Risk & Ensemble Research Status INTERNAL / OBSERVATIONAL

← Performance Breakdown · Research Watchlist

Paper-Only Risk Limits Overview

Draft, inactive, paper-only risk packages exist for PACE-3/10 (Champion Review) and XGB-GATE-20 (Research Watchlist — ML). Every package sets paper_only: true, real_orders_enabled: false, and execution_candidate: false, references no order adapter, and is not read by the running risk engine. No real orders; no execution candidate.

PACE-3/10 risk package DRAFT — INACTIVE

paper_onlyTrue
real_orders_enabledFalse
execution_candidateFalse
requires_manual_approvalTrue
Max contracts / market10
Max daily / weekly cost (USD)80.0 / 300.0
Max open markets3
Daily / weekly stop-loss (USD)15.0 / 50.0
Max-drawdown stop (USD)75.0
Consecutive-loss cooldown3 losses → 60 min

Full limit set (market-quality, execution, and kill-switch gates) lives in the config file. The running risk engine does not read this file; activating any of it would require a separate, explicit, human-reviewed commit.

XGB-GATE-20 watchlist risk package DRAFT — INACTIVE

paper_onlyTrue
real_orders_enabledFalse
execution_candidateFalse
requires_manual_approvalTrue
Max contracts / market10
Max daily / weekly cost (USD)40.0 / 150.0
Max open markets2
Daily / weekly stop-loss (USD)10.0 / 30.0
Max-drawdown stop (USD)50.0
Consecutive-loss cooldown3 losses → 60 min

Full limit set (market-quality, execution, and kill-switch gates) lives in the config file. The running risk engine does not read this file; activating any of it would require a separate, explicit, human-reviewed commit.

Ensemble Selector Research Status

Simple fire/agreement selectors over the six BTC paper strategies were tested with a chronological validation/locked-test split (validation selects, locked test evaluated once) and a search-adjusted reality check over all candidates vs the PACE-only benchmark. q=10 only; no leakage; no same-day-future information.

CandidateKindValidation PnLLocked-test PnL Test tradesTest max DDSelected
pace_default_dove_overrideselector+375.68+84.80605-70.5042yes
pace_sieve_precision_filterselector+22.21+2.7915-19.8
pace_dove_agreementselector+18.86-3.6117-23.1
pace_sieve_agreementselector+22.21+2.7915-19.8
pace_daily_stop_gateselector-15.18-27.85321-101.09
baseline_pace_onlybaseline+65.90+72.60603-70.5042
baseline_dove_onlybaseline+139.90+4.9019-19.4
baseline_sieve_onlybaseline+112.14+13.6017-15.0
baseline_same_universe_all6baseline+347.96-53.702680-453.4473

Best Ensemble Result

Selected on validationpace_default_dove_override
Locked-test net PnL (selected)+84.80
Locked-test net PnL (PACE-only)+72.60
Beats PACE-only on locked testTrue
Search-adjusted reality check p0.6235
VerdictNOT_PROMOTED

search-adjusted reality check p=0.6235 >= 0.05

XGB-GATE-20 Forward Overlay (descriptive only)

Status: INSUFFICIENT_SAMPLE — XGB-GATE-20 scores exist only from forward inception (2026-07-07T17:37:28Z); ~1 day of gated decisions is far too thin for selector evaluation. Descriptive only; not eligible for any verdict.

Gated trades since inception: 30, net -20.40.

Boundaries

No real orders. Nothing runs against real money. No execution candidate. No ensemble was promoted; the strongest possible outcome of this track is an internal review flag, which itself is not a promotion. Production trading config was not modified. Forward PnL is never backfilled.

Research-only paper trading simulation. No real orders placed. Not investment advice. Generated 2026-07-09T12:24:27Z.