* SOTA attempt

* Improve score on SXM

---------

Co-authored-by: spokane-way <spokane@way>

* SOTA attempt

* Improve score on SXM

---------

Co-authored-by: spokane-way <spokane@way>

* SOTA attempt

* Improve score on SXM

---------

Co-authored-by: spokane-way <spokane@way>

…ity plateau confirmed

Patches 15/16/21 still uncontested in 150+ open + 10 closed PRs (6 consecutive
audits). PR openai#1430 stable OPEN, 0 comments, no comp owner activity for 16h.

After 13 research fires and 6 audits, the picture is clear: training-time
tweaks are exhausted at our 22M/1500-step scale. All 4 post-fire-9 ports
(Mousse/MuonEq-R/Depth Recurrence/QK_GAIN=5.0) are neutral within the
champion noise band. The "neutrality plateau" at 3.27-3.30 is the empirical
ceiling for training-time changes at our compute budget.

Best remaining moves (in expected value order):
1. H100 escalation of CHAMP_L4_seed42+EL stack with EMA+Tilt+INT6 GPTQ bundle
2. Coprime stride implementation (task openai#58) — only data-side direction
3. BPE-8192 ngram tables build (task openai#49) — enables tokenizer A/B

Spend ~$3.55/$36 (10% utilization). Pod healthy at 7h uptime.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ker identified

First tokenizer-side fire (0/24 patches in this category). Subagent found 3
candidates (BPE-Dropout, Complementary Weighting, Three-Tier Classification)
but ALL are blocked by our pre-tokenized .bin file pipeline.

BPE-Dropout requires live re-tokenization at training time → infeasible.
Complementary Weighting subagent incorrectly cited our MLX prototype, not
the H100 train_gpt.py. Three-Tier is PR openai#1402 pending validation.

Architectural insight: SP1024 may actually be optimal for our 22M architecture
(smaller embedding = more params for model body). Top PRs use SP8192 because
their depth-recurrence stack benefits from finer tokens. We may not need
BPE-8192. Task openai#49 deferred indefinitely.

Cross-domain coverage update (16 fires):
  training: 5, optimizer: 2, eval: 3, compression: 1, data: 2, tokenizer: 1,
  hardware: 0. Hardware still uncovered.

Per user instruction: queued, not shipped. No code patches.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>