XGB-GATE-20 (xgb_side_filter_q20_phase2) INTERNAL ONLY

← Secret Lab index · xgb_gate20

Definition

Family	ml
Evidence tier	T2_PRIVATE_WATCHLIST
Public status	RESEARCH_WATCHLIST_ML
Lab status	ACTIVE_MONITOR
Effective status (this eval)	INSUFFICIENT_FORWARD_SAMPLE
Lab inception (forward start, no backfill)	2026-07-07T17:37:28Z

Historical Evidence (separate from forward)

source_report	reports/research/ml_10k_feature_discovery/xgb_gate_20_watchlist_activation.md
source_pr	18
historical_net_pnl	—
historical_max_dd	-72.41
historical_sharpe	—
historical_trade_count	—
validation_pnl	+186.75
locked_test_pnl	+315.55
search_adjusted_p	+0.01
reality_check_status	PASSED_PR18_SINGLE_SPLIT

Forward Evidence (since lab inception only)

forward_trades	135
forward_markets	100
forward_active_days	2
forward_net_pnl	-27.84
forward_win_rate	+0.49
forward_max_dd	-27.84
forward_worst_day	-16.75
forward_best_day	-11.10
forward_daily_pnl	{"2026-07-07": -11.0999, "2026-07-08": -16.7451}
forward_cumulative_pnl	-27.84
forward_sharpe_active	—
forward_sharpe_calendar	—
latest_forward_timestamp	2026-07-08T23:15:32.441000+00:00

Risk Metrics

No risk overlay on this candidate.

Internal Ranking

forward_net_pnl_rank	2
forward_drawdown_rank	2
forward_stability_rank	—
forward_trade_count_rank	1
composite_private_lab_score	+1.67

Internal sorting only (lower is better). This score cannot promote a candidate or change any public status.

Why this tier

Strong historical evidence or established public forward series; reality check: PASSED_PR18_SINGLE_SPLIT.

T2 Private-Watchlist Policy Status

Policy requirement	Status	Detail
valid scoring bundle / reproducible definition	PASS	scored
forward trades >= 100	PASS	has 135
forward active days >= 10	NOT MET	has 2
forward net PnL > 0	NOT MET	has -27.845
no safety violation	PASS	paper_only, real_orders_enabled=false, no backfill (validator-enforced)

Not eligible for private-watchlist review yet — the unmet lines above are why.

Promotion Checklist (human only)

Human review requested and recorded (never automatic)
Forward sample meets the target tier's policy minimums
Historical vs forward evidence consistent (no backfill, no locked-test rows counted as forward)
Risk metrics sane (drawdown, worst day, concentration)
Search-adjusted evidence disclosed (required for public claims, not for monitoring)
No alpha/profitability language in any proposed public copy

SECRET LAB — INTERNAL ONLY. Paper simulation bookkeeping. No real orders, no live trading, no execution candidate, no promotion from this page. Forward evidence starts at lab inception and is never backfilled. Not on the public dashboard; do not link from public_site.