Report a reference strategy beside campaign candidates #263

New Issue

2026-07-13T18:06:14+02:00

claude commented

2026-07-13 18:06:14 +02:00

The pipeline grades candidates against zero, not against a baseline on the same matrix. On a secular-trend instrument, pure beta sails through every stage: an always-long bias=+1 blueprint on XAUUSD 2018-2026 (vol{60, k6}, slippage-only cost, seed 0) records a pooled WF-OOS mean of +0.81 R/trade with prob_le_zero 0.002 and overfit probability 0.003. Trials-deflation guards against selection luck, not confound luck.

The same hand-authored control exposed an apparent breakout survivor on the same instrument as diluted beta (candidate OOS +0.12–0.25 at P(≤0) ≈ 0.08 — strictly worse than the no-signal control on every stat).

A campaign-level reference-strategy slot — a strategy ref whose per-cell stats are recorded beside every candidate's (or subtracted as an alpha reading) — would make the beta control structural instead of a per-project authoring convention.

The pipeline grades candidates against zero, not against a baseline on the same matrix. On a secular-trend instrument, pure beta sails through every stage: an always-long bias=+1 blueprint on XAUUSD 2018-2026 (vol{60, k6}, slippage-only cost, seed 0) records a pooled WF-OOS mean of +0.81 R/trade with prob_le_zero 0.002 and overfit probability 0.003. Trials-deflation guards against *selection* luck, not *confound* luck. The same hand-authored control exposed an apparent breakout survivor on the same instrument as diluted beta (candidate OOS +0.12–0.25 at P(≤0) ≈ 0.08 — strictly worse than the no-signal control on every stat). A campaign-level reference-strategy slot — a strategy ref whose per-cell stats are recorded beside every candidate's (or subtracted as an alpha reading) — would make the beta control structural instead of a per-project authoring convention.

claude added the idea label 2026-07-13 18:06:14 +02:00

claude referenced this issue

2026-07-23 20:56:56 +02:00

Build a deployment arena: a standing proving ground where source-blind agents exercise the released binary #322

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: Brummel/Aura#263