# Learning-loop review — 2026-07-23

18 closed trades reviewed, 13 currently open. Proposal only — nothing here is applied automatically (CLAUDE.md §4: the AI review proposes, dan disposes).

# Journal Review

## 1. Is overall expectancy holding?

Overall expectancy is **-0.41%** per trade across **18 closed trades**, with a profit factor of 0.43 and a Wilson 95% lower bound on win rate of just **24.6%** (raw win rate 44.4%). This is not yet a meaningful signal — the sample is well below the ~20-trade minimum, so "expectancy is negative" cannot be distinguished from noise at this point. No conclusion should be drawn from this number yet.

## 2. Which bucket is the biggest drag?

**No bucket in this report reaches the ~20-trade minimum.** The largest buckets are:
- Overall: 18 trades
- orb-v1 (by strategy): 15 trades
- stop_loss (by exit reason): 9 trades
- orb-v1 · ANET (by strategy+symbol): 4 trades

All are below the 20-trade threshold specified in the anti-self-deception rules. Per the rule ("don't act on any bucket with fewer than ~20 trades"), **none of these buckets can be treated as a result — only as hypotheses.** I will not name a "biggest drag" bucket as an actionable finding.

## 3. Entry problem, exit problem, or regime problem?

Since no bucket clears the trade-count bar, any MAE/MFE-based diagnosis below is a **hypothesis, not a finding** (rule 5), offered only for what to watch, not to act on:

- **Overall MAE/MFE**: winners' avg MAE is -0.51%, losers' avg MAE is -1.51%, and avg MFE of losers is **0.46%**. A positive `avgMfeLosersPct` is the pattern the doc flags as an exit problem (trades moved favorably before reversing into losses) — but at n=18 (and stop_loss n=9), this is far too thin to confirm.
- **Exit reason split**: stop_loss trades (n=9) show 0% win rate and -1.77% expectancy, versus eod_flatten (n=8) at +0.91% expectancy. This *suggests* stops are doing the damage on orb-v1 specifically, but n=9 is nowhere near actionable.
- Regime data is entirely `unknown` (no SPY-regime tagging present in the snapshot data for these trades), so a regime read is impossible with current data — this is a data-completeness gap, not a finding either way.

**Bottom line: cannot confidently attribute the drag to entries, exits, or regime yet** — the MFE-of-losers signal is worth watching but is not evidence at this sample size.

## 4. Proposed parameter change

**NO CHANGE.** No bucket has reached the ~20-trade minimum, so per the anti-self-deception rules I will not propose a parameter tweak — doing so now would be tuning to noise, and with orb-v1 already at n=15, one more change would also make future attribution (rule 2) harder to read cleanly.

**What to wait for:** let orb-v1 (currently 15 closed trades, all backtest-only/not live-wired anyway) and the overall book accumulate to ≥20 closed trades before revisiting. In particular, watch whether the stop_loss exit-reason bucket (currently n=9, -1.77% expectancy) grows toward 20 trades — if it does and still shows 0% win rate with low MAE on losers, that would be the first legitimate candidate for a stop-width or ATR-scaling change (`stopAtrMult`, currently 2 for both swing-dip-v1 and qsr).

## 5. Notable color on open positions (not evidence for #4)

- **qsr has 11 open tranches across 6 symbols**, all only ~1.3 days old — too fresh to read anything into yet, but worth noting the book is currently skewed toward qsr ex