⚠️ SUPERSEDED HISTORICAL SNAPSHOT (Mar 2026, 87-trade cohort)

The numbers below are a frozen snapshot from the 87-trade cohort (Mar 2026). Live state: ~117 closed trades, WR ~37%, net +$212. Canonical sources: docs/STRATEGY_PLAYBOOK.md §2 + docs/FINDINGS.md.

Financial Projections & Assumptions (Revised After 87 Trades)

Date: March 9, 2026 Data source: 87 closed trades (84 clean, 3 bug-affected excluded) from Jan 7 — Mar 7, 2026 Purpose: Single reference document capturing all strategy analysis findings and implemented changes at the 87-trade milestone. Next re-evaluation at 150 trades.

⚠️ Apr 25, 2026 — partially superseded

This document captured the state at trade #87 (Mar 9, 2026). It remains useful as a historical snapshot of the Mar 9 implementation push (Finding #202) and the projections drawn from that data. For current numbers and the ranked queue of next changes, see:

• docs/STRATEGY_PLAYBOOK.md §1 (verdict at the 87-trade snapshot) and §2 (combined Action Queue).
• Dashboard 🎯 Action Queue tab (live, recomputed on every report run).
• docs/FINDINGS.md Finding #228 (Apr 25 — v3 Finding A retraction).

The Improvements / next-steps sections below are stale — many of the 6 recommendations from STRATEGY_IMPROVEMENTS.md (which is now a redirect) have been shipped, retracted, or evolved. Do not treat the projections in this file as the operational forward plan.

1. Strategy Overview (Post-Revision)

Changes Implemented (Finding #202, Mar 9, 2026)

Signal Acceptance Rules (Mar 2026 snapshot)

2. Performance Summary (84 Clean Trades)

3. Direction x Timing Analysis

Key Findings

• LONG AMC structural disadvantage: 13 of 23 AMC stops trigger at 16:01-16:17 ET — immediately after earnings release, not from pre-earnings noise. Only 17% survive to scheduled exit.
• When LONG AMC survives: 100% winners (+$184 on 6 trades, EV +$30.70). The edge exists but survival rate is too low.
• SHORT AMC best ROI: +3.75% ROI per dollar deployed. Large wins when miss predictions are correct.
• BMO vs AMC cohort backtest (Q3 2025, winsorized): BMO +1.23% EV (significant), AMC ~0% EV (no edge).

4. Conviction Analysis

Performance by Conviction Band (10% buckets)

Performance by Conviction Band (5% buckets)

Q3 2025 Backtest Validation (5% buckets)

Key Finding

• 90%+ is profitable in Q3 2025 backtest but fails catastrophically in live 2026 trading (78% stop rate)
• The divergence is caused by higher pre-earnings volatility in 2026 and tighter stops triggering more often on high-conviction trades where the stock barely moves
• 75-80% dip in live data is statistically insignificant (14 trades, CIs overlap with adjacent bands)
• The 70-80% band is the best in both datasets

5. Stop Loss Analysis

Current vs Scenario Comparison (Using Actual Polygon Exit Prices)

Stop Reversal Analysis (Polygon Actual Prices)

Of 41 stopped trades with correct predictions (prediction matched earnings outcome): - 10 would have been winners at scheduled exit (+$113 total) - 19 would have been losers at scheduled exit (-$473 total) - Holding to exit would have been WORSE (-$360 vs -$399 from stops) - "Correct prediction" does NOT guarantee favorable stock movement

Rationale for 3%/4%/5%

The moderate scenario reduces stops from 35 to 13 while improving total P&L by +$115. Going wider (4%/5%/6%) shows diminishing returns as the remaining 2 stop triggers lose more per trade.

6. Prediction Accuracy

Volume-Prediction Correlation (34 trades with data)

Note: the $1,847 figures below are the median-split analysis buckets for this cohort, not a trading threshold. The live volume gate is $1,000 (Finding #311). - Low volume (<$1,847): 62.5% accuracy - High volume (>=$1,847): 37.5% accuracy - Higher volume correlates with lower accuracy (-12% difference) - Sample too small for statistical significance

7. Market Regime Performance

8. Backtest Scenario Comparison (Q3 2025, 370 Markets)

Key Backtest Findings

• BMO-only (scenario 15): PF 1.90, ROI +14.0% — highest ROI of any day-close single-filter scenario
• Combined filter (scenario 16): PF 2.17, EV +2.14% — highest profit factor and EV per trade
• 1h/2h exit scenarios show inflated ROI due to AMC outlier (+902% return) — use winsorized cohort data instead

9. BMO/AMC Cohort Timing (Winsorized, Q3 2025)

BMO Cohort (151 trades)

AMC Cohort (122 trades)

Current exit timing (BMO ~1h after open, AMC ~day close) is near-optimal.

10. Projected vs Actual Performance

11. Implemented Recommendations (Mar 9, 2026)

#1 CRITICAL — Skip 90%+ Conviction (IMPLEMENTED)

• Data: 29 live trades at 90%+ conviction: 14.3% WR, -$116 P&L, 78% stop rate
• Mechanism: Stock already priced in the expected outcome. Pre-earnings volatility triggers tight stops on stocks that barely move.
• Implementation: Config.SIGNAL_LONG_MAX_THRESHOLD = 0.90. Added to signal_generator, priority.py, conviction_calculator, dashboard.
• Expected impact: Avoid ~29 losing trades, save ~$116 in losses + $40 in commissions

#4 HIGH — Widen Stop Losses to 3%/4%/5% (IMPLEMENTED)

• Data: What-If analysis with Polygon actual exit prices: moderate stops improve total P&L from +$51 to +$166
• Mechanism: Fewer stops trigger (13 vs 35 of 46), letting more trades survive to scheduled exit (89% WR, +$20/trade avg)
• Implementation: Config.STOP_LOSS_LARGE_CAP = 0.03, STOP_LOSS_MID_CAP = 0.04, STOP_LOSS_SMALL_CAP = 0.05
• Expected impact: +$115 improvement in stop-related P&L

#5 MEDIUM — Enable Small Cap Longs at 70-80% (IMPLEMENTED, Finding #207)

• Data: Q3 2025 backtest with proper MCAP classification: 70-80% conviction small cap longs have 66.7% WR and +$30.68 EV (6 trades). 80-90% has only 30% WR and near-zero EV.
• Implementation: Changed small cap LONG rule from "require 90%+" to "allow 70-80% only". Full position sizing (15% of capital, Finding #211). Priority score 2/6.
• Rationale: Previous rule made small cap longs impossible (needed 90%+ but 90%+ was blocked). The 70-80% band captures genuine surprise moves in small caps.

#5b MEDIUM — Increase Position Sizes (IMPLEMENTED, Finding #211, Mar 10, 2026)

• Data: 95 closed trades analyzed. 90%+ filter reduces trade count by ~36% (26 trades eliminated), freeing capital. Position size quintile analysis shows Q5 (largest, avg $490) generates best absolute returns (+$139, $7.70/trade). Q1-Q2 (smallest, $157-$226) are net losers.
• Implementation: Large 25%→30%, Mid 18%→22%, Small 10%→15%.
• Rationale: Fewer trades post-filter means more capital available per trade. Larger positions amplify the edge on the remaining high-quality trades. On $2,450 capital: Large $735 (+$123), Mid $539 (+$98), Small $367 (+$122).

Other Recommendations (Not Yet Implemented — as of Mar 2026)

See docs/STRATEGY_PLAYBOOK.md §2 for current ship status; many of these have since shipped, been retracted, or evolved. - #2 CRITICAL: RED FLAG: Low win rate (32.1%). Root cause: 90%+ conviction and LONG AMC. Implementing #1 and #3 would raise effective WR to ~44%. - #3 HIGH: LONG AMC structural disadvantage. Two options: (A) skip all LONG AMC, (B) only trade at 70-80% conviction. Monitoring. - #6 MEDIUM: Commission optimization (natural from fewer trades after #1).

12. Data Sources and Methodology

13. Related Documents