Valid target tokens seen
Approved continued-training run scale.
Measured evidence
Continued training
Loss signals, stability controls, generation gates, reload behavior, and token-scale receipts.
Recorded finding: The training continuation path is instrumented and produces governed stability evidence.
On this page
Currency and next evidence gate
Decision signals
Key metrics and controls
The recorded values, with their controls and scope.
Eval-loss reduction
First-to-last public aggregate evaluation signal.
Checkpoint reload checks
All published reload checks passed.
Generation gate
6 hard failures across the approved prompts.
Source-linked visual evidence
What the evidence shows
Open any chart’s data table to inspect the exact values, or download the CSV.
Status matrix
Generation gate results
Passed and hard-failure counts in the current public generation gate.
Finding: The current generation gate fails; no prompt in the published set passed. That held result remains visible.
Passed
Hard failures
View source values
Scroll sideways if any columns are out of view.
| Group | Count |
|---|---|
| Passed | 0 prompts |
| Hard failures | 6 prompts |
Method
How to read this result
- Training, post-training evaluation, checkpoint reload, and generation approval are evaluated as distinct controls.
- Improved loss signals do not override a failed generation-quality gate.
What this result establishes
Generation quality and broad model capability remain under review and are not mature public claims.
- Aggregate sealed-continuation metrics only.
- Parameter and token units are normalized; token counts are shown in M where charted.
- No checkpoint paths, checkpoint hashes, tokenizer hashes, command arguments, source paths, dirty status, raw eval rows, raw sample outputs, prompts, generations, token IDs, logits, private traces, model weights, checkpoints, source internals, or mechanism details are bundled.
- Generation gate is intentionally published as failed/held, so public capability and production claims remain blocked.
- No dense-baseline superiority, downstream quality, production-readiness, unrestricted reproducibility, source-release, or proven-breakthrough claim.
Inspection and reuse
Downloads and provenance
Source data
Machine-readable summary
The recorded JSON artifact is staged from the governed public evidence pack during the build. Its currency notice determines whether it may be read as current.
Open source JSONProvenance
Build and evidence receipt
Use the public build receipt to compare the deployed site source, evidence digest, claim posture, and inference release state.
Open build receiptContinue reviewing
Related evidence
measuredreviewed
Refresh required- Observed
- Small controlled probes provide measurement signals suitable for engineering decisions and follow-up experiments.
- Supports
- Step counts, throughput ranges, parameter footprint, and pair-work reduction from bounded probes.
- Does not support
- Probe-scale behavior is not a substitute for matched end-to-end production benchmarks.
- Next decisive evidence
- Regenerate from the repaired exact artifact identity and close or retain the phase-41 contract-discovery failure.
Decision relevance: research · engineering
Review evidencemeasuredpartial
Held- Observed
- The integrated training path runs with observable safety and quality signals.
- Supports
- Bounded throughput, loss, safety-counter, warning, and claim-boundary evidence from the full training path.
- Does not support
- Current evidence does not approve commercial product readiness or unrestricted generation claims.
- Next decisive evidence
- Select one exact run and bind its metrics, receipts, model, data, configuration, and review before refresh.
Decision relevance: research · diligence
Review evidence