Methodology. How every number on this site is found, checked and dated: the filing fact behind each financial figure, the report behind each operating figure, and the rules for both.
Revised Corrections 339 under the policy of Filings read
Metric definitions
Every number the site shows is one of these metrics. Concepts are tried in order; the latest period end wins, ties go to the first concept. Values come from the SEC companyfacts API (XBRL as filed) and are never derived unless kind is derived. The seven filing metrics come first: each is computed the same way for every company from XBRL alone, so they compare. They are not the seven operational metrics a profile opens with (realized hashrate, fleet efficiency, cost per bitcoin with depreciation, bitcoin held, EV per EH/s, revenue per MW, months of liquidity). Those 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. The Compare tool opens with them, with the direct cost beside the cost with depreciation, and says in a company's column when the pass did not cover it.
How a filed duration becomes a period, by its length alone and never by a filer's own label, and when a value goes stale.
How a filed duration becomes a period, by its length alone and never by a filer's own label, and when a value goes stale.
Rule
Quarter
duration of 80 to 100 days
Half-year
duration of 170 to 190 days
Fiscal year
duration of 350 to 380 days
Year to date
any other duration; excluded from period metrics
Stale after
400 days
Revenuerevenuehero
Technical detail
Total revenues as reported for the period.
10-Q and 10-K income statement facts for domestic filers; 20-F, 40-F and 6-K facts for foreign private issuers. Latest period end across the concepts; a total tagged as Revenues outranks a contract-revenue component.
CleanSpark tags revenue under a company-specific element in its 10-Qs since fiscal 2025, which the companyfacts API omits; its latest tagged total is older and marked stale.
Reporting schedule: quarterly for 10-Q filers; semiannual or annual for 20-F and 6-K filers.
From the income statement: total revenue.
Unit
Dollars
Reports
Quarterly, semiannual, or annual
Net income (loss)net_incomehero
Technical detail
Net income or loss attributable to the company for the period, as reported.
Income statement fact for the latest period end. Net income attributable to the company outranks the figure available to common holders, which outranks the total including noncontrolling interests; the concept used is named on hover.
MARA tags the quarter's figure as available to common stockholders rather than NetIncomeLoss, so its quarterly value comes from that concept.
Reporting schedule: quarterly; semiannual or annual for IFRS filers.
From the income statement: profit or loss.
Unit
Dollars
Reports
Quarterly, semiannual, or annual
Cash and cash equivalentscashhero
Technical detail
Cash and cash equivalents at the balance sheet date.
Balance sheet fact at the latest instant. The restricted-cash total is used only when no unrestricted cash concept is tagged, and is labeled by its concept.
From the balance sheet: cash and cash equivalents.
Unit
Dollars
Reports
Quarterly
Digital assets at fair valuedigital_assets_fair_valuehero
Technical detail
Crypto assets held, measured at fair value under ASU 2023-08, at the balance sheet date.
Balance sheet or crypto-asset note fact at the latest instant. Pre-2024 cost-less-impairment carrying values (intangible assets) are a different measurement and are not shown under this metric.
Includes every crypto asset the company holds, not only bitcoin, as tagged.
Concepts are checked in this order:
us-gaap:CryptoAssetFairValue
us-gaap:CryptoAssetFairValueCurrent
Reporting schedule: quarterly.
From the balance sheet: the fair value of crypto assets.
Unit
Dollars
Reports
Quarterly
Digital asset units helddigital_asset_unitshero
Technical detail
Number of crypto asset units held at the balance sheet date, in the unit the filer tagged (bitcoin for miners; ether for Bit Digital).
Crypto-asset note fact at the latest instant; the unit string is shown as tagged.
Nine of eighteen companies tag this concept; the rest disclose holdings in prose, which the site does not extract. The companyfacts API drops dimensional context, so a filer's non-dimensional units fact may cover a subset of its holdings; the concept and document are named on hover. A value that fails the scale check against the prior period is held for review (methodology, validation).
Concepts are checked in this order:
us-gaap:CryptoAssetNumberOfUnits
Reporting schedule: quarterly.
From the crypto asset note: units held.
Unit
As reported
Reports
Quarterly
Shares outstandingshares_outstandinghero
Technical detail
Common shares outstanding at the most recent date reported.
Cover-page count (dei) first because it is dated closest to the filing date; balance sheet count otherwise. The dilution series uses every dated value across these concepts.
Strategy (MSTR) tags its share classes dimensionally, which the companyfacts API omits, so its count is missing here.
Concepts are checked in this order:
dei:EntityCommonStockSharesOutstanding
us-gaap:CommonStockSharesOutstanding
ifrs-full:NumberOfSharesOutstanding
Reporting schedule: on every periodic filing.
From the filing cover or balance sheet: common shares outstanding.
Unit
Shares
Reports
Each filing
Total assetstotal_assetshero
Technical detail
Total assets at the balance sheet date.
Balance sheet fact at the latest instant.
Concepts are checked in this order:
us-gaap:Assets
ifrs-full:Assets
Reporting schedule: quarterly.
From the balance sheet: total assets.
Unit
Dollars
Reports
Quarterly
Cash flow after property and equipment purchaseshistorical_free_cash_flowderived
Technical detail
Reported operating cash flow minus payments to acquire property, plant and equipment for the same filing period.
Both inputs must have the same exact start and end dates, accession, filing date, USD currency and accounting taxonomy. Select the latest eligible anchor period; never substitute an older matched pair. Conflicting values and missing document links refuse the calculation. Not trailing twelve months or a forecast.
Capital expenditure must be a nonnegative payment amount. This bounded cash flow measure excludes acquisitions, intangible purchases and other investment spending; it is not a company-adjusted free cash flow figure.
Concepts are checked in this order:
Reporting schedule: on each eligible filing period.
Calculated from operating cash flow less spending on property and equipment.
Unit
Dollars
Reports
Each filing
Cash from financing activitiesfinancing_cash_flow
Technical detail
Net cash provided by or used in financing activities, as tagged for the stated period.
Latest eligible reported quarter, half-year or fiscal year; other year-to-date durations are excluded. USD only. The exact period and source filing are shown.
Only captured US GAAP tags are covered. Missing or unsupported tags remain unavailable; no estimate or foreign-currency conversion is made.
Reporting schedule: quarterly; cash flow may be cumulative.
From the cash flow statement: cash from financing.
Unit
Dollars
Reports
Quarterly
Cash from investing activitiesinvesting_cash_flow
Technical detail
Net cash provided by or used in investing activities, as tagged for the stated period.
Latest eligible reported quarter, half-year or fiscal year; other year-to-date durations are excluded. USD only. The exact period and source filing are shown.
Only captured US GAAP tags are covered. Missing or unsupported tags remain unavailable; no estimate or foreign-currency conversion is made.
Reporting schedule: quarterly; cash flow may be cumulative.
From the cash flow statement: cash from investing.
Unit
Dollars
Reports
Quarterly
Cash from operationsoperating_cash_flow
Technical detail
Net cash provided by or used in operating activities for the period.
Cash flow statement fact; 10-Q cash flow facts are year-to-date, so the site shows the latest fiscal-year or first-quarter value and labels the period.
From the cash flow statement: cash from operations.
Unit
Dollars
Reports
Quarterly and annual
Cost of revenuecost_of_revenue
Technical detail
Cost of revenue for the period, as reported. Company definitions differ (power only versus power plus depreciation); the concept is named on hover.
Income statement fact for the latest period end.
Concepts are checked in this order:
us-gaap:CostOfRevenue
us-gaap:CostOfGoodsAndServicesSold
ifrs-full:CostOfSales
Reporting schedule: quarterly.
From the income statement: the reported cost of revenue.
Unit
Dollars
Reports
Quarterly
Depreciation and amortizationdepreciation
Technical detail
Depreciation and amortization for the period, as reported.
Cash flow statement or income statement fact for the latest period end; cash flow facts in 10-Qs are year-to-date and are excluded unless they cover one quarter.
Concepts are checked in this order:
us-gaap:DepreciationDepletionAndAmortization
us-gaap:DepreciationAndAmortization
us-gaap:Depreciation
ifrs-full:DepreciationAndAmortisationExpense
Reporting schedule: quarterly.
From the cash flow or income statement: depreciation and amortization.
Unit
Dollars
Reports
Quarterly
Diluted earnings per shareeps_diluted
Technical detail
Diluted EPS for the period, as reported.
Income statement fact for the latest period end.
Concepts are checked in this order:
us-gaap:EarningsPerShareDiluted
ifrs-full:DilutedEarningsLossPerShare
Reporting schedule: quarterly.
From the income statement: earnings per diluted share.
Unit
Dollars per share
Reports
Quarterly
Dilution, one year (share count change)dilution_1yderived
Technical detail
How much the number of shares grew or shrank in a year. Growth spreads the company over more shares, so each share owns less: dilution.
(latest shares outstanding / shares outstanding at the dated value closest to 365 days earlier) - 1, shown with the absolute share change. Both facts are cited.
Computed from the dilution series (every dated share count across the listed concepts, one value per date, latest filing wins). Missing when the series has no value within 300 to 430 days of the latest point.
Concepts are checked in this order:
dei:EntityCommonStockSharesOutstanding
us-gaap:CommonStockSharesOutstanding
ifrs-full:NumberOfSharesOutstanding
Reporting schedule: on every periodic filing.
Calculated from two filed share counts about a year apart.
Unit
Percent
Reports
Each filing
General and administrative expensesga
Technical detail
Selling, general and administrative expense for the period, or general and administrative where that is the tagged line.
Income statement fact for the latest period end.
Concepts are checked in this order:
us-gaap:SellingGeneralAndAdministrativeExpense
us-gaap:GeneralAndAdministrativeExpense
Reporting schedule: quarterly.
From the income statement: general and administrative costs.
Unit
Dollars
Reports
Quarterly
Income tax expense (benefit)income_tax
Technical detail
Income tax expense or benefit as tagged for the period, not cash taxes paid.
Latest eligible reported quarter, half-year or fiscal year; other year-to-date durations are excluded. USD only. The exact period and source filing are shown.
Only captured US GAAP tags are covered. Missing or unsupported tags remain unavailable; no estimate or foreign-currency conversion is made.
Concepts are checked in this order:
us-gaap:IncomeTaxExpenseBenefit
Reporting schedule: quarterly; cash flow may be cumulative.
From the income statement: income tax expense or benefit.
Unit
Dollars
Reports
Quarterly
Long-term debtlong_term_debt
Technical detail
Long-term debt as tagged at the balance sheet date; noncurrent portion where the filer tags it separately.
Balance sheet or debt note fact at the latest instant. Convertible notes tagged under their own concept are not added in.
Concepts are checked in this order:
us-gaap:LongTermDebtNoncurrent
us-gaap:LongTermDebt
ifrs-full:LongtermBorrowings
Reporting schedule: quarterly.
From the balance sheet or debt note: long-term debt.
Unit
Dollars
Reports
Quarterly
Net marginnet_marginderived
Technical detail
100 x net income or loss / revenue, from the same filing and period.
Both inputs must have the same exact start and end dates, accession, filing date, USD currency and accounting taxonomy. Select the latest eligible anchor period; never substitute an older matched pair. Conflicting values and missing document links refuse the calculation. Not trailing twelve months or a forecast.
Revenue must be positive. Net income follows the registered concept priority, which can include income available to common holders; the selected concept is shown with the inputs.
Concepts are checked in this order:
Reporting schedule: on each eligible filing period.
Calculated from net income divided by revenue.
Unit
Percent
Reports
Each filing
Operating income (loss)operating_income
Technical detail
Operating income or loss for the period, as reported.
Income statement fact for the latest period end.
Concepts are checked in this order:
us-gaap:OperatingIncomeLoss
ifrs-full:ProfitLossFromOperatingActivities
Reporting schedule: quarterly.
From the income statement: operating profit or loss.
Unit
Dollars
Reports
Quarterly
Operating marginoperating_marginderived
Technical detail
100 x operating income or loss / revenue, from the same filing and period.
Both inputs must have the same exact start and end dates, accession, filing date, USD currency and accounting taxonomy. Select the latest eligible anchor period; never substitute an older matched pair. Conflicting values and missing document links refuse the calculation. Not trailing twelve months or a forecast.
Revenue must be positive. The operating income and revenue concepts are shown with their source inputs.
Concepts are checked in this order:
Reporting schedule: on each eligible filing period.
Calculated from operating income divided by revenue.
Unit
Percent
Reports
Each filing
Property, plant and equipment, netppe_net
Technical detail
Net property, plant and equipment at the balance sheet date.
Balance sheet fact at the latest instant.
Concepts are checked in this order:
us-gaap:PropertyPlantAndEquipmentNet
ifrs-full:PropertyPlantAndEquipment
Reporting schedule: quarterly.
From the balance sheet: property and equipment after depreciation.
Unit
Dollars
Reports
Quarterly
Public floatpublic_float
Technical detail
Aggregate market value of voting and non-voting common equity held by non-affiliates, as stated on the 10-K or 20-F cover.
Cover page fact, dated at the end of the second fiscal quarter.
Concepts are checked in this order:
dei:EntityPublicFloat
Reporting schedule: annual.
From the filing cover: the market value of shares held by the public.
Unit
Dollars
Reports
Annual
Purchases of property and equipmentcapex
Technical detail
Payments to acquire property, plant and equipment for the period.
Cash flow statement fact; year-to-date in 10-Qs, so the latest fiscal-year or first-quarter value is shown with its period.
The spec's seven hero metrics live in production reports, earnings exhibits and prose, not XBRL. Two dated research passes compile them with a source per figure: Tier 1 covers 9 companies (Sep 6, 2026) and Tier 2 6 more (Sep 6, 2026), 15 in all. Status words: reported (the company's own figure and document), computed (a formula with every input sourced, drawn hatched), computed proxy (a stand-in where inputs are not disclosed), null (not stated in what we read, shown as words). "Not comparable" means the company defines the figure differently; the note says how. Computed figures that need a share price use the as-of close from the feed named in the source list.
Hashrate, realizedhashrate_eh_realizedhero
How much computing power the company actually ran in the period, in exahash per second; more hashrate earns a larger share of the bitcoin the network pays out.
Self-mining hashrate from the latest production or results report, in exahash per second.
Unit
EH/s
Fleet efficiencyfleet_efficiency_j_thhero
How much electricity the machines use for each unit of computing, in joules per terahash: fewer joules per terahash means less electricity for the same computing.
Fleet-weighted joules per terahash as the company reports it; fewer joules per terahash means less electricity for the same computing.
Unit
J/TH
Cost per bitcoin with depreciationcost_to_mine_all_in_usdhero
What the company says it costs to mine one bitcoin once the wear on its machines (depreciation) is counted, as the company states it: the figure's card says what else it counts and leaves out. Set against the company's revenue per bitcoin mined in the same period, it is that cost as a share of what each bitcoin mined brought in; set against the spot price, which is a later day, it compares that period's cost with that price.
The company's own figure, for the period it states; the figure's card says what it counts and the period. It is not the site's all-in method (Method B: mining costs plus payroll, general and administrative expense and depreciation, divided by bitcoin mined), which the workbench works out from the filings; Method A (direct power and hosting only) is the second line.
Unit
USD per BTC
Direct cost per bitcoincost_to_mine_direct_usd
What the company says it costs to mine one bitcoin before machine depreciation: the power bill and the other direct mining costs it counts, as the figure's card gives them; a figure that counts the power bill alone says so.
The company's own figure, on its own definition and for the period it states; the figure's card gives both. The site's Method A (power and hosting costs divided by bitcoin mined) is worked out in the workbench.
Unit
USD per BTC
Bitcoin heldbtc_heldhero
How many bitcoin the company held at the date, as it reported them.
Treasury balance from the latest production report or filing.
Unit
BTC
EV per EH/sev_per_eh_usdhero
What the market pays for each exahash of running capacity: enterprise value (market value plus debt minus cash) divided by realized hashrate. It compares price tags across miners of different sizes.
Enterprise value (market capitalization plus debt less cash) divided by realized hashrate; every input is named in the method.
Unit
USD per EH/s
Market capitalizationmarket_cap_usd
What the market values the company's shares at on the as-of date.
Closing price on the as-of date times the latest filed share count.
Unit
USD
Revenue per MW, annualizedrevenue_per_mw_usd_annualizedhero
Revenue for each megawatt of energized capacity, at the latest quarter's pace times four. Higher means more earned from each unit of power the company controls.
Latest quarterly revenue times four, divided by energized megawatts.
Unit
USD per MW per year
Months of liquiditymonths_of_liquidityhero
How many months the company could cover fixed costs and debt payments from cash and bitcoin at the as-of price if revenue stopped. A proxy where the company does not disclose the inputs.
(Cash plus bitcoin at the as-of price) divided by monthly fixed costs and debt service, as a proxy where the company does not disclose the inputs.
One row per announced HPC or AI colocation, lease, hosting, cloud or GPU agreement, with the status the document supports and nothing stronger. The six status words: operating, signed and serving; contracted, a definitive agreement signed with delivery possibly ahead; letter of intent, non-binding; announced intention, interest stated and no counterparty signed; sold, the company sold its stake and the agreement goes on without it; terminated, an agreement that ended. A figure on a row that is not an agreement carries its status word, because a megawatt read alone would look contracted.
Megawatts come on two denominators and are never added across them: critical IT load is what a tenant can draw; gross is the site before conversion losses and overheads. A contract value is what the document discloses. The deal page's totals count only rows that disclose one, so they understate the book, and exclude agreements no longer held (sold or terminated). Both headline counts are read from the research file, not summed from the table.
A company with no agreement is a verified negative, not a blank: the documents read are listed and the sentence saying so is quoted. The deal book.
A guidance row is a sentence a company published about its own plans or results, kept verbatim, with what it projected and the document it came from, and, once a later document reports the result, what happened and that document. A statement enters only when its quote was found in the cited primary document; statements whose quotes could not be found were dropped, not paraphrased. A row is marked verified by the check described next, and a row that did not pass says so.
A row is marked verified when its quote is found in the document it cites as one contiguous string. A quote that joins passages with ellipses is checked passage by passage, and each passage must be found as a contiguous string at a word boundary. A quote cut inside a word is not found, and neither is one cut inside a number: 2027 shortened to 202, 1,210 MW shortened to 210 MW, and 3.4bn shortened to 4bn are all refused. A passage that stops at punctuation, such as a full stop or a closing bracket, stands. The search ignores differences in spacing, capital letters, curly or straight quotation marks and the kind of dash; the words and numbers must match exactly. A verified quote can be part of a longer sentence, because the check asks whether the quoted words are in the document, not whether they are a whole sentence. The search was run on every row, the latest on Oct 3, 2026. Every row, whichever research pass it came from, is marked verified only if, in addition, every figure its projection names is printed in its quote, with the same unit where the quote gives one. A figure is a number with a unit (MW, EH/s, a percent sign, a dollar sign, million, billion) or a number of more than one digit. A year on its own and a date are not figures, but a year-sized quantity such as 2000 MW is. A figure printed with another unit does not count: 50% is not $50. No row was held back. A row that stores the sentence that reports its outcome has that sentence searched the same way in its outcome document. 42 rows store the sentence that reports their outcome, and the check found each word for word in its outcome document. 15 link an outcome document without a sentence, so their outcomes were not string-checked, and their cards say so. The check does not read whether a sentence supports the row's Actual line: the Actual line and the delivered verdict are the research pass's reading of the outcome document. Nor does it compare target dates with the quote. A row's target date is the research pass's reading of the statement, and the quote may state no year or a different one: IREN's statement of Nov 6, 2025 says only "by end of Q1", and its target date of Mar 31, 2026 is the research pass's reading of that phrase. The search of each filing runs on a separate machine that reads the document. The check that runs on every change to the site does not read the filing again: it compares each row with the stored result of that search and recomputes the figure test from the row's own quote and projection.
Specificity decides what scores. A specific statement (a number and a date) scores; a directional or vague one is shown but never scored. A reported result is not scored either: it is a statement whose result was already in hand on the day it was made, such as the preliminary figures of a quarter or year that had already ended, or a target the company had already reached. It reports what happened and promises nothing, so the ledger shows it, marks it a reported result and never scores it. A variance shows only where projection and outcome are both comparable numbers; where the outcome is in words the column says "not computed" and why, and where the target date has not passed it says "not resolved". Neither is a zero. The ledger.
Every row carries the channel its statement came through, read from the source document's words and URL host. Formal register (the company on the record): filings and press releases (annual and quarterly reports, 8-K and 6-K exhibits, shareholder letters, investor-site releases), investor presentations (decks furnished as exhibits) and earnings calls (furnished transcripts). Informal register (people in their own voice): X posts and interviews, which hold nothing yet; X waits on a paid API purchase still to be decided, interviews on a transcription step not yet built. Behind the ledger sits the statements corpus: every document held for a company in the 12-month exhibit window, sorted onto a channel, with each sentence that pairs a forward-looking word with a quantity or dated target kept verbatim. The rule is mechanical and unscored; a candidate becomes a ledger row only through the research pass above. What the corpus holds.
The honesty score on the ledger is guidance accuracy alone, one of the five components the full score is to weigh, computed once per register: the share of a company's settled, specific statements it delivered. The formal score covers filings and press releases, investor presentations and earnings calls; the informal score covers X posts and interviews, empty until captured. Every settled row carries a verdict (yes, no, late or unclear) read off words the research pass had already written (the recorded outcome, a slippage in days, or the word the outcome sentence opens with), never off the numbers; the sentence stays on the row so the verdict can be checked. Late counts as not delivered, because a statement made with a date is a statement about the date. Where the words settle neither way, the verdict is unclear and the older rule applies: the reported figure against the projection, where both are comparable numbers; delivered means it reached the projection or, for a cost-type metric, stayed at or under it. Bands: green 80% and up, yellow 60 to 79, orange 40 to 59, red under 40; a register with fewer than two scorable statements is not banded and the bar says why. Directional, vague and open statements, and reported results, are counted and shown but not scored. The other four components (revision behavior, facility delivery, insider alignment, statement consistency) are not yet computed.
Coverage matrix
Which company tags which metric. A row shows, for each hero metric, the period end of the latest tagged value or why there is none, then how many of the other 19 metrics it tags and which it does not. Each company's page shows every metric.
Each company's latest tagged period end for every hero metric, and how many other metrics it tags.
the company's XBRL filings do not tag it, so it cannot be read here.
In review:
a filed value fails a check against the previous period and is held until checked (Validation and review).
No SEC data:
the company files on SEDAR+ only.
Days old:
printed on a company's row when its oldest tagged value ended more than 400 days before the data date; every cell keeps its own period in its title.
The home map plots every active facility with coordinates, sized by energized megawatts and solid, hollow or faint by stage (operating, building, planned). Coordinates are gazetteer centroids at the precision each row names (site, city, county, state or country), never survey points; coarser precision draws a wider, softer mark. A few rows are placed more exactly from a primary source: where a filing names the power plant a site is built beside, the point is that plant's published location (not the mining hall), and where a company page gives a street address, the point is the national address register's point for it. Where a state or county keeps its own address points (Texas, Tennessee, Kentucky, Walker County in Georgia and New York do), the point is the address as that office records it for the filed address: not the building and not the entrance, and for Kentucky the layer's own placement field says Unknown. New York's point for Lake Mariner sits on the main structure at its address and was last edited in 2018, before the data campus was built. Where a company document places a site on a county land parcel, the point is the centre of that parcel, not the building. A parcel can be much larger than the site and belong to its landlord, so the row names the owner of record and the size of the parcel. Where the county draws one parcel as several separate pieces, the point is the centre of the one piece the row names, never the centre of all the pieces together. Where a second map source agrees on a street address within 300 metres, that confirms the address, not where the buildings stand. Each of those rows says so in its panel and links the document that names the place. A row whose filing names no place is not drawn, and the legend counts them. Sold and terminated sites are off the map and the energized total; the explorer lists them under Exited.
Dates on file for 131 of 151 sites: 130 announced, 45 energized, 27 expected online. Announced is the first public filing or release found that names the site; energized and expected are as a document states them, as precisely as it does. Every date keeps its source sentence; a site with none has no date, and an energized site keeps no expected date.
Validation and review
Values are XBRL facts as filed. Nothing is estimated, interpolated or hand-corrected; a fact the site cannot trust is held for review and shown as words, never a number.
Scale check: an instant value that moves more than a hundredfold from the previous dated value of the same concept and unit is held for review with its citation.
Plausibility check, three pairs. A scale error that persists never moves a hundredfold from the period before, so only another figure in the same filings exposes it.
Public float against total assets, held above 100 times. CleanSpark's cover-page float has read in the trillions since fiscal 2023 against total assets of $2.7B; the largest honest ratio here is MicroStrategy's, at two times.
Digital assets at fair value against total assets, also 100 times. The honest maximum is one, since the first is a line item in the second; the largest here is MicroStrategy at 0.95. The ceiling stays far above that on purpose: it targets filer scale errors, and one near the maximum would fire on a restated total.
Bitcoin units against digital assets at fair value on the same balance-sheet date, which implies a price per unit. Five companies state both, implying $58.5K to $66.5K; a count outside 20 times that range is held. No price series is used; the band is what filers state. It catches a units error the others miss: a count off by a thousand is not a large number but a bitcoin worth a few hundred dollars.
A fourth pair, cost per bitcoin against revenue per bitcoin mined, was measured and left out. Its halves are a quarter's revenue as tagged in XBRL over the bitcoin mined in that quarter's three monthly releases, and a cost per bitcoin for the same quarter. For the four quarters ending Sep 30, 2025 to Jun 30, 2026 no company had all three on file (probe run Sep 10, 2026, and again Oct 9, 2026), and a band needs several filers. RIOT, CLSK, WULF and VIP print their own revenue per bitcoin mined beside their cost to mine, for the quarter ended Jun 30, 2026: the Companies page shows them. Those are the companies' own ratios, one filer each, and not this pair. Revenue over bitcoin mined is also not a price per bitcoin for a filer with hosting or HPC revenue. It becomes buildable when filers state mining revenue alone, or bitcoin mined for the quarter.
Unit check: dollar metrics accept USD only, share counts shares only.
Period check: quarter, half-year and fiscal-year durations only; year-to-date facts are excluded from period metrics.
Restatements: each distinct value ever filed for a period is counted and shown on hover as versions; the latest filing wins the display.
Facts in review (6)
CLSK, Digital asset units held:The 10-Q filed Aug 6, 2026 tags 1,719,000 Bitcoin at Jun 30, 2026, against 1,641 at Mar 31, 2026: the scale check fails, so the value is held for review instead of shown.document
DGXX, Digital asset units held:The 10-K filed Mar 31, 2026 tags 9 Bitcoin at Dec 31, 2025, but 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 the 10-K's text says it held approximately 133 bitcoins at that date. The newest count it states is about 217 bitcoin held, quarter ended Jun 30, 2026. The tagged count is held for review instead of shown.document
CLSK, Public float:The 10-K filed Nov 25, 2025 states public float of $1.97T at Mar 31, 2025, against total assets of $2.7B at Jun 30, 2026: 728 times, where the cover-page float is the market value of the same company whose balance sheet is on this page. The plausibility check fails, so the value is held for review instead of shown.document
HIVE, Public float:The 10-K filed Jun 2, 2026 tags 950,740,231,000 USD at Mar 31, 2026, against 950,740,231 at Sep 30, 2025: the scale check fails, so the value is held for review instead of shown.document
ARBK, Stockholders' equity:The 20-F filed May 15, 2025 tags -29,485,000 USD at Dec 31, 2024, against 158,000 at Dec 31, 2023: the scale check fails, so the value is held for review instead of shown.document
MARA, Total liabilities:The 10-K filed Mar 10, 2022 tags 765,221,042 USD at Dec 31, 2021, against 3,924,170 at Sep 30, 2021: the scale check fails, so the value is held for review instead of shown.document
Number audit of Sep 6, 2026
A reader-side audit re-read published values against their cited documents.
6.7% (10 of 150), 95% interval 3.7 to 11.8. The draw over-samples small datasets, so this is the rate over the sample, not the site.
Rate weighted to the whole site
Withdrawn. The estimate is 0.41%, but one stratum holds 34,273 of 35,856 values and got 9 draws; at the top of its own interval the weighted rate is 29%, so 0.41% measures nothing.
Values from tagged XBRL facts (quarterly facts, share counts)
1 of 60 not verified, 1.7%, 95% interval 0.3 to 8.9
Values from the research pass (hero, tier 2, monthly, ATM, facilities, deals, guidance)
9 of 90 not verified, 10.0%, 95% interval 5.4 to 17.9
Errors found and fixed
All 10 (MARA's direct cost per bitcoin among them), plus 2 more direct costs per bitcoin a census found missing from their page the same way and 2 share-count series that began before a merger. Fixed Sep 10, 2026 and recorded one entry each among the 14 corrections log entries of that date below.
What changed so it does not recur
A share-count series is compared only inside one capital structure. Every deal and facility figure called quoted was found word for word in its cited document (checked Sep 10, 2026); one not found there prints as not quoted.
As-of and staleness
Every figure carries its period, source document and that document's filing date. The "filings read" date (Oct 14, 2026, 16:10 UTC) is the day the SEC APIs were last read, and the as-of date the rules measure from.
A value whose period end is older than 400 days on the as-of date carries a stale badge with its age in days; it stays shown because it is still the latest tagged value.
Timestamp alignment
Cross-asset computations key on the NYSE session date. Bitcoin, a 24/7 asset, is sampled at 16:00 ET each trading day; series are inner-joined on session dates before any pairwise computation; timestamps are stored in UTC but aligned on session dates, never UTC days. A unit test asserts that aligned miner and bitcoin return series share an identical date index and that a one-day misalignment is detected. The analysis prices tab follows this rule; bitcoin's UTC daily close, which goes back further, appears beside it only with a note that it is not read at the same moment as a miner's close.
Universe rule and log
Inclusion rule: any company with current SEC or SEDAR reporting obligations for which self-mining of bitcoin is, or was at listing, a primary reported segment. Treasury companies are benchmarks, not miners. The rule is re-run quarterly.
Universe as of Sep 6, 2026: 18 miners, 2 benchmarks. Basis: the company list of the site's specification (section 4.2 and Appendix A), each company checked against its EDGAR submissions on Sep 6, 2026.
ANY is now DRK (rename or redomicile since the spec was written).
BITF is now KEEL (rename or redomicile since the spec was written).
DGHI is now DGXX (rename or redomicile since the spec was written).
GREE is now VIP (rename or redomicile since the spec was written).
MIGI is now BGDE (rename or redomicile since the spec was written).
Removed SDIG: Stronghold Digital Mining merged into Bitfarms (now Keel) 2025-03; 15-12G Mar 25, 2025; deregistered.
BTBT: Exited bitcoin mining for an ETH treasury strategy, reported Jun 25, 2025; included because it mined bitcoin when it listed
DMGGF: No SEC periodic reports (Form D only under CIKs 0001719754 and 0001725460, 2017 to 2021); TSXV DMGI, OTCQB DMGGF
BGDE: Nasdaq delisting determination Apr 17, 2026 (8-K Apr 23, 2026 Item 3.01); still listed per EDGAR Sep 5, 2026
DMGGF: no SEC record; No SEC periodic reports (Form D only under CIKs 0001719754 and 0001725460, 2017 to 2021); TSXV DMGI, OTCQB DMGGF.
BTC: no SEC record.
MSTR: no share-count concept in companyconcept; MSTR share classes are dimensional and omitted from companyfacts; fallback is the instance document or the Financial Statement Data Sets.
Pending extraction
Hashrate (EH/s), fleet efficiency (J/TH), energized MW, bitcoin mined, cost to mine (Methods A and B), power cost per kWh, guidance: metrics that live in prose and exhibits, not in XBRL, so this set does not carry them; the profiles show them from the releases and filings read, and as not covered where none was read, never estimated.
The operational metrics above come from two dated research passes covering 15 companies. The automated extraction that will refresh them on every filing has not run; the other companies show the filing set only.
Sources
SEC EDGAR submissions API (data.sec.gov/submissions/): the filings index, one request per company, declared user agent, 2 requests per second.
SEC XBRL companyfacts API (data.sec.gov/api/xbrl/companyfacts/): every tagged fact per company; the site keeps the 72 concepts its metrics use and links each fact to its primary document.
SEC XBRL companyconcept API: the share-count series behind the dilution chart.
Facilities: rows compiled from 10-K, 10-Q, 8-K and 6-K text by the research pass, each with its document, field and access date. Coordinates: US Census Bureau 2025 Gazetteer and GeoNames cities1000 centroids (for a region or country, the mean of the gazetteer cities inside it), as the row's geocode source says. Rows placed at a power plant use the U.S. Energy Information Administration's Form EIA-860 plant file (public domain; Source: U.S. Energy Information Administration, Form EIA-860, Sep 2026); a row placed at a filed street address uses the Geonorge address register in Norway ((c) Kartverket, CC BY 4.0), the U.S. Census Bureau Geocoder (public domain), the address points that Texas 911 districts keep, published by the Texas Geographic Information Office (TxGIO, CC0), for a rural Texas road address Census could not place; the address points that Tennessee's 911 offices keep, published by the Tennessee Emergency Communications Board (State of TN, Tennessee Emergency Communications Board; the state says there are no restrictions on the data use); the Kentucky 911 Site Structure Address Points (CC0; Kentucky 911 Service Board, Kentucky's PSAPs, via KyGovMaps Open Data Portal, Kentucky Division of Geographic Information); the Walker County, Georgia E-911 address points (county public record, no licence stated); and the address points that New York State keeps with its counties and local governments, published by the New York State Office of Information Technology Services (NYS ITS, CC0; the state's page says access is governed by Creative Commons 0); the Quebec rows placed earlier with OpenStreetMap Nominatim keep those points and their credit ((c) OpenStreetMap contributors, ODbL), and this site no longer uses that service; the point is the address, not the building, and each row says which service placed it. A row placed on a Texas county land parcel uses the StratMap Land Parcels layer that TxGIO builds from the county appraisal districts' rolls (CC0). The row gives the parcel number, the owner of record and the acres, and the point is the centre of the parcel. TxGIO presents the layer as the appraisal districts delivered it, for map use and not for detailed analysis, which is how the row uses it. The Clinton, Mississippi row uses the Hinds County parcel layer instead (a county public record that states no licence, credited that way); that layer draws the parcel as three separate pieces, so the point is the centre of the one piece that fronts the road, and the row says so.
Operational hero metrics: the research pass over 8-K and 6-K exhibits, monthly production updates and 10-K or 10-Q text, each figure with its document, page or field, access date and grade; computed figures name every input.
Monthly production: two passes over the same window, one reading the EX-99 exhibits every company filed, the other reading newswire listings, company newsrooms and sitemaps for releases that never reach EDGAR. For all 18 companies the second records where a monthly release is published, which feeds were requested and when. Both read hashrate, bitcoin mined, bitcoin held, energized megawatts, fleet efficiency and cost per bitcoin; each value keeps the label the release printed and the sentence or row it came from, checked back against that text. 9 of the 18 companies published a monthly release in the 12 months from August 2025 to July 2026 (4 file it as an EDGAR exhibit, 5 distribute it on a newswire): 3 in every one of the 12 months and 6 in only 1 to 6 of them; 7 of those state a bitcoin figure. The other 9 carry the feeds checked, how far back each index reaches and what each host answered. A month no index covers is recorded as not looked at, which is not the same as looked at and found empty.
At-the-market programs and shelves: 424B5, S-3 and F-3 documents of the window, classified by cover-page language rather than form; resale and selling-securityholder supplements are excluded. Program size is the authority the document states, with its basis; remaining capacity appears only where the document states one.
Prices: each company's latest daily close, the one the price tape and its page print, from Polygon.io's daily market summary. Daily closes since 2021 for the miners, bitcoin, the indices and the ETFs come from Polygon.io or Yahoo Finance, each marked with its source (Statistical comparison, below, counts them). Network hashrate and difficulty come from mempool.space.
Report an error to [email protected] with the page, the figure and the document that contradicts it.
Every report is checked against the primary document within 48 hours.
A figure that was wrong is corrected on its page, the page carries a note naming the change and the date, and the log records the date, the page, what changed and why.
A filer's own error (a mis-scaled or mis-tagged fact) is not corrected by hand: it is held for review and named on the methodology page until the filer amends or the reviewed value arrives.
A data refresh that moves an as-of date, and a wording change that changes no figure and no claim, are not corrections and are not logged.
MinerTerminal publishes data and educational analysis compiled from public filings. Nothing on the site is investment advice or a recommendation to buy or sell any security. All content is provided for informational purposes; readers must do their own diligence.
How to cite
Cite the figure, the company, the as-of date and the page, for example:
MinerTerminal, revenue, CLSK, as of Oct 14, 2026, https://www.theminerterminal.com/companies/clsk
Free to cite with attribution; licensing for redistribution: [email protected].
The formulas behind every statistic on the analysis page, the pairs it tests and why, and the rules that decide what is computed, what is caveated and what is flagged for the Learn section.
Formulas
Formula
Note
mean
m=n1∑ixi
the arithmetic mean of the n values
sample variance
s2=n−11∑i(xi−m)2
n minus 1 in the denominator; the standard deviation is its square root
median, quartiles
linear interpolation between order statistics
the convention numpy calls type 7
OLS slope
b=∑i(xi−mx)2∑i(xi−mx)(yi−my)
y regressed on x by ordinary least squares
OLS intercept
a=my−bmx
SST
∑i(yi−my)2
the total sum of squares
SSE
∑i(yi−a−bxi)2
the residual sum of squares, what the fit leaves
SSR
SST−SSE
the sum of squares the fit explains
R squared
R2=SSTSSR
the share of the variance of y the line accounts for
standard error of b
se(b)=∑i(xi−mx)2SSE/(n−2)
t-statistic of b
t=se(b)b
on n minus 2 degrees of freedom
p-value
two-sided, P(∣T∣≥∣t∣) under Student t with n−2 degrees of freedom
computed through the regularized incomplete beta function; the self-test pins it against scipy
a size pair is also tested on the logs; a non-positive value is dropped and named
rebased to 100
yt=100xfirstxt
x_first is the first quarter with a positive value; the overlay's second mode
log change
rt=lnxt−1xt
consecutive calendar quarters only, both positive; the overlay's third mode and the pooled tests' variable
annualized filed figure
quarter×4,half-year×2,fiscal year as is
so companies on different reporting cadences are set on one footing; the period is printed per company
gross margin
revenuerevenue−cost of revenue
both facts from the same period; a ratio, so nothing is annualized
derived fourth quarter
FY−(Q1+Q2+Q3)
only where all four documents exist; marked derived wherever it appears
daily log return
rt=lnPt−1Pt
consecutive sessions of one series; the variable of every price study
alignment
bitcoin at 16:00 ET on each NYSE session date
the miner's close and bitcoin's price read at the same moment; the UTC daily close is shown beside it with the caveat
correlation on returns
r and ρ of rt(miner) and rt(bitcoin) on the sessions both carry
with R squared and the two-sided p; windows of 30, 90 and 365 sessions and all in common
correlation on levels
r of Pt(a) and Pt(b)
shown because readers ask for it; two prices that both rose correlate whatever their days did (the spurious-correlation caveat)
rolling correlation
r over the w sessions ending at t, by prefix sums
w of 30, 90 and 365; one value in five sessions, the last 750 sessions
green-day table
2×2 of miner up or not against bitcoin up or not;P(miner up∣bitcoin up)=a+ba,P(miner up∣bitcoin not up)=c+dc
a zero return counts as not up and is counted
chi-square
χ2=∑E(O−E)2 on one degree of freedom, no continuity correction
p from the upper tail of chi-square(1); a cell expecting under 5 is flagged
leverage factor
the OLS slope b of rt(miner) on rt(bitcoin)
the empirical measure of how much of bitcoin's move the share carried; reported with R squared and the 95% interval b plus or minus t(0.975, n-2) se(b)
annualized volatility
sd(rt)252
sample standard deviation over the window
beta
the OLS slope of rt(miner) on rt(index)
against bitcoin (the leverage factor) and the S&P 500 separately
joint betas
rt(miner)=a+b1rt(bitcoin)+b2rt(S&P 500)+e
by the normal equations; each slope with its standard error and p on n minus 3 degrees of freedom
SMA n
mean of the last n closes
EMA n
Et=kPt+(1−k)Et−1,k=n+12
seeded with the SMA of the first n closes
RSI 14
100−1+G/L100,Gt=1413Gt−1+gaint
Wilder's smoothing; G and L the average gain and loss: simple means over the first 14 changes, then the rule shown, the same for L
MACD 12, 26, 9
line=EMA12−EMA26;signal=EMA9 of the line;histogram=line−signal
Bollinger 20, 2
SMA20±2sd
population standard deviation over the 20 closes, the Bollinger convention
Ichimoku 9, 26, 52
tenkansenkou Achikou=2H9+L9,kijun=2H26+L26=2tenkan+kijun,senkou B=2H52+L52=the close drawn 26 sessions back
The historical study uses highest and lowest closes. The interactive chart uses sourced daily highs and lows; missing bars invalidate that window. Its cloud stops at the last recorded session.
swing levels
a close that is the lowest (highest) of the 5 sessions on either side is a swing low (high)
over the last 250 sessions; levels within 2% merge; the most-touched of each kind is drawn
indicator study
f=lnPtPt+h,h∈{5,21}
forward return, split by whether a signal was on at the close of t; pooled across the miners; the difference of means with Welch's t and p; overlapping windows, so p overstates
Rules: a test needs at least 3 points; under 8 it is shown with the caveat that one company can move it; a finding at p under 0.01 on at least 8 independent points is flagged as an article candidate. Pooled quarter-to-quarter tests are never candidates: their points are not independent draws.
Levels against levels can correlate through size alone, so every size pair is also shown on logs and no sentence on the page says one figure explains another.
Every sentence the job writes says what the numbers are and where they lie, never what a company or the market is doing. A p-value under 0.05 is worded as the chance the strength would arise from nothing, not as proof.
Prices: one series of daily closes per ticker, each marked with its source. 12,457 of the 34,324 closes on file come from Polygon.io: its daily market summary for shares and ETFs, its aggregates for bitcoin and the indices (an index Polygon does not cover is shown through an ETF that tracks it, and the series is named so). The other 21,867 were filled once from Yahoo Finance's daily bars; a reader cannot check those against a documented source, the page says so wherever they are drawn, and a Polygon close replaces a Yahoo one for the same session the night it arrives. Share and ETF closes are adjusted for splits: every close before a split or reverse split is multiplied by its ratio, so a series stays on one share basis; each split comes from the company's filing or Polygon's list of splits, and a later split adjusts every earlier close the night the first price on the new basis arrives. Bitcoin's daily closes sit on the UTC day they close.
The technical overlay draws every indicator in the formulas above, computed in the browser by the site's own code; the nightly job computes the same indicators for its study, and a test holds the two to four decimals over sixty sessions. The study, not the chart, says which indicator went with the returns that followed, with the pooled caveat.
The job: a nightly script, run after the data refresh, using only Python's standard library. It writes the cross-section, the quarterly series and the findings, each stamped with when it ran and the code version. Its self-test pins every statistic against SciPy reference values. Last computed Oct 15, 2026 from data as of Oct 14, 2026.
How the tier list places a company's figure among the companies that report it, when a letter from S to F would print beside it, and why none does yet.
A category is one figure read one way, lower or higher. A company gets at most one place per category, printed beside the figure and the sentence it can be checked from; there is no whole-company letter.
A figure counts when it is reported, computed or computed-proxy. One the research marks as defined differently is not comparable and never ranked. One whose source measures something else (revenue over another kind of capacity, a subsidiary's fleet) prints with that difference and no rank, its bar drawn as an outline, and stops its category printing letters. Ranks leave these out: where the computation reads "3rd of 11", the page reads "1st of 8".
A figure counts only when dated within 400 days of the data's as-of date, the site's stale rule.
A company's place p is the share of the others it ranks ahead of, a tie counting half: p = (others it is better than + half the others it ties with) / (n - 1). A rank ignores the size of a gap, so one very large figure cannot squeeze the rest.
S is a figure ranked ahead of at least 90% of the other companies with a comparable, current figure, and F ahead of at most 10%; A to E divide the rest at 75, 60, 40, 25 and 10 percent. At the floor of 14 companies that is the top two and the bottom two.
Letters print only where at least 14 companies report the figure comparably, because below that seven letters need two companies each; below 14 at least one letter can only ever hold one company. Below it a category prints each rank alone: "7th of 11 that report it".
A missing figure is never a letter or a rank. The row says which of four is true: not reported, not comparable, not current or not on file, each with its reason.
Each row prints the figure with its period, its rank of the n that report it comparably, its as-of date and its source; the page prints the data's as-of date and the day the ranks were computed, again on every data change. The table sorts by its headers (by company and date only where the category does not rank, since ordering by figure there would be the ranking withheld). The whole sentence is the figure's title, letter first where printed. Where the site computes a figure, its arithmetic is in the row's note.
Never ranked or lettered: size, strategy, price and valuation, and results that bitcoin's price decides. Listed without a rank until the site can compute them the same way for every company: the share count change, the two costs to mine and months of liquidity.
A letter prints only where a category passes every rule above (n >= floor, the category is not gated, and none of its ranked figures carries a definition flag). None does on the data of Oct 14, 2026, so the page prints ranks.
A rank is within this set, not a verdict on anyone. Every input is the company's own disclosed figure, and nothing here is a view on its shares.
Who holds an executive office now: every change filed after the year a company's pay table covers, read from its report, and the officers the annual report of a company with no US proxy names; and where each named officer's pay falls among all of them.
The window opens the day after the latest fiscal year in a company's proxy pay table ends. From then, every 8-K or 8-K/A with Item 5.02 (a principal or named executive officer takes or leaves an office) is read in full, and so is every 6-K, the form a foreign issuer files its news on, which has no item numbers; Bitfarms filed on it until it became Keel on April 1, 2026. A change records an office taken, left or exchanged, with its effective date and the sentences it was read from, in the change's title. An office left for a non-executive role, such as an adviser's, reads "left as" the office, then the role. Directors' seats, pay plans, results and financings are listed as reports with no change of executive office.
A Form 20-F filer has no proxy and no Item 5.02, so its 6-K reports since the annual report are read instead, and the executive officers its annual report names (Item 6.A) are listed with any later change. A company not on EDGAR shows the officers its annual information form names, from the copy OTC Markets serves, and nothing later.
A former officer is shown as the proxy prints the position. A company's report count is every such report in its filings index after the window opens, by form. An unread report is listed as not yet read, never as no change, and while one is unread no officer on that roster is said to have no change filed. Offices below the principal officers (a senior vice president named only in a press release) are not covered.
People pages offer twelve months, twenty-four months and All captured sale periods; All captured keeps imported ownership filings across refreshes and claims no lifetime coverage. Company totals count each transaction once; a joint filing counts in full for each reporting owner. USD totals sum known values only; shares with no stated price stay in total shares and show separately. Reviewed amendments replace their original rows before any period is summed. Pay rankings compare total compensation; base salary is in each officer's proxy table.
The pay percentile and pay against the median compare each named executive officer's total pay (latest year in the Summary Compensation Table, the figure the people list prints as latest pay) with every other named officer's: 61 officers at 15 miners, newest table filed Jul 29, 2026. The percentile is the share of other officers paid less, 100 times the officers paid less divided by the officers less one; equal pay shares a percentile, so the lowest total reads 0 and the highest 100 unless tied. It is taken again within the officer's role, read from the table's position: chief executive (CEO, PEO, co-chief, interim and former included), chief financial officer (CFO, PFO), or other (every other position, or blank); a position naming both counts as chief executive. Every percentile states its n, and a group under 5 prints words instead. Against the median is total over the median of all 61 totals, with the dollar difference beside it and the same two against the mean in its title; the median is the reference because a few very large totals pull the mean up. Pay over market capitalization is ranked the same way among the 35 officers whose company has one in the research pass. A former officer is ranked in the office the table names, and a final-year total can include severance or part of a year. Fiscal years differ: the latest years on file are 2025 and 2026, and fiscal years end on 31 March, 30 June, 30 September and 31 December, so the ranks compare overlapping years. A percentile is a rank within this set, not a verdict.
Pay over market cappay_to_market_cap
100 x total pay / market capitalization, so a higher figure is more pay per dollar of market value
An officer's total pay in their latest year in the company's Summary Compensation Table over the company's market capitalization in the research pass (the closing price on its as-of date times the latest filed share count); for a company's top-paid officer the pay is the top officer's total pay. A company with no market capitalization in the pass prints words. The pay year and the market capitalization's date differ, and both are stated.
The people list prints it for every named officer, a company page for its top-paid officer, and the home operations table for each company's top-paid officer with the company's place by it among the companies that report both a top pay and a market capitalization: 1st is the largest share, the count of those companies follows the place as the tier list prints it, and two equal shares are tied. It is a place among those companies, not a percentile, because a percentile of so few companies claims a precision the sample does not have.
Which aircraft the public record ties to a covered company, where they flew, and the one rule: the pages record aircraft, never people.
An aircraft is tied to a company by the FAA's releasable register and the company's filings: confirmed when registered to the company by name or named in a filing, probable when registered to an LLC or trustee tied to it by address, officer or filing. Flights are cut from public ADS-B positions (adsb.lol, Open Database License), and each flight names its source file. Who was aboard is not in the data and no page states it; no individual is named, and a filing quote that names a person shows the role instead.
Filings are searched for aircraft wording well beyond the register, and each search below covers only what it names. EDGAR full-text search on aircraft and charter terms covered the 17 covered companies that file with the SEC; it returns annual and quarterly reports and exhibits such as press releases, shareholder letters and employment agreements. Proxy statements and their amendments filed since June 2022 were read for the same wording, for the companies that file them. The reports and exhibits that four of those companies file as foreign issuers (forms 6-K, 20-F and 40-F) were searched as well. Bankruptcy schedules were read for one company, Core Scientific, and for no other. A company that discloses an aircraft without a tail number is listed as a disclosure with its quote, never matched to a registration by guesswork. Every statement that no further aircraft could be tied to a company rests on the FAA register as it stood on Sep 22, 2026: the FAA's download now refuses automated requests, so that copy was not read again, and a registration made after that day is not seen. Not read: Canada's register and earnings-call audio (both forbid automated access); TeraWulf's sustainability report and CleanSpark's investor-relations site (both refused access); state lien records (paid or behind a CAPTCHA); and proxy statements filed before June 2022.
An aircraft whose owner or operator asked the FAA to limit display of its flights (LADD) has its flights listed with a note saying so, since they were broadcast in the clear. An aircraft registered to several names or to a trustee has its flights listed as the aircraft's, not attributed to the company, and no share of the company's figures is computed for it.
How a flight's estimated variable cost is computed
The FAA's average variable operating cost of a flight hour for the aircraft's class (fuel and oil, maintenance and engine reserves, and crew), in 2018 dollars, times the flight's airborne time as these pages measure it, from the first to the last airborne position received, which leaves out taxiing; rounded to the nearest $100. It is a class average weighted by the number of aircraft registered, not the figure of this aircraft's own type, and it understates a large-cabin jet such as a G450, since most registered jets are smaller. It is the cost of that flight time for an aircraft of the class, whoever flew it: not what the company spent, and not what a fractional co-owner is billed.
To have an aircraft's flights reviewed or removed, or to report an error a document contradicts, write to [email protected] with the registration and the page. A person reads each request; corrections follow the corrections policy.
These are hypothetical calculations, not forecasts or price targets. Presets combine the latest figures on file, from different periods; each links to its dated source. Empty fields stay missing until you fill them.
Production
Your share of network hashrate, 144 blocks a day.
An average month is 365.25/12 days.
Uptime starts at 100 percent and scales realized hashrate.
The subsidy halves in the chosen month.
Bitcoin price and fees stay fixed.
Network hashrate compounds by your monthly growth.
Power
MW=EH/s×J/TH.
Facility MW caps the fleet, existing machines first.
With facility power empty there is no cap and no facility construction cost, so the dilution chart spends its equipment allocation on machines alone.
New capacity starts after the installation delay.
Equipment is modeled by capacity, efficiency, price per TH and useful life, with no manufacturer specifications.
EBITDA
EBITDA subtracts power, hosting, SG&A and other costs.
Changing chief executive pay shifts SG&A by the difference from the pay baseline (initially proxy total compensation).
A replacement needs a known baseline, so enter one if the preset has none.
Adjusted SG&A cannot go below zero.
This is a sensitivity assumption: proxy awards are not necessarily cash paid or expense recognized in the same fiscal year.
Cost per bitcoin and free cash flow
Method A divides power and hosting by bitcoin mined.
Method B adds SG&A, other costs and depreciation.
Free cash flow subtracts interest and new equipment and facility spending upfront.
Excluded: taxes, working capital, maintenance capital, non-mining revenue.
Missing existing-debt interest counts as none and is flagged in results.
Financing
At-the-market issuance sells at your share-price input less its discount.
Debt adds principal and interest.
A convertible converts fully at the premium, or stays debt, and its coupon is charged throughout the horizon.
There is no conversion date, debt maturity or repayment schedule.
Cash on hand raises nothing.
Negative cash after spending is an unfunded gap.
Value per share
value per share=shares after financingmultiple×EH/s built−debt+cash
EV per EH/s value: the fleet once built, less debt plus cash after financing and spending, over shares after financing.
It does not discount the installation delay or add bitcoin holdings separately.
Cash-flow value adds bitcoin holdings, cash and discounted operating cash flows, subtracts debt, and divides by shares.
Capital spending is deducted once.
No terminal or equipment resale value is counted.
Reading the results
The market-close tile always shows the sourced historical close; your share-price input changes issuance arithmetic only.
Charts and scenarios vary one input and hold the rest fixed.
The dilution chart always uses at-the-market issuance.
No comparison is a recommendation.
Starting assumptions where no company figure is on file
Network growth: 0 % a month. Held flat unless you set it: the network's growth is not in the files.
Uptime: 100 %. Of the realized hashrate, which already carries the company's own downtime.
Power price: 50 $/MWh. Companies report a cost per bitcoin, not a price per MWh.
Pool fee: 2 %. Set 0 for a company that runs its own pool.
Hosted share of the fleet: 0 %. Machines at another company's site, paying its price.
Hosting price: 70 $/MWh. All-in, power included.
Other costs a year: 0 $M. Anything the lines above leave out.
New capacity: 0 EH/s. None unless you add it.
Machines' price: 15 $/TH. A round price per terahash; set the one you know.
Build price per MW: 0.5 $M. Power built beyond what is energized today.
Months to install and energize: 6 months. The new machines hash from the month after.
Machines' life: 3 years. New machines depreciate evenly over it.
Amount to raise: 0 $M. None unless you raise it.
Discount to the close: 3 %. What the sale gives up to the close, fees included.
Coupon: 2 % a year. The note's interest.
Conversion premium: 30 %. Over the close.
Interest rate: 10 % a year. On the new debt.
Discount rate: 12 % a year. What a dollar a year out is worth today.
These are hypothetical calculations from your inputs and figures on file, not forecasts, price targets or advice. A raise is accretive when the price per share after it is above the close, so each raised dollar became more market value than the break-even; destructive when below. These words name the hypothetical cases, never a company.
The scenario starts from a company's filed share count and its latest close, each dated and linked, the way the workbench starts. New shares are issued at the close less your discount: new shares=close×(1−discount)amount. The whole raise uses one issue price, with no fee beyond the discount and no market move during the transaction.
Money made per diluted dollar is the market value each raised dollar becomes. It is your input. At 3.00, one raised dollar becomes three dollars of market value. market cap after=market cap before+amount×money made per diluted dollarprice per share after=shares aftermarket cap after Nothing else moves the price.
Each raised dollar's market value splits two ways. The new shares it paid for are worth sale priceprice after. Earlier holders keep the rest, which equals amountshares×(price after−close). Break-even is sale priceclose. At that value the price per share does not move. With no discount, break-even is exactly one dollar.
The graph draws the same raise at every amount from none to at least the market cap before it, in round steps, so a line and the step for your amount agree where the marker stands.
The spot bitcoin preset sends the proceeds into bitcoin at the bitcoin close, in the style of MSTR, the site's benchmark for raising to hold bitcoin rather than to mine it. Bitcoin held per share rises only when shares are issued above the value of the bitcoin each share already holds. The premium of market value over the bitcoin bought is an assumption, not a company figure.
The replay divides monthly hashrate and bitcoin held by the filed share count on the same date. A count dated on a month's last day joins that month's release; a count on any other day joins nothing, and no figure is matched to a nearby date. The company page supplies the share counts, starting after the latest break in capital structure; nothing is divided across a consolidation, reverse merger or disputed date. Peak and installed hashrate are left out. Every figure in one series must use the same label.
Hashrate per share is in PH/s per million shares (1EH/s=1,000PH/s); bitcoin held per share is in BTC per million shares. The indexed view sets each date as 100×value on the first datevalue on the date. The per-share index is always the release figure's index over the share count's. Tagged crypto-asset units are not used as bitcoin held because some cover only part of the holdings stated in releases.
The cross-company view of dilution against hashrate growth is drawn once at least eight miners have two aligned dates for hashrate per share. A miner the replay cannot use is listed with the reason in words.
CLSK
CORZ
HUT
IREN
MARA
WULF
MSTR
BTBT
KEEL
DMGGF
VIP
DGXX
DRK
BGDE