Skip to the guide

Proof

Every score, map and model on this site is graded by rule against what happened next. The grades are public, the poor ones too.

  1. 143%Liquidation mapof its estimate liquidated
  2. 5 of 29Indicator signalsheld up
  3. 1.19×Leverage gaugethe usual move after extremes
  4. 10 of 10Market regime scoredays published, none revised
  5. 81%Price rangesof 80% ranges held
  6. 4 of 15Strategiesbeat holding the coin
  7. 23Correctionspublished
  8. 91.0%Data on timeof hours fully on time

Our liquidation map estimates too little: exchanges liquidated 143% of what it showed on hours it had never seen. Of 29 signals from 16 well-known indicators, 5 held up and 9 showed no edge or worse. One-year price ranges held 81% of the time against the 80% they promise. 4 of 15 published rules beat holding the coin out of sample. The regime score has 10 days on its record, none revised. 23 corrections to our own figures are published.

Recomputed 17 Nov 2026, 00:57 UTC from live data. No figure on this page is typed.

Liquidation map

Measured against what exchanges liquidated

143%of its estimate was liquidated

Estimates too little

On 10,800 hourly windows from 23 to 27 Oct 2026 that its calibration never saw, Bybit liquidated 143% of what the map estimated at the levels the price reached, so the map shows too little, not too much. For Bitcoin and Ether the share was 127%. 100% would be an exact estimate.

Scored . Shown on: Liquidation heatmap.

  1. Every coin143%
  2. Bitcoin and Ether127%
  3. Every other coin189%
  4. An uncalibrated map6%
Share of the map’s estimate that Bybit liquidated within 60 minutes at the levels the price reached, on the 10,800 held-out hourly windows of calibration version 3. The uncalibrated map has fixed leverage weights and no fitted size.

Indicator signals

Every past signal, graded

5 of 29signals held up

9 showed no edge or worse

29 signals from 16 well-known Bitcoin indicators are graded on what the price did next and on how close each came to a cycle top or bottom: 5 held up, 15 are mixed or have fired too few times to judge, and 9 showed no reliable edge or did worse than a typical day.

Graded . Shown on: Top indicators, Cycle score.

  • Held up 5
  • Mixed 14
  • Too few signals to judge 1
  • No reliable edge 3
  • Worse than a typical day 6
IndicatorTop signalBottom signal
MVRV Z-ScoreHeld upWorse
Pi Cycle TopHeld up–
Puell MultipleWorseMixed
NUPLHeld upWorse
Rainbow chartHeld upMixed
Power lawMixedHeld up
Bitcoin Fear & Greed IndexWorseMixed
Bitcoin funding rateWorseNo edge
Altcoin Season IndexNo edgeNo edge
Mayer MultipleMixedWorse
200-week moving averageMixedMixed
Reserve RiskMixedMixed
Hash Ribbons–Mixed
Mixed
SOPR–Mixed
Pi Cycle Bottom–Mixed
Cycle Composite ScoreToo fewMixed
One mark a signal, at the level people quote. A signal is graded on the price a year and 90 days later against a typical day of the same cycle, and on whether it came within 90 days of a cycle turn.

Leverage gauge

What followed every hourly reading

1.19×the usual move, after extreme readings

Bigger moves followed

Across 30 coins, the median move in the 24 hours after an extreme reading was 1.19 times the same coin’s usual 24-hour move, and 0.92 times after a low one, over 1,185,673 hourly readings since 2020. The record grades the size of the move, never its direction.

Recalculated . Shown on: Leverage gauge.

  1. After a low reading0.92×
  2. Elevated0.97×
  3. High1.05×
  4. After an extreme reading1.19×
Bars: the median 24-hour move after each band, every coin together, as a multiple of the coin’s usual move. Ticks: each coin after its own extreme readings. 29 of the 31 coins with a day or more of extreme readings moved more than usual after them.

Market regime score

Kept as published

10 of 10days published, none revised

Never revised

Since 7 Nov 2026 each closed day’s score has gone into an append-only record exactly as it was published: 10 days so far, the latest 33 (defensive) for 16 Nov 2026. A row is never revised, even when an input is corrected later, and a day we fail to publish stays empty.

Latest entry . Shown on: Market regime.

  1. 447
  2. 438
  3. 449
  4. 4410
  5. 4411
  6. 4012
  7. 3713
  8. 3214
  9. 3415
  10. 3316
  11. 3317
Each square is one closed day’s score as published, with its day of the month. The gold outline is today, still open; an empty hatched square would be a day we failed to publish.

Price ranges

Past ranges against where the price ended

81%of one-year ranges held, against the 80% promised

As promised

On 592 past days, one a week since July 2014, the resampled-path model gave Bitcoin an 80% range for a year later. The price ended inside it 81% of the time, above it 3% and below it 16%.

Recomputed . Shown on: Price prediction, Every coin’s ranges.

Bitcoin’s 80% ranges from the resampled-path model, one a week, by how far ahead they looked. The 50% ranges held: 1 month 56%; 3 months 52%; 6 months 51%; 1 year 49%.

Strategies

Published rules against holding the coin

4 of 15rules beat holding, out of sample

9 grew slower than holding

17 published rules are simulated on our own daily data, after costs. On the years each rule’s levels were not fitted to, 4 of the 15 with an annual growth figure grew faster than holding the coin, 2 matched it and 9 grew slower; 9 fell less far at their worst. Each rule’s live state has gone into an append-only record since 18 Oct 2026.

Recalculated . Shown on: Strategies.

  1. Valuation bands+19.8
  2. Cycle temperature+7.9
  3. Short-term holder cost trend+7.0
  4. ETH 200-day trend+1.8
  5. Miner capitulation recovery0.0
  6. Conviction bands0.0
  7. Bitcoin cash-and-carry−1.1
  8. ETH cash-and-carry−1.1
  9. Price stretch−4.7
  10. Bitcoin 200-day trend−8.0
  11. ETH value zone−13.1
  12. ETH short-term holder cost trend−15.2
  13. Loss-selling reset−21.9
  14. Long-term holder turns−26.0
  15. ETH above holders’ cost−35.8
Each bar is one rule: its annual growth minus that of holding the coin over the same out-of-sample years, in percentage points a year, after costs. 2 rules with no annual growth figure (bought options) are not drawn.

Corrections

Every correction, published

23corrections published since 7 Nov 2026

Listed first on the changelog

Every correction to a figure we had published is listed with what was wrong, from when, and what the figure reads now: 23 corrections since the log began on 7 Nov 2026, the latest on 13 Nov 2026.

Latest correction . Shown on: Data changelog.

  • Bitcoin on exchanges restated: seven wallets no exchange lists are removed
  • Exchange-to-exchange flows recomputed on the corrected exchange labels
  • Ether flows by holder segment recomputed on the corrected exchange labels
One square a correction, on the day it was published.

Data on time

Checked every ten minutes

91.0%of hours with every reading on time, the 10 days since checks began on 8 Nov 2026

A whole group late in 5.1%

Every ten minutes each data group’s latest published hour is checked against the latest completed hour. Over the 10 days since checks began on 8 Nov 2026, every coin and feed of a group was within its stated delay in 91.0% of the group-hours checked; in 3.8% some were late, and in 5.1% the group as a whole was. The lowest was ETF issuer files, at 51.6%.

Checked . Shown on: Data status.

  • Bitcoin97.5%2.5%
  • Ethereum79.7%20.3%
  • Tokens on Ethereum77.7%20.8%
  • Litecoin98.0%2.0%
  • Dogecoin100.0%0.0%
  • Bitcoin Cash100.0%0.0%
  • Market data94.1%3.0%
  • Derivatives97.7%0.0%
  • Options chains100.0%0.0%
  • ETF issuer files51.6%0.0%
  • TradFi perpetuals99.2%0.8%
One column a day, oldest first: the share of group-hours in each state. A group counts as late at its own stated delay, two to eight hours for chain data and each feed’s own due time.

Where our numbers differ from ones you know

Three of our readings sit apart from figures that are widely quoted. Each counts something specific and leaves something out, and says so.

  1. Bitcoin on exchanges

    1.08 million BTC

    As of 17 Nov 2026, 00:29 UTC

    Counts only the wallets of the seven exchanges that publish their own addresses (Binance, Bitfinex, Bybit, Crypto.com, HTX, KuCoin and OKX), read on our node. Exchanges that publish none, Coinbase and Kraken among them, are left out, so the figure sits well under totals that estimate those wallets too: 2.4 to 3.0 million BTC as quoted in November 2026.

  2. Fear & Greed Index

    30 of 100, fear

    As of 17 Nov 2026, 00:00 UTC

    Built only from data we collect. Derivatives positioning (funding, open interest, taker flow and the long/short ratio) carries 40 of its 100 points and each part is ranked against its own last two years; it uses no survey or social-media input. When traders add leverage into a falling price it can read greedier than indexes that lean on momentum and sentiment.

  3. Market regime score

    33 of 100, defensive

    As of 17 Nov 2026, 00:48 UTC

    The plain average of ten lenses, from dollar liquidity and fund flows to leverage and price structure, each scored against its own last two years. It is not a sentiment gauge: it can read neutral on a day sentiment reads fear, when flows, leverage and trend pull against each other. Today’s ten lens scores are shown beside it.

Check it yourself

Each record has a page of its own with the method and every row behind the figure, linked beside it. The figures above are read from the same documents those pages show, so the two always agree. Where a record is young, its start date is beside it and the count is what it is.

A grade is fixed by a rule written down before the result: an indicator signal is graded on the price a year and 90 days later against a typical day of the same cycle; the liquidation map on hours its calibration never saw; a price range on where the price ended; a strategy on years its levels were not fitted to. How every chart is made is set out in our methodology.

Everything on this page is free to quote and reuse under CC BY 4.0, crediting CryptoKeyhole with a link to this page. If you think a figure of ours is wrong, write to hello@cryptokeyhole.com: a figure we correct is listed on the data changelog with what was wrong and what it reads now.

  • /api/v1/proofEvery record: its figure, sentence, date and the data behind its drawing.
  • /api/v1/proof-linesThe one-line record shown under each verdict and chart on the site.