Controlled scientific evidence execution ledger

[FlorianPfaff]

Purpose

This issue is the append-only operator ledger for the permanent Paper evidence contracts workflow. Exact commands posted here may request one reviewed, target-free controlled study from main after the trusted comment-dispatch gate is merged.

Allowed studies:

• simulation-based-calibration
• synthetic-benchmark-sbc
• recursive-corruption
• all-controlled

Appending replay requests byte-identical same-environment replay. Commands do not open Deform360 confirmation objects, DLO4/DLO5, held-v8, Prob4D targets, real-provider outcomes, or Causal4D physical executions. A completed workflow artifact remains controlled evidence until separately reviewed and promoted.

[github-actions]

Exact merged recursive-corruption v2 evidence

• source commit: b201c4c64622279623a32376dc296616469675d4
• launcher commit: e7e0daa108b9c30f9d31dbd63ac858a242d9d7d0
• result ID: 5df4c418f518192bb4272f2bc54b40ada2a2a38395fc300497e76e1f9a5462bc
• manifest ID: 65234dea7064ad7579ea7b922691d66a39e3fe3e703f3948b34f4ace2774f186
• independent seed-domains: 200
• retained scalar records: 12,000
• deterministic scientific replay: byte-identical
• all registered co-equal criteria passed: true
• retuning on the fresh roster authorized: false

Primary 1 — guarded recursion minus matched guarded residual, stress full-sequence RMSE: -2.118 mm (-2.230, -2.007 mm paired 95% interval).
Primary 2 — guarded recursion minus unguarded Gaussian recursion, materially harmful accepted updates per seed-domain: -123.92 (-127.71, -120.17 paired 95% interval).
Registered companions — corruption-window RMSE: -2.886 mm; recovery-window RMSE: -1.939 mm; clean-control cost: 0.045 mm.

Scientific boundary: controlled synthetic mechanism evidence only. No real provider, Deform360 confirmation outcome, Causal4D physical execution, DLO4/DLO5 reserve, or held-v8 target was opened.

Artifact: https://github.com/IPS-Stuttgart/BayesianPhysTwin/actions/runs/32816532073