docs: refresh CLAUDE.md and companion docs for PRs #40-#50 #51

Merged
alee merged 3 commits from docs-sweep-prs-40-47 into main 2026-09-07 02:54:08 +00:00
Owner

Docs-only sweep, no code changes. Covers:

  • Per-provider quota measurement + credit-aware routing attenuation (#40)
  • The :path routing fix for vendor/model-shaped ids (#42)
  • Quota UI billing-shape clarity (#43)
  • Pinch's protected-window size cap (#44 -- docs/pinch.md verified accurate against what actually shipped, not duplicated)
  • OpenRouter as an opt-in allowlist provider (#45), with the x-ai/*/openai/*/anthropic/* exclusions stated as settled and non-negotiable, not "currently excluded"
  • The models-page deprecated/stale toggle (#46)
  • The confidence_threshold validation incident, the gated-model crash, and the natural-language-labels accuracy fix (#47-#50) -- extended into the existing classifier.mode section since it's directly continuous with the "local_encoder is live but not yet the settled default" note already there

Note on process: this branch was executed by an opencode session that degraded mid-run (a hallucinated "the diff is empty, reject everything" verdict, followed by a genuine repetition-loop failure and an upstream stream interruption). Independently verified the actual file contents were correct before finishing this myself -- the underlying work was sound, the reviewing agent wasn't. Also caught and fixed one real inaccuracy along the way (a swapped telemetry/polled provider example in docs/admin-portal.md) and deduplicated a section that landed twice across the merge with main.

Full suite: 1503 passed, 0 failures.

Docs-only sweep, no code changes. Covers: - Per-provider quota measurement + credit-aware routing attenuation (#40) - The `:path` routing fix for vendor/model-shaped ids (#42) - Quota UI billing-shape clarity (#43) - Pinch's protected-window size cap (#44 -- docs/pinch.md verified accurate against what actually shipped, not duplicated) - OpenRouter as an opt-in allowlist provider (#45), with the `x-ai/*`/`openai/*`/`anthropic/*` exclusions stated as settled and non-negotiable, not "currently excluded" - The models-page deprecated/stale toggle (#46) - The confidence_threshold validation incident, the gated-model crash, and the natural-language-labels accuracy fix (#47-#50) -- extended into the existing `classifier.mode` section since it's directly continuous with the "local_encoder is live but not yet the settled default" note already there Note on process: this branch was executed by an opencode session that degraded mid-run (a hallucinated "the diff is empty, reject everything" verdict, followed by a genuine repetition-loop failure and an upstream stream interruption). Independently verified the actual file contents were correct before finishing this myself -- the underlying work was sound, the reviewing agent wasn't. Also caught and fixed one real inaccuracy along the way (a swapped telemetry/polled provider example in docs/admin-portal.md) and deduplicated a section that landed twice across the merge with `main`. Full suite: 1503 passed, 0 failures.
alee added 3 commits 2026-09-07 02:15:31 +00:00
Covers per-provider quota measurement + credit-aware attenuation (#40),
the :path routing fix (#42), quota UI billing-shape clarity (#43), the
pinch protected-window size cap (#44, docs/pinch.md verified against
what actually shipped), the OpenRouter opt-in allowlist (#45, with the
firm x-ai/openai/anthropic exclusions stated as settled, not pending),
and the models-page deprecated/stale toggle (#46).

Also extends the classifier.mode section with the #48-#50 incident
chain (confidence_threshold validation, the gated-model crash, and the
natural-language-labels accuracy fix) since it's directly continuous
with the local_encoder-is-live-but-not-yet-default note already there.
The merge from origin/main brought in #48's own confidence_threshold
paragraph, which duplicated this branch's own independent write-up of
the same incident. Kept the more detailed version, removed the
redundant one, and added the #50 natural-language-labels accuracy fix
(5/9 -> 8/9 correct, confidence 0.15-0.90 -> 0.39-0.999) since it's the
direct continuation of the same incident thread.
alee merged commit 0d44c3dcf7 into main 2026-09-07 02:54:08 +00:00
Sign in to join this conversation.
No Reviewers
No Label
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: alee/6krrt#51