An indicator describes market activity. An edge is a condition that has shown positive expectancy after costs, over a sufficient sample. Everything the Order Flow engine scores is registered here as a claim to be tested — not asserted.
Every figure below carries its own sample size and an explicit evidence label. A hypothesis with too few matched trades reports INSUFFICIENT_SAMPLE — never "weak" and never "rejected", because calling a thin sample rejected is the most common way a research process talks itself out of a real effect. Results come from the engine's own paper ledger, which books fees and slippage on every leg.
1 of these cannot be answered from the assessments recorded so far — every matched trade lacked the features the hypothesis reads, because those rows predate them. That is a gap in coverage, not a weak result, and waiting alone will not close it for the rows already written.
Computed 2026-09-11T07:17:35Z over 41079 journalled assessments and 215 resolved paper trades.
A score of 92 is not a 92% chance of winning. It is a composite of evidence. Whether a 90 actually beats a 70 is measured here, against this engine's own history.
INSUFFICIENT — fewer than two bands have a reportable sample; the score is not yet calibrated and must not be presented as a probability
| Score band | Trades | Win rate | 95% interval | Mean R | Expectancy | Evidence |
|---|---|---|---|---|---|---|
| <50 | 1 | not reportable | — | -0.052 | +7.730 | INSUFFICIENT |
| 50-59 | 0 | not reportable | — | — | — | INSUFFICIENT |
| 60-69 | 0 | not reportable | — | — | — | INSUFFICIENT |
| 70-79 | 0 | not reportable | — | — | — | INSUFFICIENT |
| 80-89 | 0 | not reportable | — | — | — | INSUFFICIENT |
| 90-100 | 0 | not reportable | — | — | — | INSUFFICIENT |
A band with fewer than 20 matched trades shows no win rate at all. A rate quoted off three trades is an anecdote wearing a percent sign.
Passive buyers absorbing aggressive selling within half an ATR of the value-area low should produce better expectancy than the same engine's other longs.
0 treatment / 1 control trades — below the 20 minimum. 104 of 105 matched trade(s) were unanswerable (the row lacked the features), so the shortfall is partly coverage rather than volume.
Features: absorption_type, price, val, atr_pct
Tests the brief's own indicator-versus-edge example: a bullish CVD divergence on its own against the same divergence corroborated by buyer absorption.
0 treatment / 3 control trades — below the 20 minimum. 99 of 105 matched trade(s) were unanswerable (the row lacked the features), so the shortfall is partly coverage rather than volume.
Features: cvd.divergence, absorption_type
Whether stacking location conditions on the value pillar adds expectancy over holding above VWAP by itself.
0 treatment / 105 control trades — below the 20 minimum; no conclusion is drawn
Features: price, vwap, val, absorption_type
Acceptance beyond a low-volume node, rather than a poke into one, should mark an efficient auction higher.
1 treatment / 0 control trades — below the 20 minimum. 104 of 105 matched trade(s) were unanswerable (the row lacked the features), so the shortfall is partly coverage rather than volume.
Features: lvn, lvn_accepted, price
Heavy sell-side aggression below a level, followed by a reclaim of it, should precede continuation more often than the engine's other longs.
0 treatment / 1 control trades — below the 20 minimum. 104 of 105 matched trade(s) were unanswerable (the row lacked the features), so the shortfall is partly coverage rather than volume.
Features: delta_bars, ib_low, val, pdl
Whether the confirmation layer's edge, if any, is concentrated in participating markets.
0 treatment / 1 control trades — below the 20 minimum. 104 of 105 matched trade(s) were unanswerable (the row lacked the features), so the shortfall is partly coverage rather than volume.
Features: rvol
UNTESTABLE TODAY. No options/GEX feed is wired to this repo, so every row answers None and the hypothesis reports INSUFFICIENT_SAMPLE by construction. Registered rather than omitted so the gap is visible and sized.
NOT YET ANSWERABLE — 105 trade(s) matched a journal row and every one lacked the features this hypothesis reads, so none could be assigned to either group. That is a blind instrument, not a thin sample: these rows predate the features and waiting will not make them countable. Needs 20 of each group from rows journalled after them.
Features: gex
Acceptance or reclaim of the initial-balance high, with the confirmation layer agreeing.
0 treatment / 101 control trades — below the 20 minimum. 4 of 105 matched trade(s) were unanswerable (the row lacked the features), so the shortfall is partly coverage rather than volume.
Features: ib_state, confirmation_score
Removing the single best symbol turns the result negative — the honest test of whether an edge is broad or is one lucky name.
