We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Track record / methodology
/ conditional_probability
Typical horizon: INTRADAY • Season 1 — all time
What it does
Fuses the calendar and volatility-regime edges into one joint conditional distribution: the body distribution of, e.g., Friday 14:00 UTC candles that follow a big move has its own shape. Per open range bucket it scores that joint P(settle in bucket) against the ask and buys the largest edge, falling back through (hour,regime) -> (dow) -> (hour) -> global marginals when a joint cell is too thin to trust.
Key parameters
behavior-shaping settings only| min_edge | 10 |
| symbols | ["BTC","ETH","SOL","XRP","DOGE"] |
Live trading began 2026-08-30; paper record retained above. This strategy has not traded real money; its record below is paper (simulated).
Record (fee-inclusive, this season)
Calibration
predicted vs realized win rate, by confidence bucket| bucket 2 | 173 signals | p̂ 0.18 → 0.04 |
| bucket 3 | 531 signals | p̂ 0.25 → 0.11 |
| bucket 4 | 376 signals | p̂ 0.35 → 0.19 |
| bucket 5 | 213 signals | p̂ 0.45 → 0.33 |
| bucket 6 | 91 signals | p̂ 0.55 → 0.4 |
| bucket 7 | 55 signals | p̂ 0.65 → 0.55 |
| bucket 8 | 47 signals | p̂ 0.75 → 0.6 |
| bucket 9 | 46 signals | p̂ 0.85 → 0.72 |
| bucket 10 | 17 signals | p̂ 0.94 → 0.76 |
Fill realism: maker markout
mid vs fill price after a paper maker fill, last 14d| horizon | marked | adverse | avg markout | median |
|---|---|---|---|---|
| 1 min | 92 | 69.6% | -4.59c | -3.5c |
| 5 min | 65 | 61.5% | -2.65c | -3.0c |
| 15 min | 24 | 54.2% | -2.56c | -1.75c |
| 60 min | 0 | — | — | — |
The mid keeps moving against maker fills — 2.56c adverse at 15 min — so the paper maker fills carry an adverse-selection tax the quoted edge must clear. Fills are the paper maker simulator's, not exchange truth; negative = the mid moved against the fill.
Data quality policy
Every known gap in the market data behind these figures is disclosed publicly, with the dates, what was missing, which figures it may have affected and which it did not. Position and P&L records are never rewritten to paper over a gap; model-derived figures are recomputed once the input is restored, and gaps that cannot be recovered stay marked as such. The full list is on the track record under data quality notes.
This strategy trades off the public tape. 2026-08-30 → 2026-08-30: Live trading began 2026-08-30 (live) — it saw little or no flow in that window, so its signal count and record for those days reflect missing input, not a decision to sit out.