Personal edge is the set of conditions under which the trader's BEST trades cluster differently from the WORST. Ranking is by R-multiple (falling back to net P&L when R isn't computable). The detector takes the top 10 % vs bottom 10 % and counts how often each feature value appears in each decile. Features compared:
- Weekday (Mon-Sun)
- Hour-of-day bucket (00-06 / 06-12 / 12-18 / 18-24 UTC)
- Hold-time bucket (<30min / 30min-2h / 2-8h / 8-24h / >1d)
- Symbol
- Market regime (bull/bear/sideways via TRA-226)
- Setup tags (user-defined)
- Mistake tags (user-defined)
- Psychology tags (user-defined)
For each (dimension, value) pair the detector computes a lift: how much more often that value appears in the top decile vs the bottom decile, normalised so the deciles' size doesn't distort. Lift ≥ 2 = best-features list. Lift ≤ 0.5 = worst-features list.
Across 80 scored trades, decile size 8: best-side surfaces (Weekday=Tue, top 5/bottom 1, lift 4.6), (Symbol=NVDA, top 6/bottom 1, lift 5.5), (Regime=bull, top 7/bottom 2, lift 3.0). Worst-side surfaces (Hour=18-24 UTC, top 1/bottom 6, lift 0.18), (Mistake=fomo, top 0/bottom 5, lift 0.10).
Why this is structurally different from declaring criteria up front. A trader who declares 'I trade Tue NVDA bull-regime' is committing to a hypothesis. The Onyx-Engine has the data; it tells you what your hypothesis ACTUALLY is, derived from where you've been making and losing money. The fingerprint is more honest than the self-narrative.
Reading lift values. Lift = 1 means the value is equally common in both deciles (no signal). Lift = 2 means twice as common in the top — surfaceable signal. Lift = 5+ means the value is overwhelmingly concentrated in the top decile — a real, actionable edge condition. The smoothing nudge (+0.5/decile_size on both sides) prevents divide-by-zero from inflating lift on rare features without distorting moderate-count comparisons.
Two interventions:
- Best-features list = protect these conditions. The simplest journal rule: 'When at least 2 of my best-feature conditions align, size up by 25 %.' Don't change the strategy; concentrate capital where the data says it works.
- Worst-features list = either build a separate playbook for them, or sit them out. Trying to grind through with the same setup is the most common path to plateauing.
Limitations.
- Decile-based, not statistical. A formal test (chi-square, Fisher's exact) would yield p-values. The detector uses lift instead because lift is more interpretable for traders ('twice as common in my best trades' beats 'p < 0.05').
- Tag dependence. The setup/mistake/psychology tag dimensions only surface signal if you're tagging consistently.
- 30-trade minimum. Below this the deciles are too small for the math.
Tier: Pro. Wave 7 (Personal Edge Discovery) — first detector in the final wave. Pairs with TRA-235 (Top mistakes — what costs you most) and TRA-226 (Regime performance — bull/bear/sideways breakdown).
How to read the card: two side-by-side lists. Best-trade conditions (left, green header) and worst-trade conditions (right, red header). Each row shows the dimension label, the value, the per-decile counts (top ↔ bottom), and the lift. Re-look quarterly — feature distributions stabilise over months, not days.
Tier: Pro.