Skip to content

Research/v13 confirmatory - #43

Merged
Pentexa merged 2 commits into
mainfrom
research/v13-confirmatory
Nov 10, 2026
Merged

Pentexa merged 2 commits into
mainfrom
research/v13-confirmatory

Conversation

@Pentexa

@Pentexa Pentexa commented Nov 10, 2026

Copy link
Copy Markdown
Owner

No description provided.

…) + failure classification

Preregistered confirmatory study on independent seeds 21800-21807
(8 x 100 per cell, PYTHONHASHSEED=0), separate from pilot artifacts.

Primary L endpoints (candidate - baseline, paired bootstrap 95% CI):
- stable exact-D1 13-12: +36.88 pp [13.13, 69.38] (excludes 0)
- novel  exact-D1 13-12: +17.88 pp [6.38, 31.63] (excludes 0)
- rolling-generator ablation 13-gen equals the deltas exactly; -notp 0
- per-seed nonzero: stable 6/8, novel 6/8 (not single-seed driven)
- success: stable +0.50 [0.13, 0.88]; novel -9.62 [-24.13, 0.25]
  (no measured novel success effect; discovery != success)

Preservation at 8x100 (13 == 12 exactly on every seed): F noisy_obs,
H stable, R stable; exploratory R -nodc recovery 80.38 vs 68.38.

Mission C failure classification (8x40 instrumented): rolling generation
eliminates C2 (candidate not generated: 27.5%/15.6% in v12 -> 0.0% in
v13); remaining failures are C1 (behavioral floor) and C3 (verification
latency); zero C4/C5.

Mission D: H/R regressions are execution-limited, not learning-limited
(>=96% episodes learn the correct rule) and unchanged by v0.13.

Artifacts: results/v13_confirmatory/ (raw per-seed JSON, JSONL
checkpoints, diagnostics, analyses, rendered tables). 11 stamped
Quality v0.5 manifests; all verify with 0 FAIL. Smoke/profiling used
throwaway seeds 90000+; confirmatory seeds untouched by tuning.

Tests: full python suite 224/224 (214 legacy + 10 new tool tests);
audit suite 170/170; artifact gate PASS.

See docs/RESEARCH_V13_CONFIRMATORY.md (protocol + results).
…it-blob hashes)

Manifests now record commit_sha=e4972b6 and canonical git-blob hashes of
the committed artifacts (verifier: 0 FAIL on all 11 manifests; per-seed
means/CI deltas recompute exactly from raw vectors).
Copilot AI balanced review requested due to automatic review settings November 10, 2026 13:07

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

@Pentexa
Pentexa merged commit d001a00 into main Nov 10, 2026
6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants