This report is directed only at investment professionals within article 19 of the Financial Services and Markets Act 2000 (Financial Promotion) Order 2005. Any investment or investment activity to which it relates is available only to such persons. Persons who do not have professional experience in matters relating to investments should not rely on this report. It must not be distributed to, or relied upon by, retail clients. This report is provided for information only and is not investment advice or a personal recommendation. Bitcoin and other cryptoassets are high risk and volatile, and an investor may lose all capital invested.
Synopsis of drawdown
Bitcoin's drawdown began at the cycle peak of ~$125,000 on 7 October 2025. Three forces met at the top. First, on the 10th of October, a leverage-liquidation cascade triggered the largest digital-asset deleveraging event on record, resulting in approximately $19bn of liquidations across all digital assets and more than $24.9bn of open-interest reductions among major assets. This was compounded by Satoshi-era HODLers steadily distributing their coins throughout 2025, a period dubbed Bitcoin's "IPO era", and by fears that quantum computing could one day "hack" Bitcoin wallets, together eroding sentiment and trust in Bitcoin's proposition as a store of value. Meanwhile, institutional demand via ETPs and treasury companies also started to decelerate into year-end.
Into the new year, a reflation trade took hold. A China-led credit impulse and a business-cycle upturn, buoyed by improving expectations for AI productivity and manufacturing output, sparked a commodity rally in precious metals (gold and silver) and oil that took the wind out of Bitcoin's sails. Bitcoin fell to the cycle's lowest level at that point in time, around $62k in early February, with the 5th of February flash crash marking the largest nominal daily swing on record, roughly $10.2k intraday. That capitulation began a seven-month "chop-solidation" within a ~$62-82k range.
The second quarter appeared to flip the backdrop on its head, with the market pricing implied cuts around incoming Fed Chair Warsh's swearing-in. That dovishness never materialised. Instead, resulting in a hawkish June FOMC with an inflationary backdrop taking centre stage from continued Middle East tensions and rising expectations that AI productivity would feed through to growth. Q2 was further riddled with sovereign bond stress: JGB yields rose to multi-decade highs, driving the yen-carry-trade unwind, while the US 30yr hit its highest since 2007, wringing any remaining juice out of Bitcoin. With sentiment at rock bottom and investors disengaged, these pressures culminated in what was likely a possible cycle low of $58k on 30 June.
Bitcoin’s market positioning changed over the summer: relative to precious metals and oil, where many investors had been burned in the Q1 rally and faced an uncertain outlook from the US-Iran conflict, and relative to AI equities, which, though subject to heavy hoarding and index concentration, faced mounting headwinds from decelerating token demand, circular vendor financing, Chinese competitors and open-source models. August arrived amid record-low volumes and volatility, yet Bitcoin took a run of negative headlines in its stride: the Coldcard hack, the Strategy overhang, and the CLARITY Act missing its Senate floor-vote window before the August recess (a procedural vote on 15 September could still bring it to the floor as pending business).
On-chain indicators suggested that both the ‘price pain’ and the ‘time pain’ may have been moderating. Looking at previous cycle bottoms, many metrics looked very similar. A few include the number of coins redistributed from cycle top to bottom, the share of supply in loss, the Long-Term Holder supply at all-time highs; and technical measures such as volumes and bear-market pricing levels all pointed to a bottoming process potentially drawing to a close.
Macro backdrop
Bitcoin as the canary in the coal mine sniffed out tightening liquidity conditions
After the Q4 2025 liquidation beat-down with no reprieve from financial conditions, liquidity continued to tighten in 2026 as rate-hike expectations increased in late March-April, then further into genuine sovereign-bond and real-rate stress by June-July. Hyperscalers and software stocks dependent on debt financing also took a beating as financial conditions tightened. What added to Bitcoin’s 50%+ drawdown, is now working its way through the rest of the system.
Bond markets are cracking now
The strain has since moved into US sovereign bonds, with the "debasement trade" rearing its ugly head on fiscal credibility concerns, rather than liquidity concerns. Bond vigilantes had driven the long end of the curve (30-year) up to ~5.3%, and on Wednesday 19 August the Treasury announced it would at least double its long-end buybacks, from $2bn to at least $4bn per operation, to stem the decline in bond prices. Within 24 hours, the 30-year yield had retested ~5.2%. This quasi yield-curve control, together with announcements of the potential drawdown of the ~$950bn Treasury General Account (TGA), and the joint FX intervention of the US Treasury department with the Japanese Ministry of Finance showed the US Treasury’s preference to monetise its own debt at the expense of its currency while letting deficits balloon. That backdrop is fundamentally dollar-bearish and coincided with Bitcoin's and gold's +24% and +6% weekly returns respectively. A decisive fiscal reversal – for example, cuts to spending and smaller deficits – could alter that backdrop. This is a longer left-tail scenario but would likely weigh on risk assets.
On-chain Model
When people talk about market bottoms, they usually focus on the buy side. They wonder: “when will the new buyers come in?” But market bottoms occur when residual demand exceeds exhausted supply. This model incorporates both angles using Long Term Holder Spent Output Profit ratio for the sell-side leg, and the Accumulated Trend score for the demand-side leg. It produces a probability score to determine the odds of a bottom relative to previous cycles. To learn more, see the methodology section.
Results
The current readingOn 30 June the calibrated probability score of a bottom peaked at 42%, which a seven-day average smooths to 30%. This is a robust reading, albeit lower than at previous cycle lows. It is extremely elevated compared to readings over the past two years. It reflects long-term holders realising losses more deeply than 90% of its history. This is stretched, though short of the close to-100% percentile capitulation seen at previous lows.
About two-thirds of the reading came from LTH-SOPR and one-third from ATS, even though the demand leg sat at its most aggressive level on record. This is a deliberate choice where the model weighs on the capitulation leg more heavily, such that strong buying alone cannot lift the score to bottom-like levels until sellers are exhausted. With capitulation only stretched rather than extreme, the score sat below past cycles rather than alongside them.
Blue = LTH-SOPR push, green = ATS push (log-odds vs average). Dashed verticals = cycle lows.
In the 2022 cycle, the first calibrated score fired roughly five months before the post-FTX low, clearing the 30% threshold as the Terra-Luna complex collapsed. The signal then peaked again at the confirmed November low. On both occasions the model rose above the 4% base rate, a measure of how common cycle bottoms is as a share of all days. The model therefore needs monitoring and is better read as a reference for market conditions at the time.
The February 2026 drawdown to $62.7k, almost $4k above the 30 June low, never cleared the 4% base rate, not even during the 5 February fall, the largest single-day open-to-close decline on record at $10.2k. Long-term holders rarely react to fast, volatile moves. The 30 June reading, by contrast, sat well above the baseline, marking it out from an ordinary day.
Top = BTC price (log); dots = 2022-11-21 bottom, Jun-2026 low.
Bottom = calibrated OOS P(bottom): faint = raw, bold = 7d smooth.
Dashed verticals = those lows. Dashed line = base rate (~4%).
Even so, the forward returns from those live signals varied considerably: a median (middle-of-sample) outcome of 54% over six months and 71% over a year, with hit rates (the share of signals followed by a positive return), of 50% and 75% across the four prior bottoms. Note that these figures are drawn from a small sample so may not be repeated.
Peak/Trigger = first date smoothed P sets a new high >=30%; Episodes = distinct >=30% clusters (>=14d apart); 6m/1yr fwd = BTC return +182d/+365d from that date.
With full hindsight, taking each cycle's peak reading, the forward returns were a median of 74% over six months and 114% over a year, and were positive in all four cases. This is illustrative rather than robust, resting on only four cycles and on peaks identified after the fact. Even so, those peaks landed close to the eventual low: within a few days in 2018 and 2022, and within about two weeks in 2015.
Peak/Trigger = date of the global peak in smoothed P; Episodes = distinct >=30% clusters (>=14d apart); 6m/1yr fwd = BTC return +182d/+365d from that date.
We read the 30% as potentially constructive signal rather than a definitive one. It sits at a level where the model has historically begun to identify conditions associated with previous cycle-bottom windows, and the two forces that define a bottom are both pointing the same way: sellers were visibly stretched and demand was as strong as it had ever been.
Since 30 June the market has also absorbed a run of bad news, from the war in Iran, contentious Bitcoin upgrade proposals and the Coldcard hack to the Strategy overhang and the CLARITY Act failing to pass before the August recess, without breaking down.
Set against an improving macro backdrop for risk assets and continued institutional adoption, we lean to the view that the low may be behind us. In short, the case for a bottom may look stronger than the score alone implies. This is a qualitative interpretation layered on top of the model and is not an investment recommendation or a forecast of future performance. A renewed sell-off or deterioration in macro conditions could, of course, invalidate that thesis.
Is the market behaving as if the low is in?
Bitwise's on-chain analysis asks whether seller exhaustion and demand conditions make a market bottom plausible. The CFB Full Confirmation Engine (Model A) asks the next question: after a low, is the wider market behaving as if it can hold? It looks for confirmation in price trend, participation, market structure, flows, derivatives and the macro backdrop.
The two approaches were developed independently for different purposes, use different evidence and are not combined into a single score. Their increasingly constructive readings after the June low are consistent with the possibility that June 30 was likely the cycle low, while leaving confirmation work still to do.
CFB evidence from June through August
The technical evidence did not call a bottom on June 30. It strengthened in the weeks that followed. In the fully covered research replay, the CFB publication label moved from Clearly bearish at the low to Clearly bullish by August 26 as participation, trend, flows and derivatives improved together. The engine is designed to identify whether a recovery is broad enough to deserve confidence, not to pinpoint the exact turning point.
Source: CFB Full Confirmation Engine (Model A) point-in-time research replay through 3 September 2026. The current reading uses the model's reported weighted coverage and confidence rather than treating missing inputs as neutral.
How the CFB Full Confirmation Engine works
The Full Confirmation Engine is a 45-signal framework organized into eleven categories. It is designed to distinguish a broad, durable advance from a price-only move or a rally supported by narrow, fragile conditions. The category mix is fixed and shown below.
Within each category, inputs retain their documented relative weights. Each input combines its current level (34%), three-month direction (33%) and 10-year historical z-score (33%) into a score from -1 to +1. The category scores are then combined using the fixed weights above. Missing inputs reduce reported coverage rather than being treated as neutral; breadth and momentum are critical coverage categories.
The reader-facing four-state publication labels are separate from the underlying decision-layer buckets used in the replay. The decision layer also applies a breadth and momentum gate, state persistence and documented fragility or transition overrides. Coverage and agreement among category scores determine confidence, so a high composite with incomplete inputs does not carry the same weight as a fully covered reading.
Macro conditions in the confirmation engine
Bitcoin does not trade in isolation from the macro environment. Bond markets, real yields, the dollar and risk appetite shape the liquidity conditions in which crypto assets trade. In this episode, the macro portion of the CFB framework improved alongside the rest of the technical picture, rather than carrying the conclusion by itself.
Inflation and policy expectations can still change the backdrop quickly. A durable bottom needs the price trend, momentum and breadth to agree, with macro support reinforcing that broader evidence rather than replacing it.
Confirmation has historically arrived after the low
In December 2018, the price-only H0 trend framework did not turn risk-on until February 22, 2019, 69 days after the Bitcoin low. In November 2022, the reduced historical CFB core remained in bearish consolidation through the low and first turned bullish consolidation on February 19, 2023.
Those observations are not retrospective recommendations and they are not evidence that every recovery will persist. They explain why the CFB framework should be read as a way to test whether a low is being ratified, not as a tool that claims to pinpoint a low before it happens.
The Full Confirmation Engine chart
The chart below shows only the CFB Full Confirmation Engine (Model A), the 45-leg, 11-category Broad Cap engine, from 11 April 2023. It deliberately excludes the CFB Historical Core, so the visual history matches the full model's available data rather than suggesting that the full engine has been backfilled.
Source: CFB Full Confirmation Engine (Model A) point-in-time research replay from 11 April 2023 through 3 September 2026. Regime labels are research-replay outputs and do not include the live state machine or override actions. The full engine is not a prior-cycle backtest.
How the on-chain and technical models complement each other
The on-chain and technical frameworks do not answer the same question, so their agreement is not a mechanical average. Bitwise assesses whether supply pressure and demand conditions make a reversal plausible. The CFB Full Confirmation Engine assesses whether the market later shows the breadth, leadership and structure that would support that reversal. The Bitwise reading was constructive around the June low; the CFB reading strengthened as price, participation, flows and derivatives improved afterwards.
Bitwise found elevated capitulation that remained short of earlier-cycle extremes, alongside record demand-side accumulation. Its 30% smoothed on-chain score is below earlier cycles partly because the model assigns more weight to capitulation than demand and because larger, more widely adopted markets may experience shallower drawdowns. Bitwise therefore does not treat the lower score alone as evidence that the bottoming process is incomplete.
Taken together, the evidence is consistent with the possibility that the June 30 low may prove to be the cycle low. The CFB evidence lends additional support to that possibility, but it does not establish a durable cycle bottom. Broad participation, fresh coverage and a stable macro backdrop are still needed for greater confidence.
Evidence that would strengthen or weaken the case
Fresh coverage across breadth, flows, rotation and leverage would strengthen the case. So would a positive CFB Bitcoin Trend Check (H0) through a retest and a live-engine reading that reflects a broad advance rather than a score driven by a reduced set of inputs. Constructive volume confirmation would add confidence, but it cannot create the state on its own.
A confirmed H0 risk-off transition, renewed breadth deterioration once current data are available, persistent adverse rotation or leverage, or a failed retest that reduces the post-June move to a short-lived price-only recovery would weaken the case. Another leg down remains possible. Bitwise identifies falling G10 excess liquidity and wider lower-rated high-yield spreads as risks to risk appetite, alongside the possibility of a sell-off in sovereign bonds or equities.
Bitwise's longer-term hard-assets thesis provides separate strategic context: high global debt burdens can create incentives to ease financial conditions and may support demand for Bitcoin and gold. That thesis does not substitute for the CFB confirmation signal or determine the near-term price path.
Detailed Full Confirmation Engine evidence around the June low
Path Forward
Based on CF Benchmarks' and Bitwise's analysis, the combined indicators are consistent with the possibility that 30 June marked the cycle low. However, they do not establish that conclusion, and a further material decline remains possible.
Another leg down is possible. Not only is this supported by both models, but this scenario could be driven by the macro backdrop. Despite correlations between Bitcoin, and the S&P 500 and the Nasdaq 100 continuing to drop significantly, reading now below 0.4, there are forces that could drain risk appetite. Leading indicators such as G10 excess liquidity have turned negative as lower-rated corporate bond high-yield spreads have continued to widen. Sovereign bonds should continue to see sell-offs, as US equities could soon see outflows catalysed by AI-related fears. Bitcoin would likely take collateral damage, albeit temporarily.
However, if the current trajectory persists, high global debt burdens and policy measures intended to ease financial conditions may affect demand for hard assets, including Bitcoin and gold. The direction, timing and magnitude of any such effect are uncertain, and these conditions may also produce significant losses across risk assets. This scenario analysis does not express a preference for any asset or recommend an allocation. Bitcoin, gold and broader cryptoassets have experienced differing outcomes during previous episodes. Those historical relationships have not been consistent and may not recur.
Appendix
Methodology for the CF Benchmarks Confirmation Framework
The main text describes what the evidence means for the current cycle. This appendix explains the three CFB research layers and their boundaries for readers who want the detail. They are complementary, not interchangeable.
Bitwise on-chain analysis retains its own stated source cutoffs.
Historical Core and Full Confirmation Engine availability
For historical context, the CFB Historical Core (Model B Tier B1) is a deliberately reduced historical core through 10 April 2023, while the CFB Full Confirmation Engine (Model A) begins on 11 April 2023. The earlier history is not a backfilled Full Broad Cap record.
Historical price-only check
The CFB Bitcoin Trend Check (H0) uses the CFB Bitcoin reference rate and compares price with its 20-, 50- and 100-day moving averages, plus the 50/100-day relationship. It needs several aligned observations to establish a trend, which is why it is best understood as a delayed structural confirmation layer.
Source: CFB Bitcoin Trend Check (H0) research through 26 August 2026. The shading is price-only structural context and is not a multi-asset breadth conclusion.
Directional means positive after risk-on and negative after risk-off. H0 issued 25 risk-on and 24 risk-off pivots overall, but 21 of 48 adjacent pivot intervals reversed within 30 days. These results reinforce why the CFB Bitcoin Trend Check is a confirmation layer, not a timely standalone bottom caller.
Research note
This paper separates supplied Bitwise on-chain analysis from CFB technical confirmation. Each party has prepared and is solely responsible for its own sections and conclusions. The CFB Full Confirmation Engine (Model A) is the full modern engine; the CFB Historical Core (Model B Tier B1) and CFB Bitcoin Trend Check (H0) are labelled research layers rather than backfilled live production history. Analyses are descriptive, rest on a limited number of historical cycles, use overlapping forward windows where reported, and do not constitute investment advice or a recommendation to buy, sell or hold any asset. Past performance and historical model outputs are not reliable indicators of future results. All forward-looking statements involve significant uncertainty; actual outcomes may differ materially. No statement in this document should be construed as a joint endorsement by both parties of the other party's analysis or conclusions.
Methodology for the On-chain Model
Our study examines 16 on-chain metrics, spanning miners, cost bases, valuation and investor behaviour, searching for high-quality measures of seller exhaustion and emergent buyer demand. Our goal was to find metrics for each side that reliably each ranked true bottoms above ordinary days. The study showed that, within each side, most metrics move together, so rather than stack several correlated measures we kept one clean representative for demand and one for supply.
On the supply side, the best available metric was the Long-Term Holder Spent Output Profit Ratio (LTH-SOPR). It shows whether long-term holders, those who have held for more than roughly 155 days, are selling at a profit or a loss. A reading below 1 means these holders are, on average, realising losses.
On the buy side, we selected the Accumulation Trend Score (ATS). It shows whether wallets, weighted by size, are growing their balances. On a scale of 0 to 1, a higher reading indicates accumulation.
Each input is first measured as a percentile of its own history: how extreme today's reading is against everything seen before. A logistic regression, a standard technique for turning several inputs into one figure, then combines the two into a single bottom score. The score is calibrated, meaning it is anchored to how often similar readings have sat near past bottoms, so it reads relative to previous cycles. It is best understood as a gauge of how this cycle compares with earlier ones, rather than a literal claim that a bottom will occur on a given day. A raw, uncalibrated version tends to overstate the odds, so we quote the calibrated one throughout.
With only four completed cycles to learn from, we defined a bottom as any day within 30 days of a cycle low rather than the single low (day) itself. During the drawn-out process of a bottom, the market dynamics on the exact low are hard to separate from those in the days around it. Treating that whole window as a bottom is therefore a reasonable way to enlarge the sample and improve the model's reliability, even though those days are highly correlated with one another.
Finally, every cycle was tested out of sample, meaning each cycle was scored by a model trained only on the other cycles. This removes look-ahead bias, so no factor could look good simply by being fitted to the same data it was measured on.
Notes
[1] These observations are based on four historical cycle-bottom periods and a small, highly correlated sample. They are not forecasts, do not represent the full range of possible outcomes and should not be used as the basis for an investment decision. Past performance and historical model outputs are not reliable indicators of future results.
Important information
This report is directed only at investment professionals within article 19 of the Financial Services and Markets Act 2000 (Financial Promotion) Order 2005. Any investment or investment activity to which it relates is available only to such persons. Persons who do not have professional experience in matters relating to investments should not rely on it. It must not be distributed to or relied upon by retail clients in the UK.
This report has been jointly prepared by Bitwise and CF Benchmarks. The on-chain model, its calibration and the bottom-probability analysis are the work of Bitwise; the technical-confirmation framework and the market-structure analysis are the work of CF Benchmarks. Each firm is responsible for its own contribution, and neither adopts or endorses the other’s analysis or conclusions. Each party is solely responsible for its own communication, distribution or publication of this report, or any part of or material derived from it. Any such communication, distribution or publication by one party is made on its own responsibility and is not authorised, adopted or endorsed by the other party, which accepts no responsibility or liability for it.
The report constitutes a marketing communication and is provided for informational purposes only. It is intended solely for professional investors and is not suitable for retail customers distribution and reliance.
The report is intended as research and market commentary, and nothing more. It is not investment, legal or tax advice, not a personal recommendation, and it is not an offer, invitation or solicitation to buy, sell or hold any digital asset, or any financial product. Readers should reach their own view and, where appropriate, take their own professional advice.
Bitwise and members of its group issue and market investment products that provide exposure to digital assets discussed in this report and therefore have a commercial interest in investor demand for this asset class. CF Benchmarks provides digital-asset benchmark and research services and may also have commercial interests connected with the subject matter of this report.
Bitcoin and other digital assets are volatile and speculative. Their value can fall as sharply as it can rise, and an investor may lose some or all of the amount invested. A constructive reading in this report is a view on the balance of evidence, not a promise about where the market goes next.
The figures in this report (including model scores, back-tested and simulated results, historical examples and any forward-looking commentary) rest on assumptions and on a limited set of past cycles. They show how signals have behaved before, not how they will behave in future, and are not a reliable indicator of future results.
All readings are as at the report date unless otherwise stated, and the underlying model updates over time. Market conditions, data and model outputs can and will change after that date, and this report is not updated to reflect them.
This report is provided to the recipient on a strictly confidential basis for the recipient's own use only. The recipient may not forward, distribute, reproduce or make available the report, in whole or in part, to any other person without the prior written consent of Bitwise and CF Benchmarks. Any onward distribution must include all applicable risk warnings and regulatory notices contained in this report.
Distribution control: recipients must not forward, publish, reproduce or otherwise make this report available to any retail client in the UK. Any extract must retain the professional-only audience restriction and the limitations relevant to the extracted material.
