How the Bitcoin risk model turns raw data into one number.
The model combines four families of signals into a single risk score from 0 to 10. Each family — and exactly how the scoring works — is described below. The specific indicators inside each family, and their precise weighting and calibration, are proprietary.
Prefer the short version? Start with the questions a skeptic should ask.
From raw data to one number
Normalize
Each factor is converted to a 0–100 scale by ranking it against its own trailing 4-year window (a full halving cycle), not against all-time extremes. 0 means low relative to recent history; 100 means high.
Weight
Factors are combined as a weighted average, each family weighted by how reliably it has flagged past cycle turns.
Score
The weighted result is divided by 10, giving a composite risk score from 0 (deep value conditions) to 10 (historically dangerous conditions).
Each reading also blends where its level sits within the trailing 4-year window with how fast it has been moving there recently, so a signal accelerating toward an extreme registers before its level alone would catch up — this is what let the model flag the leverage-driven November 2021 top, which a level-only reading had missed.
If an external data source is unavailable, that factor defaults to a neutral 50 rather than silently skewing the score. Weights are normalized so the composite always reflects the full set.
Crypto model (Bitcoin)
Each family is weighted by how reliably it has discriminated cycle tops from bottoms in historical testing — on-chain and valuation carry most of the weight, and macro sits at the bottom as context rather than a timing signal. The specific indicators within each family are part of the proprietary model.
On-chain & Valuation
How stretched Bitcoin's price is relative to measures of what holders actually paid and what the network is realizing in value. The largest single family — historically the most reliable discriminator between cycle tops and bottoms.
When holders are sitting on large collective gains and the market is pricing coins far above what was actually paid for them, history says more supply tends to come to market. This family carries the most independent information in the model.
Cycle & Price Extension
How far price has run above its own longer-term trend, and where today's run-up sits against Bitcoin's shrinking cycle-over-cycle pattern of diminishing returns.
Price running far above its own recent baseline, with miner and network profitability running hot alongside it, has preceded most major corrections. Several of these measure closely related things, so weight concentrates on the most reliable members.
Momentum & Sentiment
What the crowd is feeling and how price is behaving. Useful confirmation, secondary to the two valuation families above.
Extreme greed and overheated momentum have historically been better moments to reduce than to add, and volatility expansions mark unstable regimes. Context and confirmation rather than a standalone timing call.
Supply & Macro
Structural supply dynamics and the broader liquidity backdrop. Weak cycle-timers on their own, so they are kept at low weight — present for context, not relied on to call turns.
Supply-side conditions and the macro backdrop shape risk appetite for everything, but their link to crypto cycle timing is loose and lagging — so they inform the score without driving it.
How the score becomes a weekly signal
The score updates as data arrives, and each Sunday it maps to one of three zones. The signal is always a risk statement about position sizing — never a price prediction.
Undervalued
Historically favorable conditions. The strategy scales its weekly buy size up as risk falls — buying most aggressively below 1.
Neutral
Neutral territory. No buying, no selling — the strategy simply waits. The widest band, so most weeks land here.
Risky
Historically dangerous conditions. The strategy trims a growing fraction of the position as risk climbs through the higher tiers.
Two ways to act on the signal
The same weekly score feeds two published, rules-based strategies. Both are mechanical — no discretion, no hand-picked entries — and both are tracked on the public record.
Anchor Accumulate
Risk-based buy & holdScales the size of each weekly Bitcoin buy up as risk falls through the accumulate zone, and pauses buying once risk leaves it. It never sells. For the holder who is not going to sell either way, but would rather not buy the same amount at $16k and at $69k. It therefore asks for more capital in low-risk stretches and less in high-risk ones — which is the rule working as intended, and why its results are always published with the capital deployed beside them.
Anchor Bridge
Active · buy engine + rules-based trimRuns the same accumulate buy engine, and additionally trims a fraction of the position into the S&P 500 when risk is elevated. Each trimmed "lot" rotates back into Bitcoin once price falls a set distance below where it was trimmed — a rules-based bridge between the two assets, not a market-timing call. For the investor willing to trade and file the paperwork: its case rests on a shallower drawdown rather than a bigger headline number, and it holds up after the tax on every sale.
See both strategies applied week by week on the public track record.
The methodology above describes the model we run today: Bitcoin. The same 0–10 scoring approach — normalize against recent history, weight by reliability, map to accumulate / hold / reduce — is designed to extend to other assets like equities and precious metals, each with its own independent factor set. Those models aren't live yet. When one launches, its factors, weights, and limitations will be documented here the same way — we won't point you at a score we haven't stood behind.
Limitations, stated plainly
A model is only worth trusting if it's equally clear about what it can't do. Here's the whole list, unsoftened.
There are two track records here, and only one of them is verifiable.
The first is informal: the founder has traded this approach live, by hand, on TradingView since December 2021 — a real forward test in the sense that money was at risk and the calls were made without knowing what came next, but a private one. There are no timestamped public entries for it, so you have no way to check it and we don't ask you to treat it as evidence. It explains where the indicators and weighting came from; it doesn't prove they work.
The second is public and verifiable: from 2026-07-09, the numeric weights in this codebase were locked and every weekly call has been published to an append-only, hash-chained live track record. That is the one you should judge us on, and it is deliberately short — it started when the freeze started.
The codified weights were frozen on 2026-07-09.
That's the date the exact numeric weights and zone bands in this codebase were locked, marking the start of the forward (out-of-sample) record for this specific implementation — see the performance page for details. It is a narrower claim than the founder's trading history above: it covers only when this software's numbers were fixed, not how long the underlying approach has been used. Everything the backtest shows for dates before it was scored by weights chosen with knowledge of those years — that is what makes it in-sample. Any future change to these weights will be justified independently of how it would have scored on past cycles, not chosen by looking at backtest results, and will be published here.
Past regimes may not repeat.
Every signal's usefulness rests on historical patterns (on-chain valuation cycles, network economics, macro tightening). A structurally new market can break any of them.
This is decision support, not advice.
The model quantifies conditions; it doesn't know your situation, and it will be wrong on individual calls. The bet is that a disciplined process beats narrative and vibes over many decisions.
Anchor Accumulate is not a strict upgrade over weekly DCA — it’s a timing trade, not a crash-protection trade.
The strategy varies how much you put in: it buys harder when the model reads risk as low, and pauses when it doesn't. So it is not running the same amount of money as plain weekly DCA, and the returns you see are per dollar deployed. Across 2018–present it deployed $87,000 against weekly DCA's $44,600 and returned more per dollar. In the 2023–2025 rally the reverse applies: risk stayed elevated, so it deployed only $3,000 against weekly DCA's $15,600 — a better percentage on far less money, ending in a materially smaller position. If you expect an uninterrupted bull market, this rule will leave you behind. And because it never sells, its drawdown stays close to weekly DCA's: it times entries better, it does not avoid the fall. See the full performance breakdown for every regime, with the capital deployed alongside each result. It is not free alpha in every regime, and treating it as one would be dishonest. We wrote up exactly what the backtest does and doesn't prove in why we call our backtests illustrative, not validated.
Methodology changes will be published.
If a weight or factor changes, the change and the reasoning will be documented — the process is only trustworthy if you can watch it over time.
See it live, or pressure-test it yourself.
The model updates weekly. The FAQ answers the questions a skeptic should ask first.
Alphabit publishes the outputs of transparent, rules-based models applied to hypothetical reference portfolios, for research and education. This is not personalized investment advice, not a recommendation, and not a prediction of future results. Past patterns may not repeat. See the full disclosure.