Overview & Experiments 55 Synthesis Roadmap Lookahead Audit

Pyramiding Hold-out — step1.0/stop3.0/regime auf ungesehenen Fenstern

Confirmed

2026-06-12 sizingpyramidingholdoutvalidationdonchian

Hypothesis

Die in G5 selektierte Variante (pyramid_step_atr=1.0, pyramid_stop_mult=3.0, regime_gate=1, max_units=3) schlägt die Baseline auch out-of-selection: Return höher UND Max-DD <= 1.6x Baseline in >= 2/3 der Hold-out-Fenster 2020/2021/2023.

Verdict

**HOLD-OUT BESTANDEN** — die fixierte Variante gewinnt 3/3 ungesehene Fenster (Return höher, Max-DD <= 1.6x Baseline). Selection-Risiko aus G5 entkräftet; deploy-würdig nach Peters Freigabe.

gate

return > baseline AND max_dd <= 1.6x baseline in >= 2/3

sizing

{'max_position_pct': 1.0, 'position_size_pct': 1.0}

variant

{'pyramid_step_atr': 1.0, 'pyramid_max_units': 3, 'pyramid_stop_mult': 3.0, 'pyramid_regime_gate': 1}

windows

['COVID_2020', 'BULL_2021', 'CHOP_2023']

holdout_wins

3/3

Pyramiding Hold-out — step1.0/stop3.0/regime auf ungesehenen Fenstern

2026-06-12 · status: confirmed · 4.8s

Hypothesis: Die in G5 selektierte Variante (pyramid_step_atr=1.0, pyramid_stop_mult=3.0, regime_gate=1, max_units=3) schlägt die Baseline auch out-of-selection: Return höher UND Max-DD <= 1.6x Baseline in >= 2/3 der Hold-out-Fenster 2020/2021/2023.

Verdict: HOLD-OUT BESTANDEN — die fixierte Variante gewinnt 3/3 ungesehene Fenster (Return höher, Max-DD <= 1.6x Baseline). Selection-Risiko aus G5 entkräftet; deploy-würdig nach Peters Freigabe.

Key metrics

metric	value
holdout_wins	`3/3`
variant	`{'pyramid_step_atr': 1.0, 'pyramid_stop_mult': 3.0, 'pyramid_max_units': 3, 'pyramid_regime_gate': 1}`
sizing	`{'position_size_pct': 1.0, 'max_position_pct': 1.0}`
windows	`['COVID_2020', 'BULL_2021', 'CHOP_2023']`
gate	`return > baseline AND max_dd <= 1.6x baseline in >= 2/3`

COVID_2020 — WIN

variant	trades	adds	return_pct	ann_pct	max_dd_pct	sharpe	win_rate	profit_factor
baseline	62	0	49.15	48.99	10.41	4.6	0.484	1.95
pyramid/regime	62	10	58.15	57.95	10.07	4.605	0.468	1.982

BULL_2021 — WIN

variant	trades	adds	return_pct	ann_pct	max_dd_pct	sharpe	win_rate	profit_factor
baseline	63	0	-15.23	-15.23	16.68	-1.683	0.381	0.781
pyramid/regime	63	2	-14.79	-14.79	17.2	-1.609	0.365	0.789

CHOP_2023 — WIN

variant	trades	adds	return_pct	ann_pct	max_dd_pct	sharpe	win_rate	profit_factor
baseline	55	0	16.81	16.81	8.99	2.332	0.436	1.421
pyramid/regime	55	24	39.03	39.03	9.84	3.014	0.418	1.665