure for the quarter ended Jun 30, 2026, and supersedes the figure and the period given here.

The 10-K for 2025 gives $95,611 as the total cost to mine one bitcoin once $41,930 of depreciation per bitcoin is added to the direct cost, and the note on file says the company defines the figure itself and that it excludes SG&A and interest. "All-in" said more than that: on this site an all-in cost counts SG&A, where officer pay and overhead are booked, and this figure leaves it out. The earlier correction for this page, which called the figure an all-in cost, was wrong for the same reason.

The Cost per bitcoin column in the Companies comparison said "all-in, includes depreciation" under CleanSpark's $96,277, Riot's $90,631 and TeraWulf's $95,611 and "power cost only" under MARA's $38,690, and the note above the table said to compare only figures with the same words. Each cell now says what its own figure counts and leaves out: CleanSpark's "includes depreciation and financing costs, excludes SG&A, owned sites only", Riot's and TeraWulf's "includes depreciation, excludes SG&A and interest" and MARA's "power cost only, owned sites only". IREN's "power cost only" is unchanged. The hover text on each figure gives the company's own definition, and the note now says that figures with different words are not comparable and that figures with the same words can still differ in scope and period. The numbers did not change. A later row of Oct 9, 2026 for the page of TeraWulf records its move to the figure for the quarter ended Jun 30, 2026, and supersedes TeraWulf's figure given here.

The first three figures include depreciation, but each leaves out SG&A, two leave out interest and one counts financing costs, so "all-in" said more than they count. MARA's figure covers its owned sites only: 1,260 of the 2,422 bitcoin it produced in the quarter, about 52 percent. The old note was also wrong to say that figures with the same words can be compared: TeraWulf's is the figure its 10-K gives for the year 2025, as of Dec 31, 2025, while CleanSpark's and Riot's are for the quarter ended Jun 30, 2026, and the cost cells show no period.

The list of sources on the Methodology page and the Explore pages for TeraWulf's Lake Mariner and Lake Hawkeye sites credited New York State's address points to a state public record with no licence stated. They now credit the NYS Office of Information Technology Services and say CC0. The points, the pins and the distances did not change.

The state's own page for the data says: "Access to NYS Address Points is governed by Creative Commons 0 (CC0)". Our check had read only the item of the map service we ask, which states no licence of its own, and took that to mean none was stated. CC0 asks for no credit, so the line is a courtesy, and it now names the publisher and the licence.

The Against the sector chart on company pages drew Bitcoin mined against a sector median of 1,310 BTC taken over seven miners whose figures cover different periods: one month for Bitdeer, Digi Power X and Keel, three months for MARA Holdings and Riot, and 12 months for Hive and DarkHorse Technologies. MARA Holdings' 2,422 BTC for a quarter was drawn at about 1.8 times that median, and DarkHorse Technologies' 112 BTC for 12 months at about 0.09 times it. The chart now compares Bitcoin mined only with figures over the same period. No period has five miners yet, so that spoke is not drawn, and MARA Holdings and DarkHorse Technologies, left with two figures, no longer show the chart. A line under the table on the Companies page now says when the Bitcoin mined column mixes periods, and the column's hover text no longer says a month or a quarter only.

Bitdeer's monthly release of Sep 16, 2026 gives 1,310 BTC mined in August 2026, one month. MARA Holdings' shareholder letter for the quarter ended Jun 30, 2026 gives 2,422 BTC, a quarter. Hive's release for the year ended Mar 31, 2026 says 2,885 bitcoin were mined during the fiscal year, 12 months. The Companies table and each figure tile already named the period beside every number, but the chart's median ignored it, so it was the middle of unlike spans, not the typical figure for any one of them, and a longer span sat above a shorter one by length alone. The chart now takes the median only over miners that state the figure over the same span as the miner shown, with the five-miner minimum the other spokes use. No figure is converted: a year is never divided into months and no period is estimated.

Bit Digital's stake in WhiteFiber was given in the present tense on Oct 3, 2026, until the fix later that day: the Deals page said Bit Digital "owns about 69.6 percent" and the Bit Digital page said it "owns about 60 percent", with no day on either. They were one stake read on two days, and it moves whenever WhiteFiber issues shares. Both pages now date each figure: Bit Digital owned about 69.6 percent of WhiteFiber at Jun 30, 2026, and held 59.9 percent of its shares after Aug 21, 2026. The fall is dilution, not a sale: WhiteFiber issued new shares and Bit Digital's own 27,043,750 shares did not change. Two earlier entries of this log and some changelog entries of Oct 3, 2026 still say "owns" and are left as written; read them as the figure of the day they were written, not as a stake today.

Each figure is right for its day. Bit Digital's quarterly report for the quarter ended Jun 30, 2026 states about 69.6 percent. Its Schedule 13D/A on WhiteFiber, Amendment No. 1, filed Aug 25, 2026 with an event date of Aug 21, 2026, states 27,043,750 shares, about 59.9 percent of the class, down from 74.3 percent, and says the fall comes solely from the increase in WhiteFiber's shares outstanding and not from any sale. That filing is the one used because WhiteFiber is its issuer. Amendment No. 2, filed Aug 24, 2026 with an event date of Aug 18, 2026, is a different filing: its issuer is Bit Digital itself and it reports holders of Bit Digital's own shares, so it states no WhiteFiber stake.

Company pages printed a figure in dollars beside the unit USD, so CleanSpark's all-in cost read $96,277 USD per BTC. The same doubling showed on EV per EH/s and on revenue per MW, in the cards behind them, in the Companies cost table and in the compare tool. They now read $96,277 per BTC, $125.4M per EH/s and $683K per MW per year.

The figure already carries its dollar sign, so the unit USD after it said the currency a second time. No figure changed, only its wording. CleanSpark's all-in cost is $96,277 per bitcoin for the quarter to Jun 30, 2026, from its quarterly report filed Aug 6, 2026, and its revenue per MW is $683,198 a year, annualized from the same quarter, from its results release of the same day.

The evidence sentence shown for IREN's statement of Aug 27, 2026 (Bitcoin mining substantially wound down by Dec 31, 2026) was a piece of a table from the notes on derivatives in IREN's annual report for fiscal 2026, the fair value of its capped call transactions, which says nothing about Bitcoin mining. It now shows the sentence that does: "As of June 30, 2026, our installed Bitcoin mining capacity was approximately 23.2 EH/s, representing approximately 380MW of data center capacity."

IREN's annual report for fiscal 2026 prints that sentence twice, in its business discussion and in its management discussion. The row's outcome line says installed hashrate fell from 50 EH/s in June 2025 to 23.2 EH/s in June 2026, and the sentence is the report's own statement of the 23.2. The verdict on the row does not change.

The evidence sentence shown for CleanSpark's statement of Jul 14, 2026 (175 MW of critical IT load at Sandersville, deliveries from Q4 2027) opened with the exhibit's file labels and ran on through the release's title, headline and dateline, a run of text the release does not print as one passage. It now shows the two headline lines that report the outcome: "Ordered and pre-paid all long-lead items to meet Sandersville RFS date ... Anticipated equity portion of Sandersville project has been fully funded".

CleanSpark's results release of Aug 6, 2026 prints those two lines among its headline points, and the row's outcome line quotes the first of them. The row stays open: deliveries at Sandersville are not due before Q4 2027.

The evidence sentence shown for CleanSpark's statement of Feb 5, 2026 (up to 890 MW of utility capacity in the Houston region) opened with the exhibit's file labels and ran on through the release's title and dateline, a run of text the release does not print as one passage. It now shows the headline line that carries the figure the outcome cites: "Doubled MW under contract year-over-year including 585 MW of ERCOT-approved capacity".

CleanSpark's results release of May 11, 2026 prints that line as its first headline point. The row's outcome line gives the 585 MW. The verdict on the row does not change.

The evidence sentence shown for IREN's statement of Nov 6, 2025 ($3.4bn of annualized run-rate revenue by the end of 2026) opened with the exhibit's file labels, a run of text the release does not print. It now shows the highlight that carries the figure the outcome cites: "Current: $4bn contracted ARR for 2026 capacity".

IREN's results release of Aug 27, 2026 prints that line among its highlights, under Annualized run-rate revenue (ARR). The row's outcome line quotes the $4bn contracted ARR for 2026 capacity. The row stays unresolved until year-end and the verdict does not change.

The evidence sentence shown for Cipher's statement of Sep 25, 2025 (Barber Lake Phase I, 168 MW of critical IT load by September 2026) was a run of a quotation, a heading and several bullet points that is not in the release as printed. It now shows the one point that reports Barber Lake: "Barber Lake data center continues to progress towards completion as tenant has commenced beneficial use of the facility, including partial occupancy of the building and deployment of network racks".

Cipher's business update of Aug 4, 2026 prints that point under Finance and Operations Highlights. The row's outcome line quotes its words on the tenant's beneficial use and partial occupancy. The row stays open until the target date and the verdict does not change.

The evidence sentence shown for MARA's statement of May 11, 2026 (no large-scale ASIC purchases, hashrate roughly flat) was about MARA's repurchase of convertible notes, which says nothing about hashrate. It now shows the sentence that gives the hashrate: "As of June 30, 2026, we owned approximately 440,000 mining rigs globally, including our share of mining rigs from our equity method investee, the Abu Dhabi Global Markets company (the "ADGM Entity"), with an energized hashrate of approximately 70.3 exahashes per second ("EH/s")."

MARA's quarterly report for the period ended Jun 30, 2026 prints that sentence in the overview of its management discussion. The row's outcome line says energized hashrate went from 72.2 EH/s at Mar 31, 2026 to 70.3 EH/s at Jun 30, 2026, and the sentence is the report's own statement of the 70.3. The verdict on the row does not change.

The evidence sentence shown for Riot's statement of Mar 2, 2026 (the Kentucky Facility reaching a total capacity of about 232 MW) was about Riot's total hashrate and bitcoin mined, which says nothing about Kentucky capacity. It now shows the sentence that does: "The Kentucky Facility currently provides approximately 192 MW of developed capacity."

Riot's quarterly report for the period ended Jun 30, 2026 prints that sentence twice, in the note that describes the business and in its management discussion. The row's outcome line says 192 MW were developed at Jun 30, 2026, and the sentence is the report's own statement of it. The verdict on the row does not change.

The evidence sentence shown for Riot's statement of Jan 16, 2026 (25 MW of critical IT load to AMD, delivered from January to May 2026) was about a later 25 MW expansion with deliveries in November 2026 and May 2027, not the delivery the statement was about. It now shows the line that reports the delivery: "Completed delivery of the initial 25 critical IT MW to AMD on time and on budget".

Riot's results release of Aug 10, 2026 prints that line among its strategic highlights, and the row's outcome line quotes its words on time and on budget. The verdict on the row, delivered on its date, does not change.

The evidence sentence shown for IREN's statement of May 7, 2026 (480MW of AI Cloud capacity by the end of 2026) was a long run of highlights about annualized revenue and contract pricing, not about delivering capacity. It now shows the two highlights that report delivery: "Horizon 1 delivered to Microsoft ... Horizon 2 commissioning underway and Horizon 3-4 in late-stage construction, targeting delivery in Q4 2026".

IREN's results release of Aug 27, 2026 prints those lines among its highlights, and the row's outcome line cites Horizon 1 delivered and Horizons 3 and 4 targeting delivery in Q4 2026. The verdict on the row does not change.

The projection printed beside CleanSpark's statement of Nov 25, 2025 read "increase from 45.6 EH/s, the September 2025 final-month average (peak 50 EH/s in fiscal 2025)", and neither figure is in the quote shown. It now reads "continue increasing computing power through calendar year 2025 and beyond; no figure is stated".

The quote, from CleanSpark's annual report for fiscal 2025, says only that the company expects to continue increasing its computing power. The 45.6 EH/s is the average the report gives for Sep 30, 2025 and the 50 EH/s the peak it reached during the period: both are results reported, not figures the company said it would reach. The row's outcome line keeps the 45.6 EH/s as the figure the later readings are set against. The row was directional and stays so, so no score changes.

The projection printed beside MARA's statement of Apr 30, 2026 (an initial AI and critical IT buildout at Hannibal, Ohio) named up to 600 gross MW and about $144 million of annualized Adjusted EBITDA, and neither figure was in the quote shown. The quote now runs on into the next sentence of the release, the one that carries the 600 MW ("MARA also has multiple paths to expand capacity at the site to up to 600 gross MW over time ..."), and the projection no longer names the $144 million.

MARA's release of Apr 30, 2026 says, in the sentence after the one quoted, that MARA has multiple paths to expand the site to up to 600 gross MW over time and will pursue them. The $144 million is what the release says the plant MARA agreed to buy earned in the second half of 2025, a past result of that plant and not a figure MARA said it would reach, so it is out of the projection. The row stays a specific statement, still open, and its outcome line is unchanged, so no score changes.

The quote shown for IREN's statement of Nov 6, 2025 (more than $500m of AI Cloud ARR by the end of Q1 2026) ended with the year 2026, which IREN's release prints with a footnote number run on to it, as 20263. The quote now ends at the words by end of Q1, so it is whole words from the release, and the year stays in the projection and the target date.

IREN's release of Nov 6, 2025 prints the line as "... supporting growth to target AI Cloud ARR of >$500m by end of Q1 20263", where the 3 is a footnote marker run on to the year. A quote that stops after 2026 stops inside that printed number, so it cannot be told apart from a quote cut off inside a number. The figure of the projection, more than $500m, is in the quote as before.

The evidence sentence shown for CleanSpark's statement of Dec 3, 2024 (about 50 EH/s from 211.5 MW under development) ended in a closing quotation mark the release does not print there, and the row dated the results release Aug 7, 2025. The sentence now shows as the release prints it, and the date is Aug 8, 2025, the day EDGAR lists the filing.

In CleanSpark's results release the CEO's words run on after the word revenue ("... nearly $200 million in revenue. We also grew our bitcoin treasury ..."), so no quotation mark closes the sentence there. The release is dated Aug 7, 2025. EDGAR lists the 8-K (accession 0000950170-25-105347) as filed Aug 8, 2025: it was accepted late on the evening of Aug 7, after the cut-off for same-day filing. The other labels on the ledger give the filing date. The verdict on the row does not change.

The projection printed beside DMG's statement of Jun 1, 2026 read "first phase of the 50 MW AI colocation capacity delivered", but the 50 MW is not in the quote shown, and the ledger counted the statement as a specific promise. It now reads "first phase of the AI colocation capacity under the letter of intent delivered", and the statement is marked directional: shown, never scored. The Specific figure on the ledger now reads 52 of 77 statements instead of 53, and no company's score changes.

DMG's news release of Jun 1, 2026 gives 50 MW where it says DMG has signed a letter of intent with a prospective tenant to provide up to 50 MW, where it describes the proposed arrangement, and in one sentence that recalls a previously announced strategy to develop a 50 MW facility and sets no date. A letter of intent is not a commitment, and the release says there can be no assurance that a definitive agreement will be completed. The sentence quoted as the target, delivery of the first phase of capacity by Dec 31, 2026, names no figure. The row has no outcome document and was never selected, so no company's score changes; DMG's own tally of specific statements falls from two to one.

The Mechanism and evidence line under How this article reads said the chart "CleanSpark closes, rebased" was drawn from daily closing prices supplied by Polygon.io and Yahoo Finance. It now says Polygon.io alone. A sentence in the worked example said the first and last closes come from a cached file that a reader cannot check against a documented endpoint. It now says Polygon.io supplied both, and that a reader can check them against its documented endpoint with a free key.

The chart draws two closes, the first and the last of CleanSpark's latest 60 sessions, and the site's price history marks every close in those 60 sessions as Polygon.io's. Yahoo Finance supplied only earlier history, which this chart does not draw. The prices page counts the closes each supplier gave and says Yahoo Finance filled the earlier history once.

The Mechanism and evidence line under How this article reads said the chart "Return and price-level correlation" was drawn from daily closing prices supplied by Polygon.io and Yahoo Finance. It now says Polygon.io alone.

The chart compares CleanSpark and MARA Holdings over their last 365 shared sessions, and the site's price history marks every close of both companies in those sessions as Polygon.io's. Yahoo Finance supplied only earlier history, which this chart does not draw. The prices page counts the closes each supplier gave and says Yahoo Finance filled the earlier history once.

The worked example printed CleanSpark's first and last closes as 13 dollars each, then the divisions 13 / 13 = 1.0000 and 13 / 13 = 0.9713. It now prints each close to the cent, so on the data of Oct 2, 2026 the divisions read 12.89 / 12.89 = 1.0000 and 12.52 / 12.89 = 0.9713.

The price study records CleanSpark's close on Jul 9, 2026 as 12.89 dollars and on Oct 1, 2026 as 12.52 dollars, both supplied by Polygon.io, and 12.52 divided by 12.89 is 0.9713. The page had rounded each close to a whole dollar for display but divided the exact closes, so the division a reader could check by hand did not agree with the figures printed beside it.

The worked example printed the number of miners with annualized revenue on file as "$16", as in "There are $16 miners" and "Divide the sum by $16". It now prints 16.

The count is the number of miners that have an annualized revenue figure on file, 16 of the 18 in the universe. A count of miners is not a dollar amount, and the page had printed it in the dollar format used for the revenue figures beside it.

The worked example printed the number of miners in the annualized revenue distribution as "$16", in "covers $16 miners" and in "$16 minus one". It now prints 16.

The count is the number of miners that have an annualized revenue figure on file, 16 of the 18 in the universe. A count of miners is not a dollar amount, and the page had printed it in the dollar format used for the revenue figures.

The worked example printed the sums of squares of the log fit as SST 1.05, SSR 0.492 and SSE 0.553, then the check SSR plus SSE as 1.0455 and R squared as 0.4706. Divided by hand, 0.492 by 1.05 is 0.4686, not 0.4706. It now prints the three sums to three decimals, SST 1.045, SSR 0.492 and SSE 0.553, and the check and R squared to three decimals, 1.045 and 0.471, which agree with them.

The analysis output for hashrate against revenue on the base-ten logs records SST 1.045457, SSR 0.492021, SSE 0.553436 and R squared 0.470628. The page had shortened SST to three significant figures and the other two sums to three decimals, but computed the lines under them from the unrounded sums.

The worked example printed the first-quarter revenues of MARA Holdings and Riot Platforms as $230.7K and $1.4M and the latest as $174.9M and $174.2M, beside the divisions 758.0648 and 121.8427. Divided by hand, $174.2M by $1.4M is 124.4, not 121.8427. It now prints the four revenues in whole dollars: 230,694 and 174,881,000 for MARA Holdings, and 1,430,000 and 174,235,000 for Riot Platforms.

Each figure is a quarterly revenue filed with the SEC in whole dollars: MARA Holdings in its 10-Qs filed May 14, 2020 and Aug 6, 2026, and Riot Platforms in its 10-Qs filed May 8, 2020 and Aug 10, 2026. The page had shortened them to a few digits for display but divided the exact figures.

The worked example printed the Pearson r of the daily log returns as 0.84 and R squared as 0.70, which does not follow from 0.84: 0.84 times 0.84 is 0.71. It now prints the two correlations to three decimals, 0.838 for returns and 0.413 for price levels, the gap as -0.425, and R squared as 0.70, which follows (0.838 times 0.838 is 0.702).

The analysis output for CleanSpark and MARA Holdings over their last 365 shared sessions records r as 0.838427, the price-level r as 0.413015 and R squared as 0.70296. The page had rounded r to two decimals but squared the exact value.

The worked example printed the Pearson r of the pooled test as 0.59 and said that squaring it gives the same R squared, 0.34, but 0.59 times 0.59 is 0.35. It now prints r to three decimals, 0.586, and 0.586 times 0.586 is 0.34.

The analysis output for the pooled revenue and cost of revenue test records r as 0.586364 and R squared as 0.343823. The page had rounded r to two decimals before a reader could square it.

The worked example printed the p-value gap for the hashrate and revenue test as -0.1021, beside the ceiling 0.01 and the p-value 0.112, so a reader's subtraction gave -0.102. The gap lines now print to three decimals: -0.102 for that test, and 0.010 for the CleanSpark leverage test, which read 0.0100.

The findings log records the p-value of the hashrate and revenue test as 0.11207683. The page printed it to three decimals but subtracted the exact value and printed the gap to four.

The article quoted the IRS list of ways to dispose of a digital asset as ending "in exchange or trade for property, goods or services." The IRS list item reads "in exchange or trade for property, goods or services in any amount"; the quotation now includes the last three words.

The IRS Digital assets page, read again on Oct 3, 2026, lists "In exchange or trade for property, goods or services in any amount" among the ways to dispose of digital assets that make the Form 1040 digital assets answer yes.

The Prices note printed CleanSpark's first and last closes as 13 dollars each beside the division 13 / 13 = 0.97. It now prints 12.89 and 12.52 dollars, so on the data of Oct 2, 2026 it reads 12.52 / 12.89 = 0.97. The Correlation note printed the Pearson r of the daily log returns as 0.84 beside an R squared of 0.70, which does not follow (0.84 times 0.84 is 0.71); it now prints r as 0.838, and 0.838 times 0.838 is 0.70.

The price study records the two closes as 12.89 and 12.52 dollars, both supplied by Polygon.io, and the correlation of CleanSpark and MARA Holdings as r 0.838427 and R squared 0.70296. The notes had rounded the figures for display but computed the lines beside them from the exact values.

The Distributions note printed the number of miners with annualized revenue on file as "$16", in "across the $16 miners on file". It now prints 16.

The count is the number of miners that have an annualized revenue figure on file, 16 of the 18 in the universe. A count of miners is not a dollar amount, and the page had printed it in the dollar format used for the revenue figures.

The Time series note printed MARA Holdings' first-quarter and latest revenue as $230.7K and $174.9M beside the division 758.06; divided by hand, those two figures give 758.13. It now prints 230,694 and 174,881,000 dollars, which divide to 758.06.

Both figures are quarterly revenues filed with the SEC in whole dollars, in MARA Holdings' 10-Qs filed May 14, 2020 and Aug 6, 2026. The note had shortened them to a few digits for display but divided the exact figures.

The "Sector average cost per bitcoin held" card on the home page showed $80,892 per BTC across 5 of the 18 miners. With the same filings it now shows $64,876 per BTC across the same 5 miners.

The old figure was a plain average of the five miners' own cost-per-bitcoin figures, so a miner holding 2 bitcoin counted as much as MARA with 35,577. An average cost per bitcoin held has to weigh each miner by the bitcoin it holds, so the new figure is the five miners' total cost basis ($3,199,574,000) divided by their total bitcoin held (49,318). The inputs are unchanged: each miner's cost basis and bitcoin held from its latest 10-Q, four at Jun 30, 2026 and one at Mar 31, 2026.

Core Scientific's "Power in development" (Core Scientific page, summary tile, and the comparison table on the Companies page) showed at least 2,000 MW. That was wrong by about 12 times. The page now shows "not stated".

The figure paired the label "Uncommitted, Leasable Power" from one box of the July 28, 2026 earnings deck (slide 5, about 170 MW at Calvert City and Grand Forks) with the number "2 GW+" from a different box (new sites under due diligence for possible purchase). Neither box is power in development, and the deck states no figure for it. Our automated check and the final review both passed the old figure; both now require a quote to come from one box.

The worked example on this page, and the Correlation note on the Prices analysis page (/analysis/prices), printed the Spearman rho of the CleanSpark and MARA Holdings daily log returns as 0.83. They now print 0.832, to three decimals like the Pearson r beside it.

The price study records the Spearman rho of the two return series as 0.832067. The notes had rounded it to two decimals while printing the Pearson r to three.

The worked example gave MARA Holdings' and Riot Platforms' quarterly revenue as bare numbers. Each of the four now ends with its unit, US dollars.

The quarterly series records revenue in US dollars; the example printed the amounts without saying so.

The Bitdeer page showed 500 MW under Power in development. That figure was a possible expansion in Bhutan under a letter of intent that is not binding, so it is not power the company controls or has planned. The page now shows 206.5 MW, the AI cloud pipeline total in the company's September 2026 release, dated August 31, 2026.

The 500 MW came from the non-binding letter of intent. The 206.5 MW is the total of the AI cloud pipeline table in Bitdeer's own release of its August 2026 production, which is a figure the company states itself.

The Bit Digital page showed about 1.5 GW of power in development and about 11 MW online. Both come from a slide about WhiteFiber, of which Bit Digital owns about 60 percent, and the 1.5 GW was defined on the slide as sites for sale and under evaluation. Neither is shown now.

The investor slide that gave both figures describes WhiteFiber's business, not Bit Digital's own, and its footnote defines the 1.5 GW as data center sites for sale and under evaluation by management. The page has no way to show that the figures are WhiteFiber's, so both were withdrawn.

The Hut 8 page showed a cash cost of $36,500 per bitcoin. That figure is the cost of American Bitcoin, a Hut 8 subsidiary, from American Bitcoin's own release, and the quote stored with it held no number. It is no longer shown.

American Bitcoin's release of Aug 3, 2026 states a cost to mine of about $36,500 per bitcoin for American Bitcoin, not for Hut 8. The page presented it as Hut 8's own cost.

The Keel page showed 693 bitcoin mined, dated December 31, 2025. The company's filing gives 693 for the three months ended March 31, 2025. The page now shows 164 bitcoin mined in October 2025, the latest period the filings give.

The filing's 693 covers the three months ended March 31, 2025, a quarter that ended more than a year ago, so it was labelled with the wrong date. The 164 is the company's statement that it earned 164 bitcoin during October 2025, which is one month, not a quarter.

The CleanSpark page showed up to 885 MW under Power in development. The company's words are up to 885 MW of secured and planned power across its Texas portfolio, and our review could not confirm that as power in development, so it is hidden. A figure the review marks unclear is now hidden in the same way.

The 885 MW is a ceiling that mixes secured and planned power under a tenant letter of intent, so it is not a clear statement of power in development.

The Riot page showed 756 MW under Power in development, and for a while it showed no hashrate. The 756 MW is no longer shown. The 44.4 EH/s of hashrate, as of June 30, 2026, is shown again.

The 756 MW came from Riot's April 2026 deck. Its August 2026 deck replaces that plan with a letter of intent of about 1 GW at Corsicana, and a letter of intent is not power the company controls or has planned, so Riot has no figure under Power in development now. The hashrate was removed in error by an earlier check, because the number sat apart from its label in the slide text. Riot's August 2026 deck gives 44.4 EH/s under Hash Rate Deployed, and its appendix chart agrees for the second quarter of 2026.

The Cipher page dated its 11.6 EH/s of hashrate December 31, 2025, then lost the figure when a check withdrew it for that date. It now shows about 11.6 EH/s as of February 24, 2026. The 700 MW under Power under contract was dated December 31, 2025 and is now dated August 4, 2026.

Cipher's deck of February 24, 2026 states about 11.6 EH/s of total hashrate at Odessa, its only mining site, so that is the date the figure belongs to. The 700 MW is stated on Cipher's deck of August 4, 2026, and the earlier date came from a different document.

The Big Digital Energy page showed 300 MW under Power in development, and 120 MW under Power online. The 300 MW was wrong. The page now shows 129 MW online and 168 MW in development.

The 300 MW belonged to a site of SixThirty.AI, a separate company whose assets are held by the Endeavor Group, not to Big Digital Energy. The 120 MW covered one site. The company's presentation of June 3, 2026 gives its own assets as 129 MW operating and 168 MW in development. The 247 MW and 752 MW on another slide include SixThirty.AI's assets, so they are not used.

Four statements on the ledger were counted as settled although their own outcome line says they are not: Riot's of Mar 2, 2026 (the Kentucky Facility reaching about 232 MW), and IREN's of Aug 27, 2026 (Bitcoin mining wound down by Dec 31, 2026), May 7, 2026 (480MW of AI Cloud capacity by the end of 2026) and Nov 6, 2025 ($3.4bn of annualized run-rate revenue by the end of 2026). Each now shows as still open. The Settled figure on the ledger reads 43 of 77 instead of 47, and no company's score changes.

The ledger called a statement still open only when the first word of its outcome line was unresolved. These four lines say unresolved or pending at the end or after a first clause, so they were counted as settled with an unclear verdict. The rule now reads the whole line for either word. None of the four was scored, so no score or ring moves; the unclear verdicts counted over the settled rows fall from 22 to 18.

DMG's statement of Jun 1, 2026 no longer carries the not verified chip it had before its projection was corrected, and DMG's count of specific statements on the ledger fell from two to one.

The chip said the quote did not print the 50 MW named in the projection. With the 50 MW taken out of the projection, the quote prints every figure the projection names, so the ledger's check passes the row, and the same change marks the statement directional, which is why DMG's specific count fell.

The Hut 8 page still showed $36,500 per bitcoin mined, in the cost summary and as the direct cost figure, and a months of liquidity figure of 142.8 months that was divided by it. The $36,500 is the cost of American Bitcoin, a Hut 8 subsidiary. None of the three is shown now; the page says not disclosed.

American Bitcoin's results release of Aug 3, 2026 states a cost to mine of about $36,500 per bitcoin for American Bitcoin, not for Hut 8, and Hut 8 states no cost to mine one bitcoin. An earlier correction removed the figure from one place; it was still shown in two tiles and still fed the months of liquidity figure.

The Bit Digital deals table showed a 10-year, 40 MW agreement with Nscale worth $865 million as Bit Digital's, and counted it in the company's 45 MW of contracted or operating capacity and in the disclosed value. The row now shows no megawatts, term or value, and the company total reads 5 MW. The correction to the Deals page of the same day removed the Nscale row with the rest of Bit Digital's deals table, so neither that row nor a company total is on the page now.

Bit Digital's quarterly report for June 30, 2026 says that in November 2025 WhiteFiber's wholly owned subsidiary, Enovum NC-1 Bidco, LLC, entered into the services agreement with Nscale for colocation at its NC-1 facility. The agreement is WhiteFiber's. Bit Digital owns about 60 percent of WhiteFiber, and a subsidiary's contract is not the parent's.

The Hive page dated its 24.0 EH/s of hashrate Mar 31, 2026. The figure is unchanged and is now dated Jun 30, 2026.

Hive's results release of Aug 15, 2026 reports the first quarter of fiscal 2027, ended June 30, 2026, and gives the average operational hashrate during it as 24.0 EH/s. Hive's fiscal year ends on March 31, and the date had been read as if the quarter were a calendar quarter.

The Hut 8 page showed 8,660 MW under Power in development. The page now says not stated.

Hut 8's results release of Aug 4, 2026 gives a development pipeline of about 8,660 MW, of which 5,400 MW is Energy Capacity Under Diligence. Hut 8's quarterly report for June 30, 2026 defines that as greenfield opportunities where it invests limited development capital to assess a site's potential, and puts the sites where it has secured site control in the next stage, Under Exclusivity. So the 5,400 MW is neither power Hut 8 controls nor power it plans. The report gives no total for the stages that are controlled or planned (Exclusivity, Development and Construction), and the release and the report split Development and Construction differently (50 and 1,330 MW against 550 and 830 MW), so no figure is shown.

The Bit Digital page showed 40 MW under Power under contract and about $1 billion of Contracted revenue. Neither is shown now; the page says not stated.

Both figures belong to WhiteFiber, not to Bit Digital. The 40 MW is the agreement of WhiteFiber's subsidiary Enovum NC-1 Bidco, LLC with Nscale at the NC-1 campus. The remaining performance obligations in Bit Digital's quarterly report for June 30, 2026 are the same table as in WhiteFiber's own report, $1,008 million in total, so all of the contracted revenue is WhiteFiber's. Bit Digital consolidates WhiteFiber, but a subsidiary's figure is not the parent's. Bit Digital's quarterly report for June 30, 2026 puts its stake in WhiteFiber at about 69.6 percent. After WhiteFiber issued new shares in August 2026, Bit Digital's Schedule 13D/A (event Aug 21, 2026) puts it at 59.9 percent.

The Hive page showed 2,885 bitcoin mined as of Mar 31, 2026, which read like a quarter. The figure is unchanged and now reads 12 months ended Mar 31, 2026, and the Companies table prints 12 months beside the number.

Hive's results release for the full year ended March 31, 2026 says 2,885 bitcoin were mined during the fiscal year. The other bitcoin mined figures in the same table cover a quarter or a month, so a date alone let a year's total read as a quarter's.

The DarkHorse Technologies page (formerly Sphere 3D) showed 112 bitcoin mined as of Dec 31, 2025, which read like a point in time. The figure is unchanged and now reads 12 months ended Dec 31, 2025, and the Companies table prints 12 months beside the number.

The company's release of Mar 9, 2026, issued as Sphere 3D, says it mined 111.6 bitcoin during 2025, which the page prints rounded to 112. A year's total next to quarterly and monthly figures needs its span stated.

The Digi Power X page showed 23 bitcoin mined as of Nov 30, 2025, which read like a point in time. The figure is unchanged and now reads month ended Nov 30, 2025, and the Companies table prints 1 month beside the number.

Digi Power X's report of Dec 3, 2025 says miners running at its facilities produced approximately 23 bitcoin during the month, so the figure covers one month.

The worked example printed CleanSpark's and MARA Holdings' annualized revenue as "$766.3M US dollars a year" and "$699.5M US dollars a year", naming the unit twice. It now reads "$766.3M a year" and "$699.5M a year".

The figure already carries its dollar sign, so the words "US dollars" after it said the unit a second time. The two figures are annualized revenue in the analysis output of Oct 3, 2026: CleanSpark's from its 10-K filed Nov 25, 2025, and MARA Holdings' from its 10-Q filed Aug 6, 2026, times four.

The Distributions note printed the sum of squared deviations of annualized revenue as 1,273,929.94T and the variance it gives as 84,928.66T, figures a reader cannot read. It now prints 1.2739 x 10^18 and 8.4927 x 10^16 dollars squared, and names the divisor, 15.

Compact notation stops at trillions (T), so a figure of a quadrillion or more printed as thousands of trillions. The variance of the 16 miners' annualized revenue is 8.49 x 10^16 dollars squared, and 15 times it is 1.27 x 10^18. The note also divided the exact sum, so its quotient could differ in the last digit from the printed sum divided by 15; it now divides the printed sum.

The worked example printed the sample variance of annualized revenue as 84,900T. It now prints 8.49 x 10^16 dollars squared.

Compact notation stops at trillions (T), so a variance of 8.49 x 10^16 dollars squared printed as 84,900 trillion. A figure of a quadrillion or more now prints as a power of ten, with its unit.

The chart "Two findings against the p-value ceiling" printed the CleanSpark leverage test's p-value as 0 in its table and its hover labels, and the other two figures as 0.11 and 0.01. It now prints "under 0.0001", 0.112 and 0.010, the way the analysis tables print a p-value.

The findings log records the p-value of the CleanSpark leverage test over 365 sessions as 4.8e-42. A test never returns a p-value of exactly zero, and the analysis tables print any p-value under 0.0001 as "under 0.0001". The chart had printed every figure to two decimals, which rounds 4.8e-42 to 0.

The card for the article on tests flagged for an article printed the CleanSpark leverage test's p-value as 0. It now prints "under 0.0001", and the other two figures as 0.112 and 0.010, where they had read 0.11 and 0.01.

The findings log records the p-value of the CleanSpark leverage test as 4.8e-42, and a test never returns a p-value of exactly zero. The card draws the same chart as the article, which printed every figure to two decimals.

The chart "Returns after a low-RSI signal" labelled its bars "On, 1059 sessions" and "Off, 22968 sessions" and printed the two mean forward log returns as 0.03 and -0.02. It now reads "On, 1,059 sessions" and "Off, 22,968 sessions" and prints 0.025 and -0.022, the figures the worked example gives. The worked example printed CleanSpark's exponential averages as 13.4 and 13.3 beside a MACD line of 0.128, with no subtraction. It now prints 13.414 and 13.286 and the line 13.414 - 13.286 = 0.128. The session counts and the figures are those of the analysis output of Oct 3, 2026; they move with each nightly refresh.

The indicator study records the mean forward log return after a low-RSI signal as 0.025165 over 1,059 sessions and as -0.021805 over 22,968 other sessions. The fixture records CleanSpark's averages of Oct 1, 2026 as 13.414195 and 13.286065, whose difference is the MACD line, 0.12813. The chart had printed counts without a thousands separator and the means to two decimals, and the example had shortened the averages so that 13.4 less 13.3 did not give the MACD it printed.

The Bit Digital deals table showed a five-year, 5 MW colocation agreement with Cerebras Systems, first disclosed in March 2025, as Bit Digital's, and it made up the company's contracted capacity total of 5 MW. The row is removed, and Bit Digital no longer has a total.

The agreement is WhiteFiber's. WhiteFiber's quarterly report for June 30, 2026 says its MTL-3 data center in Quebec supports the contracted Cerebras capacity, 5 MW of IT load, and that it has billed Cerebras since Nov 1, 2025 for the five-year contract. Bit Digital owned a majority of WhiteFiber, about 69.6 percent at June 30, 2026 by its own quarterly report, which is not the same as being the party to the agreement.

The Bit Digital deals table showed more than $540 million of new multi-year cloud services agreements, dated Aug 13, 2026, as Bit Digital's, and counted the $540 million in the disclosed value. The row is removed.

Bit Digital's own results release of Aug 13, 2026 says that "WhiteFiber has also signed new multi-year Cloud Services agreements representing more than $540 million of aggregate contract value over their initial terms". The agreements are WhiteFiber's, not Bit Digital's.

The Bit Digital deals table showed $175 million of cloud contracts, dated May 2025, as Bit Digital's, and counted it in the disclosed value. The row is removed. The date was wrong as well: the contracts are dated May 2026.

On the slide on WhiteFiber's growth journey in Bit Digital's investor deck of Jun 30, 2026 the new cloud wins stand under May 2026, and the key takeaways say "May 2026 wins represent >$175MM of recent cloud contract value". They are WhiteFiber's contracts. The row had taken May 2025 from the next line of the slide, which dates WhiteFiber's purchase of NC-1.

The Bit Digital deals table showed a $20.2 million agreement with DNA Fund, for 576 H200 GPUs over 25 months, as a contracted agreement of Bit Digital's. The row is removed.

The agreement is WhiteFiber's. WhiteFiber's annual report for 2025 says WhiteFiber entered into the master services agreement with DNA Fund on Dec 30, 2024, and that in November 2025 it terminated the agreement and all related purchase orders, so the status contracted was out of date as well. The row was dated before the 18-month window, so the figures inside the window did not move for it.

The Bit Digital deals table showed an 18-month agreement to supply the company's first cloud customer with an additional 464 B200 GPUs, first disclosed in March 2025, as a contracted agreement of Bit Digital's. The row is removed.

The agreement is WhiteFiber's. WhiteFiber's quarterly report for June 30, 2026 says that in January 2025 it entered into an agreement to supply its Initial Customer with an additional 464 GPUs for a period of 18 months. The same report says WhiteFiber signed a termination agreement with that customer in the second quarter of 2026 and lists the customer among those whose agreements were discontinued, so the status contracted was out of date as well.

The Bit Digital deals table showed a 24 MW colocation agreement at NC-1 in North Carolina, dated Feb 4, 2026, as Bit Digital's. The row is removed.

NC-1 is the campus of WhiteFiber's subsidiary Enovum NC-1 Bidco, LLC, which Bit Digital's quarterly report for June 30, 2026 describes as providing colocation at its NC-1 facility. The 24 MW is one line on the WhiteFiber slide of Bit Digital's investor deck of Feb 4, 2026, which gives its megawatts on a gross basis and names no tenant, so it is not an agreement of Bit Digital's.

With Bit Digital's seven rows removed, the Deals page counts 49 rows in the book where it counted 56, and 47 rows inside the 18 months where it counted 51. Companies with at least one row fall from 13 to 12, and companies with a signed or operating agreement from 12 to 11, so the headline now reads 11 of the 18 miners. Companies checked and found to have none stay at 5. Bit Digital is listed apart: it has no agreement in its own name, but WhiteFiber, which it consolidates, holds agreements, so it is not counted among them. The disclosed contract value inside the 18 months falls by $715 million, of which $540 million was announced in July and August 2026. Bit Digital leaves the contracted capacity ranking, where it had stood with 45 MW and, after an earlier correction, last with 5 MW. The row for the Nscale agreement, which had stayed on the table with no figures after that correction, is removed with the rest, and the company page, which read 7 HPC and AI rows on file, 7 signed or operating, now says no HPC or AI agreement of Bit Digital's own is on file.

Bit Digital's quarterly report for June 30, 2026 puts its stake in WhiteFiber at approximately 69.6 percent, says it consolidates WhiteFiber, and says WhiteFiber's subsidiary signed the Nscale agreement. WhiteFiber's own reports describe the Cerebras, DNA Fund and first-customer agreements, and Bit Digital's own release and decks present the $540 million, the $175 million and the 24 MW at NC-1 as WhiteFiber's. None of the seven rows is an agreement of Bit Digital's own, so the Deals page totals are recomputed from the rows that remain.

The Bit Digital page listed a statement of Aug 13, 2026 that its Cloud Services portfolio is expected to generate more than $200 million of annualized revenue once fully deployed. The statement is removed, and Bit Digital no longer has an entry on the guidance ledger. The ledger now holds 74 statements from 17 miners where it held 75 from 18.

The sentence is about WhiteFiber, not Bit Digital's own business. In Bit Digital's results release of Aug 13, 2026 its chief executive speaks of WhiteFiber, says WhiteFiber has signed new Cloud Services agreements, and then says that, based on contracts signed to date, "its Cloud Services portfolio" is expected to generate more than $200 million of annualized revenue once fully deployed. Bit Digital's quarterly report for the quarter ended June 30, 2026 states that it owned about 69.6 percent of WhiteFiber, and its Schedule 13D/A with an event date of Aug 21, 2026 states 59.9 percent of the class, with its own 27,043,750 shares unchanged. Owning most of a company is not the same as that company's figure being Bit Digital's own.

The test tables printed each sum of squares and variance past a quadrillion as thousands of trillions. For annualized revenue against realized hashrate the total sum of squares read 651,000T, the part the line explains 185,000T, the part left over 466,000T, and the variance of annualized revenue 72,300T. The three tables that take market capitalization as the x figure printed its variance the same way, up to 23,800,000T, and the table against EV per EH/s printed its sums of squares that way too. Each now prints as a power of ten: 6.51 x 10^17, 1.85 x 10^17, 4.66 x 10^17 and 7.23 x 10^16 for the first table.

Compact notation stops at trillions (T), so a figure of a quadrillion or more printed as thousands of trillions, which a reader cannot read. A sum of squares or a variance of a dollar figure is a square, so these run to 10^17 and 10^19. In the analysis output of Oct 6, 2026 the total sum of squares of annualized revenue on realized hashrate is 6.51 x 10^17 dollars squared and its variance is 7.23 x 10^16 dollars squared; the figures move with each nightly refresh.

The sample statistics under each distribution printed the variance of seven figures as thousands of trillions: 84,900T for annualized revenue, 2,200,000T for cash, 1,930,000T for annualized net income, 20,900,000T for market capitalization, 19,400,000T for total assets, 48,100,000T for shares outstanding and 173,000T for EV per EH/s. They now print 8.49 x 10^16, 2.20 x 10^18, 1.93 x 10^18, 2.09 x 10^19, 1.94 x 10^19, 4.81 x 10^19 and 1.73 x 10^17.

Compact notation stops at trillions (T), so the variance of a dollar or share figure, a square of figures that run to billions, printed as thousands of trillions. The variance of the 16 miners' annualized revenue in the analysis output of Oct 6, 2026 is 8.49 x 10^16 dollars squared; the figures move with each nightly refresh.

The hover text on each point of the pair charts named the unit twice: CleanSpark's annualized revenue read "$766.3M USD per year". The hover text on each row of the distribution charts did the same, and read "+72.3% percent of revenue" for the highest gross margin. They now read "$766.3M per year" and "+72.3% of revenue".

The figure already carries its dollar or percent sign, so the unit in words after it said the unit a second time. CleanSpark's annualized revenue is $766.3M for the year ended Sep 30, 2025, from its 10-K filed Nov 25, 2025.

The hover text on each point of the chart of the strongest fit named the unit twice: CleanSpark's annualized revenue read "$766.3M USD per year". It now reads "$766.3M per year".

The figure already carries its dollar sign, so the words USD per year repeated the unit. CleanSpark's annualized revenue is $766.3M for the year ended Sep 30, 2025, from its 10-K filed Nov 25, 2025.

The hover text on the bars of the article cards named the unit twice. On the cost card CleanSpark's direct cost read "$44,406 USD per bitcoin"; it now reads "$44,406 per bitcoin", and every other dollar bar on the cards reads the same way.

The figure already carries its dollar sign, so the words USD per bitcoin repeated the unit. CleanSpark's direct cost is $44,406 per bitcoin for the quarter to Jun 30, 2026, at its owned facilities, from its quarterly report filed Aug 6, 2026.

The guidance ledger said that its statements, its Generated date and the date on its headline figures were as of Sep 6, 2026, although some statements had been corrected or re-graded on Oct 2 and Oct 3, 2026. The page now shows Oct 3, 2026, the newest day any statement on it was read or corrected.

The ledger's date was written once, when the statements were first checked, and was not moved when later corrections changed statements. It now comes from the statements themselves. No statement, score or count changed.

From Oct 2, 2026 the Explore page offered a 3D flight over Core Scientific's Marble, North Carolina site. On Oct 3 the changelog said the flight was removed because Google's 3D view there is only coarse terrain, not a model of the site. That reason was wrong. The measurement behind it aimed its camera at sea level, and the ground at Marble is high enough that the camera looked over the site. Measured again with the camera aimed at the ground at the pin, the finest tile Google serves there is about 4 m across, with 14,970 triangles in each of three readings, which is a detailed model. The flight stays off: the view at the pin shows a country road beside a very large windowless building, and the two people who judged it could not tell whether that is the site.

The first coverage check took one reading per site, from a camera that could drift during a reading, and counted a 2 m tile with 4,044 triangles at Marble as a 3D model. The second held the camera still but aimed it at sea level, so on high ground it read a view that never held the site and found terrain only. The third aims at the pin's own ground height and is the one the site now uses. Google's tiles show whether a 3D model exists near a pin, not whether the pin sits on the site, so a flight also needs a person to have looked at the view and recorded what it shows.

From Oct 2, 2026 the Explore page offered a 3D flight over Keel's Scrubgrass site near Kennerdell, Pennsylvania. On Oct 3 the changelog said the flight was removed because Google's 3D view there is only coarse terrain, not a model of the site. That reason was wrong. The measurement behind it aimed its camera at sea level and looked over the site, which stands on high ground. Measured again with the camera aimed at the ground at the pin, the finest tile Google serves there is about 4 m across, with 21,705 triangles in each of three readings, which is a detailed model, and the view shows a power plant at the pin. The flight stays off until a person has watched it run, because it starts from a default ground height that has not been tested at a site this high.

The first coverage check took one reading per site, from a camera that could drift during a reading, and counted a 2 m tile with 4,885 triangles at Scrubgrass as a 3D model. The second held the camera still but aimed it at sea level, so on high ground it read a view that never held the site and found terrain only. The third aims at the pin's own ground height and is the one the site now uses. Two people who judged its view without knowing the site saw a power plant with a boiler house and coal conveyors, and rows of container-sized units on a gravel pad about 100 to 150 m to the north.

The page said IREN's June 2025 figure of $26,259 per bitcoin was the last one the company disclosed and that its fiscal 2026 filings give only an aggregate cost of revenue. It now shows IREN's net electricity cost per bitcoin mined of about $34,400 for the three months to Dec 31, 2025, and says that this figure is electricity only. The same figure now feeds the cost-per-bitcoin dataset and the cost comparison chart.

IREN's 10-Q for the quarter ended Dec 31, 2025 (accession 0001878848-26-000015, filed Feb 5, 2026) prints "Net electricity costs per Bitcoin mined" of 34.4 thousand dollars for the three months and 35.0 thousand for the six months. The 10-Q for Sep 30, 2025 (0001878848-25-000081, filed Nov 6, 2025) prints 35.4 thousand. The 10-Q for Mar 31, 2026 and the 10-K for fiscal 2026 do not print the figure.

The article said IREN last published a cost per bitcoin for June 2025, and that its fiscal 2026 filings give only an aggregate cost of revenue. It now says IREN's quarterly report filed Feb 5, 2026 gives a net electricity cost per bitcoin mined of about $34,400 for the three months to Dec 31, 2025, and that its quarterly report filed May 8, 2026 and its annual report filed Aug 27, 2026 give no cost per bitcoin.

IREN's annual report filed Aug 28, 2025 (accession 0001878848-25-000063) and its quarterly reports filed Nov 6, 2025 (0001878848-25-000081) and Feb 5, 2026 (0001878848-26-000015) each give a net electricity cost per bitcoin mined: $25,642 for the year to Jun 30, 2025, about $35,400 for the three months to Sep 30, 2025 and about $34,400 for the three months to Dec 31, 2025. June 2025 was not its last such figure. The two later reports show the cost of revenue for bitcoin mining as a total and no cost per bitcoin.

The article's chart scored Bitfarms (now Keel Infrastructure) as short of its 18 EH/s target, with 17.7 EH/s at Jun 30, 2025, and counted two of three results below target. Bitfarms is now left out of the chart and the article, which compare MARA and Bitdeer only.

Bitfarms' management discussion for the second quarter of 2025, filed Aug 12, 2025 (accession 0001213900-25-074626), says the company achieved its 18 EH/s operational target in March 2025. The target was met before Jun 30, 2025, so 17.7 EH/s at that date was a later reading and not a miss.

The table of MARA's 75 EH/s target gave Jul 1, 2025 under the heading Filed, and the article said its company figures came from filings. The document that states the target, the June 2025 production update, is a press release on MARA's website, and it was not in the sources. The heading now reads Dated, the document is named as a press release and linked in the sources, and the article says documents instead of filings.

MARA filed nothing with the SEC on Jul 1, 2025. The press release of that day says MARA is targeting 75 exahash by the end of 2025.

The article listed In review among the words printed where a metric has no filed value. It now says In review is printed where a filed value exists but failed a check against the previous period, and that the value is held back until it is checked.

The site prints In review only after it has found a filed value to hold back, as the methodology page says. A metric with no filed value prints other words, such as Not tagged or No SEC data.

The article said each filed figure on company pages carries a stale badge once its period ended more than 400 days before the as-of date. It now says this holds for figures read from the tagged data in a company's SEC filings, such as revenue, cash and shares outstanding, and that operating figures such as hashrate and cost per bitcoin show their period and source but no age badge.

The stale badge is drawn only for tagged SEC data. Operating figures come from a separate research pass and carry no badge: IREN's fleet efficiency, dated Jun 30, 2025, shows as reported with no age.

Eight statements on the ledger are now marked reported results and no longer count as promises. Six had been scored as delivered: TeraWulf's preliminary revenue for the first quarter of 2026 and the third quarter of 2025, Big Digital Energy's for the third quarter, the first nine months and the full year of 2025, and IREN's update of Jul 7, 2025 that it had reached 50 EH/s. The other two, Hut 8's list of four new sites and DMG's 75 MW utility approval, were already outside every score. Big Digital Energy and TeraWulf no longer show a score (each had read green, 100 percent). IREN moves from 67 percent, yellow (2 of 3), to 50 percent, orange (1 of 2). Seven of the 18 companies carry a score instead of nine, and the Specific figure reads 53 of 77 statements instead of 60. The delivered chart reads 13 delivered and 10 short across 14 companies instead of 19 delivered and 10 short across 16.

A statement is scored only if its target had not yet been met and its period had not yet ended on the day it was made. Five of the six were preliminary results for a quarter, nine months or year that had already closed: TeraWulf's 8-K exhibits of Apr 14, 2026 (accession 0001104659-26-043279) and Oct 28, 2025 (0001104659-25-102858) and Big Digital Energy's of Oct 17, 2025 (0001213900-25-099694) and Feb 6, 2026 (0001213900-26-012978), each filed 14 to 37 days after its period ended. IREN's 6-K of Jun 30, 2025 (0001878848-25-000047) says it had reached its mid-year target of 50 EH/s that day, a week before the update of Jul 7, 2025 that the ledger holds. A result already in hand is a report, not a promise, so counting it as a promise kept would flatter the score. Each row stays on the ledger with its reason.

The ledger's row for CleanSpark's statement of Nov 25, 2025 called 45.6 EH/s an average for fiscal 2025 in its projection and an average for the September 2025 quarter in its outcome. Both now say it is the September 2025 final-month average.

CleanSpark's 10-K for fiscal 2025 (accession 0001193125-25-297510) says that as of Sep 30, 2025 its operating mining units produced an average of 45.6 EH/s, and the note to its hashrate table says the averages were calculated from operating activity for the final month of the reporting period. It is a one-month average, not a fiscal-year or quarter one. The 42.6 EH/s in the 10-Q for the quarter ended Jun 30, 2026 is the average for June 2026, so the two figures are measured the same way.

The article counted IREN's 50 EH/s update of Jul 7, 2025 and five preliminary revenue figures (TeraWulf's two, Big Digital Energy's three) as promises, printed a 6.6 percent shortfall for CleanSpark from a statement that names no figure and a 35.6 percent shortfall for Greenidge (now Vulcan) from a plan the company superseded, called CleanSpark's 45.6 EH/s a fiscal-year average, and gave three HIVE filings as 8-Ks. Those rows are out of the table and the charts, so the article now names two dossier rows, HIVE and Argo. The article now says, in its section on reported results, that a result already in hand is a report and not a promise; in its opening section, that 45.6 EH/s is the September 2025 final-month average; in its list of sources, that the three HIVE filings are 6-Ks; and in its section on kinds of miss, that the withdrawal example is Bitfarms' 21 EH/s target. The article counts six hashrate promises from five companies instead of eight from six, and its second chart seven promises instead of fifteen.

IREN reached its 50 EH/s target on Jun 30, 2025, a week before the update of Jul 7, 2025 that the ledger holds, and each revenue figure was a preliminary result for a quarter, nine months or year that had already ended when it was stated, so the ledger now marks all six as reported results and does not score them. CleanSpark's statement of Nov 25, 2025 says only that it expects to keep increasing its computing power and names no figure, so the ledger holds its percentage back, and the outcome of Greenidge's statement opens "Superseded by the pivot", so the ledger prints no percentage for it either. CleanSpark's 10-K for fiscal 2025 says its 45.6 EH/s is an average for the final month of the period, September 2025. EDGAR lists HIVE's filings of Jun 27, Oct 28 and Dec 16, 2025 as Form 6-K.

The article's table and both charts scored Bitfarms (now Keel Infrastructure) as short of its 18 EH/s target by 1.7 percent, with 17.7 EH/s at Jun 30, 2025, and counted nine hashrate promises from seven companies, three of them short. Bitfarms is now left out of the table and both charts, a paragraph explains why, and the article counts eight promises from six companies, two of them short.

Bitfarms' management discussion for the second quarter of 2025, filed Aug 12, 2025 (accession 0001213900-25-074626), says the company achieved its initial 2025 targets of 18 EH/s operational and 19 w/TH installed in March 2025; the one for the first quarter, filed May 14, 2025 (0001213900-25-042951), already said it had achieved its first 2025 targets. The results release filed Aug 12, 2025 (0001213900-25-075061) gives 19.5 EH/s operating at Mar 31, 2025 and 17.7 EH/s at Jun 30, 2025, after power to its Argentina site was halted. The 18 EH/s target was met before Jun 30, so 17.7 EH/s was a later reading, not a miss. The same statement of Mar 27, 2025 (0001013762-25-002979) also named a 21 EH/s installed target, which the company says is no longer feasible on its original timeline; 17.7 EH/s does not measure it, so the article gives that target no percentage. The ledger counts the statement as not delivered.

Bitfarms' (now Keel Infrastructure's) statement of Mar 27, 2025 showed an outcome that opened Missed and withdrawn and quoted the Aug 12, 2025 results release, and the file held a variance of minus 1.7 percent against 17.7 EH/s at Jun 30, 2025, which the page did not print. The row now says that two of the three targets it names were reached in March 2025 and that the company calls the third, 21 EH/s installed, no longer feasible on its original timeline. It quotes the discussion that says so, files no percentage, and still counts as not delivered.

The first-quarter management discussion, filed May 14, 2025 (accession 0001213900-25-042951), says the company achieved its first 2025 targets of 18 EH/s operational and 19 w/TH installed and was still working toward 21 EH/s by Jun 30, 2025. The second-quarter discussion, filed Aug 12, 2025 (0001213900-25-074626), says those two were achieved in March 2025 and that the 21 EH/s target is no longer feasible on its originally anticipated timeline; the results release of the same day (0001213900-25-075061) gives 19.5 EH/s operating at Mar 31, 2025 and 17.7 EH/s at Jun 30, 2025. The 17.7 is an operating reading taken after power to the Argentina site was halted, so it does not measure the 21 EH/s installed target. The release the row cited before does not contain the sentence about 21 EH/s. The row still counts as not delivered because one of its three targets was not reached.

The ledger row for Bitfarms' (now Keel Infrastructure's) statement of Mar 27, 2025 covered three targets, 18 EH/s operational, 21 EH/s installed and 19 w/TH installed, and read as one miss. It now covers only the 21 EH/s installed target, the one the company withdrew, and still counts as not delivered with no percentage. The 18 EH/s operational and 19 w/TH installed targets, which the company says it reached in March 2025, are not in this row.

The second-quarter 2025 management discussion, filed Aug 12, 2025 (accession 0001213900-25-074626), says the company achieved its initial 2025 targets of 18 EH/s operational and 19 w/TH installed in March 2025 and that it is no longer feasible to achieve the 21 EH/s target on its originally anticipated timeline. The statement of Mar 27, 2025 (0001013762-25-002979) names all three targets in one sentence, and its own document already reported 18.6 EH/s under Management. Neither the discussion nor the results release of Aug 12, 2025 (0001213900-25-075061) gives an installed hashrate figure for Jun 30, 2025, so the 17.7 EH/s operating hashrate they report does not measure the 21 EH/s target and the row files no percentage.

The article said the ledger keeps only rows whose quote could be found in the cited document as an exact string. It now says that most rows were checked that way, that rows read from company dossiers were not, and that the ledger labels each of those as from a dossier and not yet re-checked. It names the three dossier rows the article draws on, HIVE and Argo in the table and both charts and Greenidge in the second chart, and says that a search of their cited filings on Oct 2, 2026 found the quote and the outcome wording of each, word for word.

The ledger's methodology note says rows from a company dossier carry the same fields but skipped the contiguous-string check, and that each says so. Three rows the article uses are dossier rows: HIVE's statement of Jun 27, 2025, Argo's of May 9, 2025 and Greenidge's of May 15, 2025. A search of their cited filings on Oct 2, 2026 found each quote in its filing, and the outcome wording of each in the filing that answers it, word for word, ignoring only line breaks and spacing. The ledger's labels and counts for those rows did not change.

The note above the selected statements said each of them has a verified quote, a resolved outcome and an outcome sentence. It now says each has a verified quote, gives how many have a settled outcome with its sentence, and says the rest are still open or have no outcome document yet.

The ledger selects up to three statements per company by its selection rule, which ranks a verified quote first and a resolved outcome after it, so a company with fewer than three better rows has an open statement selected. Six of the selected statements have no outcome document yet and one has an outcome document that reports the statement unresolved, so the note was wrong for those seven. The note now takes its count from the ledger's own outcome states.

The quote shown for Bit Digital's statement of Jun 25, 2025 (WhiteFiber targeting 76 MW by the end of 2026) stopped before the end of its sentence, after the words WhiteFiber's 1,300 MW pipeline. It now shows the whole sentence, which ends "of potential new data center projects (as of May 30, 2025)."

Bit Digital's report of Jun 25, 2025 prints that sentence in full. The ledger had copied only the part up to the word pipeline. The projected 76 MW and its date do not change.

The quote shown for HIVE's statement of Jun 27, 2025 (25 EH/s by US Thanksgiving) stopped at a dash, after the words US Thanksgiving. It now shows the whole sentence, which goes on "more than quadrupling our hashrate from March 2025."

HIVE's report of Jun 27, 2025 (exhibit 99.1) runs the sentence on past the dash. The ledger had copied only the part before it. The projected 25 EH/s and its date do not change.

The quote shown for Argo's statement of Mar 26, 2025 (GEM's 2.4 EH/s, expected to more than double Argo's hashrate) ended in the middle of a word, acce. It now shows the whole sentence, which ends "access to the GEM's hosting arrangements."

Argo's report of Mar 26, 2025 on its term sheets with GEM Mining prints the sentence in full. The ledger's copy was cut off after the letters acce. Nothing else in the row changes.

The quote shown for Vulcan's (then Greenidge's) statement of May 15, 2025 (161.5 MW of near-term mining capacity) stopped before the end of its sentence, after the words excluding any future. It now shows the whole sentence, which ends "excluding any future transactions."

The company's first-quarter 2025 results release of May 15, 2025 prints the sentence in full. The ledger's copy was cut off after the word future. The projected 161.5 MW and its date do not change.

The quote shown for CleanSpark's statement of Nov 25, 2025 ended "evaluating strategic acquisitions." It now ends "evaluating strategic acquisition targets."

CleanSpark's annual report for fiscal 2025 reads "evaluating strategic acquisition targets." The ledger's copy had different words at the end, so the quote shown was not in the report word for word. The rest of the sentence was right.

The quote shown for Cipher's statement of Nov 3, 2025 (Colchis, 1 GW energized in 2028) printed American Electric Power (AEP). It now prints American Electric Power ("AEP"), with the quotation marks the company's release uses.

Cipher's earnings release of Nov 3, 2025 puts AEP in quotation marks where it first defines the abbreviation. The ledger's copy had dropped them, so the quote shown was not in the release exactly as printed. No word or figure changes.

The document label on the ledger for HIVE's statement of Nov 18, 2025 (about $140 million of BUZZ HPC annualized run-rate) said 8-K where the company filed a 6-K. It now says 6-K. The link did not change.

EDGAR lists HIVE's filing of Nov 18, 2025 (accession 0001062993-25-016918) as Form 6-K.

The document label on the ledger for HIVE's statement of Oct 28, 2025 (35 EH/s during 2026) said 8-K where the company filed a 6-K. It now says 6-K. The link did not change.

EDGAR lists HIVE's filing of Oct 28, 2025 (accession 0001062993-25-016315) as Form 6-K.

The document label on the ledger for HIVE's statement of Jun 27, 2025 (25 EH/s by US Thanksgiving) said 8-K where the company filed a 6-K. It now says 6-K. The link did not change.

EDGAR lists HIVE's filing of Jun 27, 2025 (accession 0001062993-25-012198) as Form 6-K.

The document label on the ledger for the filing that answers HIVE's statement of Jun 27, 2025, its November 2025 production report of Dec 16, 2025, said 8-K where the company filed a 6-K. It now says 6-K. The link did not change.

EDGAR lists HIVE's filing of Dec 16, 2025 (accession 0001062993-25-017359) as Form 6-K.

The document label on the ledger for Keel's statement of Feb 6, 2026 (redomicile to the United States and rebrand as Keel) said 8-K where the company filed a 6-K. It now says 6-K. The link did not change.

EDGAR lists Keel Infrastructure's filing of Feb 6, 2026 (accession 0001213900-26-013007) as Form 6-K.

The document label on the ledger for Digi Power X's statement of Oct 1, 2025 (delivery and commissioning of the first ARMS 200 pod) said 8-K where the company filed a 6-K. It now says 6-K. The link did not change.

EDGAR lists Digi Power X's filing of Oct 1, 2025 (accession 0001213900-25-094368) as Form 6-K.

The document label on the ledger for Digi Power X's statement of Aug 14, 2025 (first NVIDIA B200 GPU cluster live) said 8-K where the company filed a 6-K. It now says 6-K. The link did not change.

EDGAR lists Digi Power X's filing of Aug 15, 2025 (accession 0001213900-25-077013), which reports the material change of Aug 14, as Form 6-K.

The quote shown for MARA's statement of Jul 1, 2025 (75 EH/s by the end of 2025) printed 1.7 gigawatts (GW). It now prints 1.7 gigawatts ("GW"), with the quotation marks the company's release uses.

MARA's June 2025 production update puts GW in quotation marks where it first defines the abbreviation. The ledger's copy had dropped them, so the quote shown was not in the release exactly as printed. No word or figure changes.

The evidence sentence shown for IREN's statement of Jul 7, 2025 (50 EH/s installed) was a piece of a table from the notes on intangible assets in IREN's annual report for fiscal 2026, which says nothing about hashrate. It now shows the sentence that does: "As of June 30, 2026, our installed Bitcoin mining capacity was approximately 23.2 EH/s, representing approximately 380MW of data center capacity."

IREN's annual report for fiscal 2026 prints that sentence twice, in its business discussion and in its management discussion. The row's outcome line says installed hashrate fell from 50 EH/s in June 2025 to 23.2 EH/s at Jun 30, 2026 as miners were taken out for AI Cloud, and the sentence is the report's own statement of the 23.2. The verdict on the row does not change.

The article's list of sources and the note on its first chart said CleanSpark's results release for its third fiscal quarter of 2025, the one that reports 50 EH/s reached in June 2025, was filed on Aug 7, 2025. Both now say Aug 8, 2025.

EDGAR lists Aug 8, 2025 as the filing date of that 8-K (accession 0000950170-25-105347). The report is dated Aug 7, and the filing was accepted late on the evening of Aug 7, after the cut-off for same-day filing.

The Mechanism and evidence line under How this article reads said the chart of mining revenue at three subsidy levels was drawn from the filed figures the sources list carries. It now says the chart is hypothetical arithmetic that applies the block reward rules the sources list describes.

The sources are Bitcoin Core's code for the block subsidy and its mainnet parameters, two Bitcoin developer guides, the Bitcoin white paper and the Bitcoin.org FAQ. None of them carries a company's filed figures. The article's own note says every worked amount is hypothetical arithmetic: a miner with 1 percent of the network, 144 blocks a day, fees of 0.1 bitcoin a block and bitcoin at 100,000 dollars.

The Mechanism and evidence line under How this article reads said the chart of interest cost before and after a cut was drawn from the filed figures the sources list carries. It now says the chart is hypothetical arithmetic that applies the interest rate mechanics the sources list describes.

The sources are Federal Reserve pages on how policy affects credit, on monetary policy and on why interest rates matter, and Securities and Exchange Commission investor pages on bonds, floating rates and credit risk. None of them carries a company's filed figures. The article's own note says every worked amount is hypothetical arithmetic: 100 million dollars of floating-rate debt at a 5 percent benchmark plus a 4 percent spread, and 25 million dollars of cash earning the benchmark.

The Mechanism and evidence line under How this article reads said the chart "Where the capital gain rates end" was drawn from the filed figures the sources list carries. It now says the chart is drawn from IRS Topic no. 409, which gives these limits for taxable years beginning in 2025.

The chart's figures are the taxable income limits of the 0 percent and 15 percent long-term capital gain rates that IRS Topic no. 409 gives for taxable years beginning in 2025: $48,350 and $533,400 for single, $96,700 and $600,050 for married filing jointly, $64,750 and $566,700 for head of household. They come from that IRS page, not from any company's filing, and the chart cites no other source.

The Mechanism and evidence line under How this article reads said the chart "CleanSpark closes, rebased" was drawn from the filed sample the sources list carries. It now says the chart is drawn from daily closing prices supplied by Polygon.io and Yahoo Finance, which the sources list names.

The chart is computed from daily closing prices, not from company filings. The methodology page says the site's closes come from Polygon.io, with the earlier history filled once from Yahoo Finance's daily bars, and the sources list names both.

The Mechanism and evidence line under How this article reads said the chart "CleanSpark bitcoin leverage" was drawn from the filed sample the sources list carries. It now says the chart is drawn from daily closing prices supplied by Polygon.io and Yahoo Finance, which the sources list names.

The chart is computed from daily closing prices, not from company filings. The methodology page says the site's closes come from Polygon.io, with the earlier history filled once from Yahoo Finance's daily bars, and the sources list names both.

The Mechanism and evidence line under How this article reads said the chart "Return and price-level correlation" was drawn from the filed sample the sources list carries. It now says the chart is drawn from daily closing prices supplied by Polygon.io and Yahoo Finance, which the sources list names.

The chart is computed from daily closing prices, not from company filings. The methodology page says the site's closes come from Polygon.io, with the earlier history filled once from Yahoo Finance's daily bars, and the sources list names both.

The Mechanism and evidence line under How this article reads said the chart "Returns after a low-RSI signal" was drawn from the filed sample the sources list carries. It now says the chart is drawn from daily closing prices supplied by Polygon.io and Yahoo Finance, which the sources list names.

The chart is computed from daily closing prices, not from company filings. The methodology page says the site's closes come from Polygon.io, with the earlier history filled once from Yahoo Finance's daily bars, and the sources list names both.

The Mechanism and evidence line under How this article reads said the chart "Two findings against the p-value ceiling" was drawn from the filed sample the sources list carries. It now says the chart is drawn from two rows of the findings log, one built on filed figures and one on daily closing prices.

One of the two rows is the CleanSpark leverage test, which sets CleanSpark's daily return against bitcoin's daily return, so it is built on daily closing prices, not on a filing. The other row, hashrate against revenue, is built on filed figures. The methodology page says the closes come from Polygon.io, with the earlier history filled once from Yahoo Finance's daily bars.

TeraWulf's Abernathy joint venture with Fluidstack read terminated; it now reads sold. TeraWulf sold its whole 50.1% stake to a group led by Fluidstack, which goes on leading the project, so the agreement no longer counts toward TeraWulf and stays out of the disclosed value as before.

TeraWulf's report of Jul 6, 2026 says it agreed to sell all of its equity interests in the joint venture for about $530 million and would then cease to own any of them; nothing in it ends the lease. The status word said the agreement itself had ended.

The article said TeraWulf's Abernathy joint venture with Fluidstack was terminated in 2026; it now says TeraWulf sold its whole stake to a group led by Fluidstack in July 2026 and that the tracker keeps the row, marked sold.

TeraWulf's report of Jul 6, 2026 records a sale of its equity interests in the joint venture, not an end to the lease, which goes on with Fluidstack leading the project.

Bit Digital's 40 MW agreement with Nscale at WhiteFiber's NC-1 campus was dated October 2024, before the 18-month window, and left out of the disclosed value; it is now dated Dec 18, 2025, inside the window, and its $865 million is counted.

WhiteFiber announced the 10-year, 40 MW colocation agreement with Nscale, about $865 million over the initial term, on 18 December 2025. The row's own source, Bit Digital's investor deck of Jun 30, 2026, dates the Nscale anchor contract December 2025; October 2024 is the date of the line after it in that deck, WhiteFiber's acquisition of Enovum.

TeraWulf's contracted or operating critical IT capacity read 839 MW; it now reads 621 MW, the sum of its contracted and operating rows that state a critical IT figure. The page had explained the gap to its in-window rows as anchor contracts alone.

The total was stored with the deal book and the corrections of Sep 10, 2026 did not reach it: the 210 MW once shown for the CB-3 and CB-4 leases reads not stated, because the release states more than 200 MW and no single figure, and CB-5 reads the 160 MW its release states, not 168. The same rule over the corrected rows gives 621 MW.

Riot's statements corpus listed five of its investor-site releases twice (the conference-call notices of Feb 20, 2026, 04-16, 07-28 and 08-09 and the production update of 04-02) and counted 25 documents, 17 of them filings and press releases; each release is now listed once, and the counts read 20 and 12.

Each pair is one page at one address, once with its closing slash as Riot's feed gives it and once without: the site rebuilt the second from the copy of the page it had saved, which drops the slash. None of the ten rows had been read, so no statement changes.

The chart of the colocation leases as a rent said they sit between 1.4 and 2.8 million dollars per megawatt a year; its reading now says between 1.2 and 2.8.

The chart's own lease of Riot to AMD at Rockdale, 311 million dollars over 10 years for 25 megawatts of critical IT, is 1.24 million dollars per megawatt a year, below the range the sentence gave.

The article said IREN's Microsoft contract, divided the same way as the leases, comes to five times any lease; it now says three and a half times the highest lease.

9.7 billion dollars over five years for 200 megawatts is 9.7 million dollars per megawatt a year. The highest colocation lease on the chart, Digi Power X's to Cerebras, 1.1 billion dollars over 10 years for 40 megawatts, is 2.75 million, so the contract is 3.5 times it.

The table of Riot's datacenter megawatts by status closed on a row adding the four to 1,241 megawatts of critical IT, 25 of them billing; the table now keeps the two measures apart: 241 megawatts of critical IT under the three leases, 25 of them billing, and 1,000 gross megawatts under the letter of intent.

The Corsicana letter of intent names 1,000 megawatts of the site's gross power, not critical IT load, and a total across the two counts one measure as the other; the site adds megawatts only on one denominator.

The headline read $117.5bn of disclosed HPC and AI contract value; it now reads $114.5bn, the sum of the deal rows the page lists. The difference is the $3.00bn once shown for TeraWulf's CB-5 lease, which the row itself has read as not stated since Sep 10, 2026.

The headline was a total stored in the deal book, and the correction of Sep 10, 2026 to the CB-5 row did not reach it: the release states no value for CB-5. The page, its share card and the home page now add up the rows themselves, so a correction to a row moves the total with it.

The globe drew 135 of the 147 active facility rows and its legend read "Operating 95 (83 mapped)": twelve operating sites, one building and two planned were on file, counted in every total, and on no map. Fourteen of the fifteen are now drawn, at the precision their filing states, and the legend counts what the globe drew with a note under it for the one that is still not drawn.

Every one of those rows but one names a place its filing states: a town (Molde), a region (Oromia, British Columbia, South Australia, Cordillera, Ontario, Manitoba) or a country (Norway, Bhutan, Canada, the United States, Iceland). None had been geocoded, so each was counted and never shown. Each is now placed from the same gazetteers as every other facility, never finer than the filing states, and marked with its precision and the cities its centroid is the mean of. MARA's row for five international sites in three countries, one of them not named, states no single place and is still not drawn.

The share-count headline read "The share count went from 325.7M at Mar 31, 2019 to 274.3M at Aug 14, 2026", a fall. The panel now compares counts only inside one capital structure: the headline reads from 165.6M at Mar 31, 2025 to 274.3M at Aug 14, 2026, the three earlier counts are not drawn or compared, and the panel says why.

The 2019 to 2021 counts are IFRS facts and the later ones are cover-page counts under a different concept, with a four-year gap and a fall to 45 percent between them: the fall was a consolidation, not a repurchase. The cited 40-F/A does not print the 2019 figure in its text at all; it exists only as a tagged fact, which the panel now says.

The share-count headline started from 1 share at Dec 31, 2022; it now starts from 112.0M at Dec 31, 2023 and says that the earlier count is neither drawn nor compared.

The 20-F for 2022 states one outstanding ordinary share as of Dec 31, 2022: Bitdeer was a pre-merger holding company, and a dilution series that starts before a reverse merger measures the merger, not issuance.

The share-count headline started from 36.9K shares at Dec 31, 2020; it now starts from 25.0M at Dec 31, 2021 and says that the earlier count is neither drawn nor compared.

The 2020 count is a pre-reorganisation IFRS fact and the 2021 count a cover-page count of a different concept, 677 times larger; a share count is comparable only inside one capital structure.

The projection printed beside MARA's statement of Aug 6, 2026 read "exit of all hosted mining (537 MW, 29.2 EH/s at Dec 31, 2025) by Q1 2028"; it now reads "exit of all third-party hosting arrangements by Q1 2028".

The quoted statement says only that third-party hosting arrangements expire beginning Q3 2027 and conclude by Q1 2028. Neither 537 MW nor 29.2 EH/s is in it, and 29.2 appears nowhere in the cited letter.

IREN's revenue statement of Nov 6, 2025 printed "not computed" with a generic reason; the reason now names the contradiction the row records: the target was more than $500m of ARR by Mar 31, 2026, the document nearest that date reports about $0.4bn under contract five weeks earlier, and the $0.5bn figure is from a release 37 days after the target date, so a single percentage would compare a projection with an outcome from a later period.

An earlier build of the row filed 0.0 percent, which its own figures do not give back; the row was repaired to file none, and the audit asked that the page keep naming why rather than fall back to a generic clause.

TeraWulf's CB-3 and CB-4 leases showed 210 MW of critical IT load and a revenue per MW derived from it; both now read not stated, and the release's own words, more than 200 MW, are quoted in the row.

The release of Aug 14, 2025 states "more than 200 MW of critical IT load (representing ~250 MW of gross capacity)" and no single figure; 210 was 42 plus 168, and the 10-Q for the quarter ended Jun 30, 2026 says the July 2026 amendment raised "the aggregate contracted critical IT load under two leases from 162 MW to 168 MW", which states no figure for any one lease.

TeraWulf's CB-5 lease showed 168 MW of critical IT load and $3.00bn of contract value; it now shows the 160 MW the release of Aug 18, 2025 states, and the contract value reads not stated.

The release states "an incremental 160 MW of critical IT load" and no value for CB-5; 168 came from the July 2026 amendment, but the 10-Q for the quarter ended Jun 30, 2026 says the July 2026 amendment raised "the aggregate contracted critical IT load under two leases from 162 MW to 168 MW", which states no figure for any one lease, and $3.0bn was $6.7bn less $3.7bn where the release states no $6.7bn.

Barber Lake (CIFR) showed 0 MW energized on the home facilities list and the profile; the energized figure now reads not disclosed.

The cited 10-K for 2025 states no energized capacity for Barber Lake; the 0 was inferred from the site's stage, and the row's own note records the tenant in beneficial use by August 2026.

Farnham (KEEL) showed 10 MW nameplate beside 10 MW energized; the nameplate figure now reads not disclosed.

The 10-K properties table gives Farnham 10 MW under the column energized capacity and states no nameplate; the row had copied the one figure into both columns.

The EV per EH/s tile linked its share-price input to a Yahoo Finance chart endpoint as if it were a document; the tile now names the provider and the as-of date, marks the price input as unverifiable by a reader, and does not link the endpoint.

The endpoint returns JSON rather than a page and its robots.txt disallows automated fetching, so neither a reader nor an auditor can check the price; the other four inputs trace to filings and the arithmetic reproduces exactly.

The EV per EH/s tile linked its share-price input to a Yahoo Finance chart endpoint as if it were a document; the tile now names the provider and the as-of date, marks the price input as unverifiable by a reader, and does not link the endpoint. The same change applies to every computed valuation figure on the site.

The endpoint returns JSON rather than a page and its robots.txt disallows automated fetching, so neither a reader nor an auditor can check the price; the other inputs trace to filings and the arithmetic reproduces exactly.

The purchased energy cost per bitcoin, $38,690 for the quarter ended Jun 30, 2026, was on file with its sentence and never drawn, because the line was conditional on the all-in tile above it, which is not disclosed; the line is now drawn whenever the figure exists.

The Q2 2026 shareholder letter states the figure outright, and the page had offered the reader the arithmetic while withholding it.

The direct cost per bitcoin, $26,259 for June 2025, was on file with its sentence and never drawn for the same reason as MARA's; the line is now drawn whenever the figure exists.

The June 2025 monthly update states the figure outright; the all-in tile above it is not disclosed, which was the condition that hid it.

The direct cost per bitcoin, about $36,500 for the quarter ended Jun 30, 2026, was on file with its sentence and never drawn for the same reason as MARA's; the line is now drawn whenever the figure exists.

American Bitcoin's Q2 2026 results release states the figure outright; the all-in tile above it is not disclosed, which was the condition that hid it.