Generate & rank bracket variations against the parity-locked engine · every variation shown, ranked by robustness floor
✓ Engine validated to your live fills. Regime 5-min reproduces 226 trades / $75,368 / PF 2.03 vs TradingView's 235 / $79,170 / PF 2.07 — within ~5% on every metric, win rate near-exact. Every number below is computed fresh from that engine, not a black box.
Strategy
Net & Max DD$ in the table are shown at this size and scale exactly linearly (2 contracts = 2× the dollars, 10 = 10×). Change it anytime — the table re-scales and re-filters instantly, no re-run needed.
Chart timeframe
5m is the parity-locked, validated setting. 2m / 15m run the same Regime engine on re-bucketed bars — useful to explore, but not yet checked against a TradingView export on those timeframes, so treat their numbers as provisional.
Fixed at 5-min. Fabio's entries are taken from your TradingView deep export (footprint signals can't be rebuilt from bars), so the engine pins those exact entries and sweeps only the exit geometry below. Contracts above scales net & DD$ live, same as Regime.
Seed export
Lookback · analysis window — ends at the latest bar/trade
Last 365 days of data. The most recent ~20% is held out as an out-of-sample tail (never used to rank); the rest is split in half for the Robust floor. The OOS column shows how each variation does on that held-out tail — the real forward test.
For Fabio this windows the pinned trades by entry date. Short windows (30/90d) leave few trades to fold — lean on 1yr, which covers the full Jun-2025→Jun-2026 export; longer settings just include the same trades.
Brackets
from
to
step
Long TP
Long SL
Short TP
Short SL
ATR multiple of the entry-bar ATR(14). e.g. TP 5.7 / SL 4.8 = the deployed Regime bracket.
from
to
step
Target-R (rMult)
Stop-swing bars
Stop-buffer ticks
Breakeven @ R
Max-hold bars
All five are exit dials — leave a row's to/step blank to pin it, fill both to sweep; they multiply into the grid. Target-R = take-profit as a multiple of the trade's stop distance. Stop-swing / stop-buffer set where the protective stop sits (whole bars / ticks — fractional steps collapse to the unique integers, no duplicate recipes). Breakeven @ R arms a move-to-entry stop once price reaches that R; tick the box to also test it disabled. Max-hold = bars before a time-stop fires; 0 = no time-stop, ride to the 15:55 EOD flatten — the parity-locked baseline is Target-R 2.8 / stop-swing 5 / buffer 2 / BE 1.2R / max-hold 0. To probe whether Fabio's edge is in holding to the flatten, open Target-R wide (e.g. to 8) and sweep max-hold 0→24.
Exits · fixed for this strategy (sweepable next)
EOD flat 16:45 ETTime-stop 30 barsRegime-reversal onTrailing act 4.7 / off 1.0 ATRBreakeven off
Editable pins applied to the whole sweep. Hours = ET hour numbers (comma-separated). Days = Mon–Fri.
Min trend score & daily loss · leave to/step blank to pin · fill them to sweep as an axis
from
to
step
Min TS Long
Min TS Short
Daily loss $
Sweeping these multiplies the combo count, same as a bracket axis.
Classifier block · the regime engine itself · sweeps as one block (recomputes the signal per combo — the slow axis)
from
to
step
Fast EMA
Slow EMA
Sticky bars
Min score
Slope mult
Leave to/step blank to hold a knob at deployed; fill them to sweep it. Slow EMA must stay above fast. Struct 20 · ATR 14 · slope LB 3/5 stay deployed. Each classifier combo recomputes the signal (~1s), so this is the slow axis — watch the recompute count.
Entry parameters · pinned — entries come straight from your export
Every exit dial is now live above — Target-R, stop-swing, stop-buffer, breakeven (and BE-off), and max-hold all sweep into the grid. What stays pinned is the entry side: Fabio's entries are taken verbatim from your footprint export, so regime-gate length and max-extension ATR can only become axes once you send a filters-off export (the unfiltered entry superset), and the footprint delta multiplier can't be rebuilt from bars at all. Until then, entries fire exactly as your export recorded them — the sweep only ever changes what happens after the entry.
Targets · all three are hard filters — rows that miss any are hidden · blank a field to disable that gate
All three are hard filters, applied over your lookback window at the contract count above: any variation whose Max DD$ exceeds the cap, whose win rate is under the floor, or whose worst-half (Robust) PF is under the floor is removed. If none qualify the table is empty and shows the lowest-DD reference row. Drawdown scales linearly with size (2 cons = 2×, 10 = 10×). Leave any target blank to disable that gate. The holdout drawer still shows both halves and the OOS tail so a rosy window can't hide a weak stretch.
Configure brackets, pick a lookback, and hit Run generation. Every variation is returned and ranked by its weakest half of the lookback window.