Backtest Desk

TRADING MODE — loads the live bot's scout criteria

▸ Advanced toggles (matrix-accuracy, phase 2/4 experiments)

EVALUATE MATRIX ACCURACY — walk every intraday bar through BTDConfluenceMatrix; measure forward returns by stage. Slower but answers "does Stage 2 actually mean building?". OFF by default — Phase 5+ work doesn't use the records; opt in only when re-tuning the BTD matrix itself.

PHASE 2: MATRIX-GATED ENTRIES — stage gate (0+2 only, reject stage 1), volume_conviction ≥ 1.5 bonus, composite as size modifier (<-1 → 1.30×, ≥1 → 0.80×). Based on 580K-record dimension analysis.

PHASE 4: RE-ENTRY COOLDOWN (bars)

PHASE 4: RVOL RANK CAP (0 = off)

PHASE 4 A2: 1m SIZE MULT (1.0 = off; 0.5 halves 1m positions) — shrinks 1m-TF position sizes; leaves 5m/10m/etc untouched. Motivated by the 580K-record dimension analysis (5m/10m had 4–5× the per-trade edge of 1m).

MIN % CHANGE

MIN RVOL (needs ≥5d history)

MIN VOLUME (shares)

MIN PRICE ($)

MAX PRICE ($)

TOP N PER DAY

TICKERS (comma or newline separated) — only for one-off ticker validation

START DATE

END DATE

PRIMARY TIMEFRAME

ALSO BACKTEST THESE TIMEFRAMES — tick any to run the window on multiple timeframes. Each tf's matrix records are tagged; dimension analysis slices by tf to show which dims carry signal at which horizon. Leave empty for single-TF run.

1m 5m 10m 15m 30m 1h 4h 1d 1w

SEED CAPITAL ($)

STOP ATR MULTIPLIER

PARITY MODE — picks how closely this run mirrors the live executor. LEGACY = today's backtest, byte-for-byte. LIVE PARITY = EQM entry gate, OPM + MatrixPulse per bar, MCM modulation, ghost exits (CONTEXT/SUPERTREND/PROFIT_LOCK/ BREAKEVEN_FLOOR/LULD), AMSM cold-start aggressive. HYBRID = you pick each feature individually for A/B attribution.

SCANNER MODE — FORWARD (new default, live-realistic) walks intraday bars chronologically and records each candidate's qualification time; entries fire at-or-after that timestamp. HINDSIGHT (legacy) ranks the WHOLE day's candidates by (change_pct × volume) using end-of-day data — the look-ahead bias that inflated prior backtests by ~$12K. Use HINDSIGHT only for direct comparison against historical runs.

BACKTEST DESK

CONFIGURE RUN

AGGREGATE RESULTS

PER-TIMEFRAME

PER-TICKER

EQUITY CURVE (FIRST TICKER)

DAILY CANDIDATE SAMPLE (LAST 20 DAYS WITH PICKS)

MATRIX ACCURACY — does "Stage N" actually mean something?

MATRIX DIMENSION SIGNAL/NOISE — is each dim actually predictive?

JOB HISTORY