Current Regime
Markup
confidence 86%markdown
7%
accumulation
2%
markup
86%
distribution
5%
What it means
Which of the four market 'states' the model thinks we're in: markup (uptrend), distribution (topping), markdown (downtrend), or accumulation (basing). Each bar is the probability the model assigns to that state today.
Why it matters
The position-sizing rule scales exposure by regime: full size in markup, half in accumulation/distribution, zero in markdown. So the regime is the single biggest lever on the trade signal.
P(Up · 7d)
55%
calibrated, ECE < 0.0530d horizon: 58%What it means
Probability the model assigns to BTC closing higher than today, seven days from now. 50% is a coin flip; the model only rarely gets far from that.
Why it matters
This is the headline directional call. Combined with volatility and regime, it drives the Kelly-style position target you see in the gauge below.
Volatility · 7d
4%
7d realized σ, HAR-RV + GBM30d realized: 11%What it means
Expected size of BTC's cumulative move over the next 7 days, as a standard deviation of returns (not annualized). Higher = wider expected price swings. The 30d figure is naturally larger because it spans more days.
Why it matters
High vol shrinks the recommended position size (more uncertainty per unit return). Low vol expands it and often precedes large directional moves.
Tail Risk · 30d
0%
P(drawdown > 20%)head not availableWhat it means
Probability of a peak-to-trough drawdown of more than 20% over the next 30 days. Built from regime, directional probability, and macro stress (VIX, DXY, real yields).
Why it matters
Sits above the headline as a 'caution check'. Even when P(up) looks fine, an elevated tail-risk number warns that left-tail outcomes are getting more likely — informing how conservative to size.
Price · 90d History + Forecast Cone
Quantile Forecast
Spot history
Median forecast (P50)
Nominal 80% interval (P10–P90)
What it means
The white line is realised BTC/USD. The accent dashed line is the model's median price forecast; the shaded cone is the model's P10–P90 quantile interval (nominal 80%). On backtest its empirical coverage runs below 80% — conformal (CQR) calibration is a pending follow-on — so treat it as an indicative, not guaranteed, band. Only the horizon endpoints (7d/30d) are model outputs; intermediate-day widths are interpolated.
Why it matters
A narrow cone means the model expects a tight outcome; a wide cone means high uncertainty even if the median is up. The cone, not the median line, is the honest read on what the model expects.
Position Target
+0.04/ ±1.00
Neutral exposure−1.0 short0+1.0 long
half-kelly signal0.05
Half-Kelly criterion: 0.5 × (2·P(up) − 1). The 0.5 is the spec-mandated safety factor. Positive favours long; negative favours short. This is the value that actually feeds the position target.vol normalization1.00
Divides the half-Kelly signal by realised vol so a high-vol day shrinks position size. Floored at 1.0 so calm days do not over-lever.regime multiplier0.89
Scales exposure by market regime — markup ×1.0, accumulation / distribution ×0.5, markdown ×0.0. Mixture-aware so the gauge moves smoothly when regime probabilities shift.What it means
A model-suggested portfolio tilt between fully short (−1) and fully long (+1), built from three components: the half-Kelly directional signal, the volatility forecast (size shrinker), and the regime (multiplier).
Why it matters
This is what the model would 'do' if it were trading. The number itself is not advice — it is the educational summary of the three signals above. The headline = (half-kelly signal ÷ vol normalization) × regime multiplier, clipped to ±1, so the three knobs reconstruct exactly how it was built — auditable, not opaque.
On-Chain Feature Z-Scores · 365d Window
Signal Heatmap
Every cell is one input feature, z-scored against its own 365-day history. Red / above zero means the metric is unusually high today; blue / below zero means unusually low. Click any cell for the full per-feature page with definition, 90-day history, and how it affects BTC price.
Whale %
-0.89σ
Share of tracked address-resolvable balances held by addresses with >=1,000 BTC.Tx Volume
-0.73σ
Total on-chain BTC moved today across all confirmed transactions.Supply in Profit
+1.85σ
Share of circulating supply whose last on-chain move happened at a lower price than today.SOPR
+2.19σ
Spent Output Profit Ratio. >1 means coins moving in profit; sustained <1 is a capitulation tell.Realized Vol 30d
-0.61σ
Rolling 30-day standard deviation of daily log-returns. Drives the position-sizing vol normaliser.Puell Multiple
-0.19σ
Daily USD issuance value vs its 365d moving average. Macro miner-revenue cycle indicator.MVRV-Z
+1.36σ
MVRV ratio (market value ÷ realized value). The heatmap σ is this ratio normalized against its own 365-day history — high σ = unusually rich vs the past year. (Distinct from the textbook MVRV-Z; the raw value on this page is the underlying ratio.)Miner Outflow
-0.65σ
Net BTC leaving coinbase-recipient (miner) addresses. Distribution signal.Log Return
-0.23σ
Daily log price change (BTC/USD). Mostly noise; included to give the model an autoregressive baseline.Hashrate
+0.02σ
Daily network hashrate derived from block difficulty. Long-horizon security + cost-of-attack floor.Fee p50
-0.54σ
Median per-byte fee rate in confirmed blocks. Tracks mempool congestion + demand.Exch. Net Flow
-0.01σ
Net BTC flowing into labelled exchange addresses. Positive = potential sell pressure.Dormancy
-0.39σ
Average age of coins moved today, weighted by volume. Spikes when long-dormant coins wake up.CDD
-0.54σ
Coin Days Destroyed. Sum of (coin age x value) on today's moves. Tracks long-term-holder activity.Active Addr.
-1.29σ
Unique addresses that sent or received at least one tx today. Network adoption proxy.|z| > 2 historically marks extremes · color intensity scales with magnitude
cool · below avg
hot · above avg
What it means
A z-score reports how far today's value sits from its own 365-day average, measured in standard deviations. Zero is dead-average; ±1 is mildly unusual; ±2 is a historical extreme.
Why it matters
These features are the model's input. The heatmap shows the current 'shape' of the BTC market in one glance — which inputs are hot, which are cool, and therefore which signals are pulling the prediction. Clicking through opens the full definition + history + price-effect explainer.
Model Health
Directional ECE
0.020target < 0.05Calibration of the directional head: the gap between predicted probabilities and reality. Lower = when it says "70% up", BTC actually rose ~70% of the time.OOF Log-Loss
0.691target < 0.69How well the directional model scores on held-out data. The 0.69 (ln 2) ceiling is the coin-flip baseline; below it means the model beats random.Feature Drift
1 / 15|z| > 2 features todayCount of inputs that look unusually extreme vs their history. A handful is normal; a third or more is the cue to retrain.Coverage 80%
0.81target [0.75, 0.85]Share of past days the realised price landed inside the 80% forecast cone. Should sit near 0.80 — far off means the cone is mis-sized.What it means
Four quick self-checks on the model's quality, measured on held-out (never-trained-on) data. For the calibration, log-loss, and coverage gates, the accent color means the check passed and red means it failed. Feature drift is a monitoring signal, not a §7.4 gate: accent = healthy, amber = drift building (watch), red = time to retrain.
Why it matters
These tell you whether to trust the headline forecasts above. If calibration or log-loss go red, the probabilities are unreliable; if drift climbs, the inputs have shifted away from what the model learned on. Full breakdown lives on the model-card page.
SHAP Contributions · 7d Directional Forecast
Why
Each bar shows how much that feature pushed the directional model's score (log-odds) away from its baseline — margin contributions, not probability points. Bearish = pushing the score down.
whale_pct
-0.093
fee_rate_50pct_z
-0.038
supply_profit_pct
-0.029
mvrv_z
-0.022
active_addresses
-0.020
puell_multiple
-0.019
Plain English
The strongest pull on today's directional read comes from whale_pct — pulling the model toward a bearish read. Combine with the regime, vol forecast and macro context above before interpreting.What it means
TreeSHAP attribution scores in log-odds (margin) space. For each feature, this is its signed contribution to the directional head's raw score — one of several inputs (alongside the vol and quantile heads) that get blended and isotonic-calibrated into the headline P(up). So these rank what drove the directional signal, not an exact split of the final probability.
Why it matters
Turns the model from a black box into an audit trail: if a feature is pulling the directional call hard one way, you can click through to its heatmap cell to see why.