# Learning-loop review — 2026-07-30

52 closed trades reviewed, 44 currently open. Proposal only — nothing here is applied automatically (CLAUDE.md §4: the AI review proposes, dan disposes).

# AI Review — Paper Trading Journal

## 1. Is overall expectancy holding?

Overall expectancy is **0.13% per trade** across 52 trades, total P&L **-$13**. That's essentially flat — expectancy is indistinguishable from zero once you account for the small sample. The aggregate n (52) technically clears the ~20-trade floor, but a mean of +0.13% on a P&L curve that nets to -$13 isn't "holding an edge," it's noise oscillating around breakeven. No conclusion of a real edge can be drawn yet.

## 2. Which bucket is the biggest drag?

Two qualifying buckets (n≥20) point at the same underlying problem:

- **orb-v1** (by strategy): n=33, expectancy **-0.22%**, profit factor **0.70**, total P&L **-$176**.
- **stop_loss** (by exit reason): n=22, win rate **0%**, expectancy **-1.82%**, total P&L **-$698**.

(The "unknown" bucket recurring at n=32 in several cuts is just trades missing snapshot fields — not a real, tunable bucket, so it's excluded.)

`orb-v1` is the strategy generating the bulk of these stop-outs (33 trades, 0.2d avg hold, PF 0.70) — the stop_loss cut is effectively showing us *why* orb-v1 is dragging the book.

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

This reads as an **entry/stop problem, not an exit problem**, based on the aggregate MAE/MFE stats we have (bucket-specific MAE/MFE for orb-v1 or stop_loss alone isn't broken out in the report, so this is inferred from the overall numbers as a proxy):

- **MAE of losers: -1.55%** — moderate, not a blowout. Combined with the stop_loss bucket's 0% win rate and short average hold (0.7d for stop_loss overall, 0.2d for orb-v1), trades are getting stopped quickly on what looks like ordinary noise rather than a thesis genuinely failing.
- **MFE of losers: 0.53%** — low. Per the doc's own diagnostic, an exit/take-profit problem shows up as losers with *high* MFE (traded worked, we gave it back). 0.53% is not "materially positive," so this is **not** a give-back/exit pattern.
- That combination (modest MAE, low MFE, fast stop-outs) matches the doc's "stop tighter than the natural noise of the name" / weak-entry-confirmation pattern rather than a target/trailing-exit issue.
- Regime doesn't look like the driver either: SPY>MA200 bucket (n=20) actually shows *positive* expectancy (0.22%, PF 1.43) — the drag isn't concentrated in a bad regime, so a regime gate isn't indicated by the current data.

## 4. Proposed parameter change

**orb-v1: `volumeConfirmMult` 2 → 3**

Reasoning: the fast, low-MAE, low-MFE stop-outs suggest orb-v1 is entering on breakouts that aren't well-confirmed and getting shaken out almost immediately (0.2d avg hold, PF 0.70, 0% win rate on the stop_loss subset). Raising the volume-confirmation multiplier demands a stronger breakout before entry, directly targeting the "demand more confirmation" fix the doc prescribes for this MAE pattern, without touching sizing, targets, or exits. This is the single change to make; measure the next batch of orb-v1 closed trades (post-change only, per rule 4) before touching anything else.

## 5. Notable open-position color (not evidence for #4)

- **qsr has a very large number of simultaneously open tranches** (AEM, HDB, TXN, UMC, TRGP, TT, CARR, APP, NXPI — many duplicated across tranches), far more concurrent exposure than the 8 closed qsr trades used to build its scoreboard. Worth watching for concentration risk given `QSR_RISK.maxOpenPositions: 100` effectively allows unlimited stacking.
- **TT positions are underwater at -3.30%** (4 tranches) and **TXN at -3.96%** — both still open, no calendar time-stop, so they can keep drifting.
- **UMC is up +8.88%** and CARR/TRGP modestly positive — nothing to act on, just noting the qsr book currently skews profitable on unrealized marks.
- **BTC/USD and ETH/USD (crypto-trend-v1)** have been open **9.2 days with no time-stop** — this strategy isn't in the closed-trade evidence base at all yet, so it's an unknown quantity; flag for monitoring only.
- **orb-v1's 3 open positions (SMCI, MARA, ANET)** all carry a 1-day time-stop and should resolve imminently, feeding back into the same bucket flagged above.