Failure ledger — negative results and non-claims
What was tested, what happened, and why it is not a claim. Prefer under-claiming. Numbers from sealed summaries / ledgers; if unsure, see the cited path.
Hard non-claim (claim-stack era): there is no matched candidate-vs-stock first-cross ΔT%. Joint stock never crossed where measured (seeds 1337 and 2029). Seed 4242 had no stock arm.
| What was tested | Outcome | Why not a claim |
|---|---|---|
| F2 short warmdown (WD 256), curated/prior/EMA finals stack | Terminal CE ≈ 3.405; no first-cross ≤ 3.3821 | Quality miss. No T_cross. |
| F2b long warmdown (WD 1024), same family | Quality HIT (cross @4736, terminal ≈ 3.377); historical wall ≈ −2.9% vs notebook F1 | Speed framing retracted. |X| below ~6% infra/noise floor; stack confound. Quality hit only. |
| Historical notebook F1 (~229.1 min) | Absolute cross exists; torch unlogged | Not the sealed denominator. Use F1_seal only for published %. |
| claim_muon_par (Muon + parallel + triage + prior + EMA) | Final val 3.386130; never crossed; slower wall than seal | Quality miss → speed N/A. Do not publish as win. |
| claim_muon_wsd | Quality HIT (~3.376 terminal); T ≈ 3.899 h | ~0.8% slower than F1_seal (3.866 h). Quality without under-baseline speed. |
| Board-v3 800-step claim-class screens | Control wins/ties; schedule fracs tax step time | No promote. Proxies ≠ full claim. |
| FineWeb cleaners / curation screens @800 | No arm promoted to full claim | Closed levers; not under-baseline recipes. |
| Adam LR 0.003 alone | Mid-horizon lead failed persistence / 2400 gates | NO-GO for claim stack. |
| Dual-polar split-QKV | ~+0.015 CE but time ratio ≈ 1.026 | CE edge eaten by step tax. |
| muon_lr 0.025 | Reverse mid-run | NO-GO. |
| Fixed interleave shard order | ~0 edge | NO-GO. |
| Muon NS6 / warmdown 300 | ~0 @400 screens | NO-GO. |
| Late EMA as train CE | Mid ema_val mirage; terminal ~0 | NO-GO for train metric. |
| Tier A mom 0.90 / 0.98, Adam WD 0.05 / 0.2, warmdown 160 | 5/5 NO-GO; zero promote | Defaults stay; not a claim path. |
| Claim-stack joint stock seed 1337 | Terminal 3.3848; first-cross NONE | Control quality miss. Matched ΔT% undefined. |
| Claim-stack joint stock seed 2029 | Terminal ≈ 3.3840; first-cross NONE | Same. Matched ΔT% undefined. |
| Claim-stack seed 4242 stock arm | Not run (candidate-only by design) | No matched pair on this seed. |
| Component bank alone (systems + cheap split @800/@2400) | Real CE/time component evidence | Through 2026-07-14: 0 under-baseline full recipes. Bank ≠ claim until full-horizon seals. |
| Cross-account cardinal % mixing research T with clock F1 (early policy) | Forbidden for publication under early FAIR_CLOCK | Public ΔT% uses F1_seal on matched GPU/torch pin; identify accounts. See ACCURACY_NOTES resolution. |
Explicit matched-pair statement
| Seed | Candidate first-cross | Stock first-cross | Matched ΔT% vs stock |
|---|---|---|---|
| 1337 | yes @4736 · 3.553 h | NONE | N/A |
| 2029 | yes @4736 · 3.567 h | NONE | N/A |
| 4242 | yes @4736 · 3.511 h | not run | N/A |
Published under-baseline percentages compare candidate T to F1_seal (and optionally F1_Claim-W2 for seed 2029), not to joint stock.
Sources: MULTI_SEED_CLAIM.md, seed summaries, TIER_A_SEAL.md, claim_muon_par/SUMMARY.md, evidence_ledger_v1, FAIR_CLOCK.md, board_v3/WRITEUP.md.