gitadmin/foxhunt
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

baseline(per-horizon-cfc): 3-seed h6000 baseline JSON skeleton + retrieval gap

Browse Source 
Per spec §4.1 + Q-3-seed-baseline. Both baseline Argo workflows
succeeded:
- alpha-perception-baseline-seed16963 (39min)
- alpha-perception-baseline-seed16964 (26min)

Precise AUC values deferred: train pod stdout is in MinIO archive
(xl.meta format, needs mc auth). The on-disk alpha_train_summary.json
on /feature-cache/alpha-perception-runs/f22f3f948/ corresponds to
seed 16963 (last to finish). Task 15 implementer retrieves via debug
pod mount of feature-cache PVC.

Known data point: seed 16962 = 0.7454 best_auc_h6000.

The Smoke 1 invariant gate threshold is signal-derived per
pearl_tests_must_prove_not_lock_observations — formulas captured in
the JSON for Task 15 to apply once values are retrieved.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
jgrusewski
2026-05-21 16:10:48 +02:00
parent fde5511338
commit c535be046f
1 changed files with 50 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

50
docs/superpowers/notes/2026-05-21-baseline-3seed-h6000.json Normal file
View File

@@ -0,0 +1,50 @@
{
"purpose": "Smoke 1 invariant gate baseline (per spec §4.1, Q-3-seed-baseline). regression_tolerance derived from per-seed variance; the threshold is invariant-based (not a hardcoded AUC value).",
"commit_sha": "f22f3f948883721b7e70ca2e2c41b3ddba665299",
"training_config": {
"epochs": 20,
"early_stop": "none (patience=0)",
"loader_mode": "random",
"n_train_seqs": 8000,
"n_val_seqs": 1000,
"seq_len": 32,
"mamba2_state_dim": 16
},
"seeds": [
{
"seed": 16962,
"argo_workflow": "alpha-perception-diag-rnd-120534",
"status": "Succeeded",
"best_auc_h6000": 0.7454,
"best_epoch": 5,
"source": "previous diagnostic run, confirmed via session telemetry"
},
{
"seed": 16963,
"argo_workflow": "alpha-perception-baseline-seed16963",
"status": "Succeeded",
"duration_min": 39,
"finished_at": "2026-05-21T15:48:08+0200",
"best_auc_h6000": null,
"source": "RETRIEVAL_DEFERRED: train pod stdout is in MinIO archive (xl.meta format, needs mc auth). The alpha_train_summary.json on /feature-cache/alpha-perception-runs/f22f3f948/ corresponds to this seed (last to finish). Task 15 implementer retrieves via debug pod mounting feature-cache PVC."
},
{
"seed": 16964,
"argo_workflow": "alpha-perception-baseline-seed16964",
"status": "Succeeded",
"duration_min": 26,
"finished_at": "2026-05-21T15:35:02+0200",
"best_auc_h6000": null,
"source": "RETRIEVAL_DEFERRED: train pod stdout in MinIO archive. PVC summary was overwritten by seed 16963's run. For Task 15, can re-submit once with a seed-keyed output dir, OR estimate via available data."
}
],
"current_known_baseline": {
"n": 1,
"value": 0.7454,
"seed": 16962
},
"task15_pre_gate_action": "Before evaluating Smoke 1, the implementer MUST retrieve seed 16963's best_auc_h6000 via debug pod (mount feature-cache-pvc, cat alpha_train_summary.json). For seed 16964, EITHER re-submit ONE confirmation run with --out keyed per-seed, OR proceed with N=2 baseline (16962 + 16963), accepting the wider stddev estimate from sample size 2.",
"regression_tolerance_formula": "(2 * stddev) / (mean - 0.5)",
"smoke1_invariant_threshold_formula": "best_h6000 >= 0.5 + (mean - 0.5) * (1 - regression_tolerance)",
"notes": "Per pearl_tests_must_prove_not_lock_observations: threshold is signal-derived, not hardcoded. Per feedback_isv_for_adaptive_bounds: every bound in the control loop derives from observed dispersion."
}