# Scarface v1.0.3 - Builder 2 ranked fixes (27 Oct 2026 AEDT). Not handed to BT1.
Code: python/aio7/plugins/scarface_s1/scarface_s1_v103.py (v1.0.2 untouched). Backup: backups/allinone_v7_pre_scarface_v103_20261027_1206.tgz
Off check: v1.0.3 with every new switch off = v1.0.2, IDENTICAL on NVDA/SPY/TSLA, 5m and 15m, base and A+ (12/12). Fire test: all 10 switch values fire on all 3 symbols (offfire_v1.0.3.json).
Scorer: STRICT (sim took it, same direction). Headline = +/-5 min; second column = +/-15 min (Builder 2's tolerance). Stop = his stop TYPE must equal the code StopMode.
Sweep: 1920 configurations (2 ranges x RetestMaxBars{off,3,5,10,20} x Confirm{Close,StrongClose} x LevelPriority{off,GapPremarket,Nearest,RangeFirst} x Window{10:30,11:00} x Stop{EntryCandle,Level,Impulsive} x Psych{0.5,1,5,Auto}). SWEEP_v1.0.3.csv.
| config | range | +/-5 min (headline) | +/-15 min |
|---|---|---|---|
| v1.0.2 A+ as is | 5m | 4/26 | 5/26 |
| v1.0.2 A+ as is | 15m | 5/26 | 5/26 |
| best: Close, RangeFirst, 11:00, EntryCandle, Psych any | 5m | 6/26 (23%) | 9/26 (35%) |
| best | 15m | 6/26 | 6/26 |
Bar 80%: STILL FAILING. The best config was picked on the same 26 trades out of 1920 tried, so it is IN-SAMPLE and optimistic (+2 over as-is is within the noise for 26 trades). No IS-free holdout exists at this sample size.
Per factor (best 5m, +/-5): RetestMaxBars hurts (3 -> 4, 5-20 -> 5); Close > StrongClose (6 vs 5); RangeFirst 6 > off 5 = GapPremarket 5 > Nearest 4; 11:00 > 10:30; StopMode EntryCandle 6, Level 1, Impulsive 0 (the wider stop fails MinRR 2 and drops the trade); Psych step does not change the pass count.
Note: v1.0.2 5m drops 5 -> 4 versus the earlier report because the stop step now requires the stop TYPE to match (stricter).
Remaining misses (best 5m, +/-5): trigger 19, setup 17, entry 17, stop 2. Mostly: no taken trade within 5 min of his frame time (the frame times are approximate, a DATA LIMIT; retest timing NOT STATED, SF-D10/D11).
New data found on the box: smt-python/data/stocks/*_duka_2025-01-01_2026-10-21 (Dukascopy CFD, RTH only, no premarket). It covers SPY Dec 2025 (T27-T31), PLTR (T33, T41) and TSLA 21 Sep (T38).
Scored separately (8 trades): 0/8 on every config. Causes: his stop is the impulsive candle (T27, T28, T30, T41), there is no taken trade within 5 min (T29, T31, T33, T38), and there is no premarket on CFD data (DATA LIMIT).
Still missing: MU, CRDO, INTC, SNDK.
Files: SWEEP_v1.0.3.csv, PER_TRADE_v1.0.3_best_5m_tol5.csv, PER_TRADE_v1.0.3_best_5m_tol15.csv, PER_TRADE_v1.0.3_best_5m_duka_extra.csv, offfire_v1.0.3.json. Run: code/forensic_v103.py (sweep), code/offfire_v103.py.