Skip to content

docs(models): update reasoning-only stream limit to input-aware bounds - #143

Merged
sre-helmcode merged 1 commit into
mainfrom
docs/watchdog-input-aware-limits
Oct 2, 2026
Merged

sre-helmcode merged 1 commit into
mainfrom
docs/watchdog-input-aware-limits

Conversation

@sre-helmcode

Copy link
Copy Markdown
Contributor

What

Updates the public "Reasoning-only stream limit" section (EN + ES) to match the deployed watchdog behavior after the 2026-10-02 tune (nan-devops commit ec88bc24, pending PR):

  • 420 seconds closes a reasoning-only turn; a stream that already produced more than 45,000 reasoning characters gets until 600 seconds.
  • The reasoning-character budget scales with prompt size: ~150,000 characters plus half the prompt's size, capped at 300,000.
  • Replaces the stale "60,000 characters / ~44,000 measured ceiling" claims (source of a member report quoting outdated numbers).

Why

The published numbers no longer match the platform; members quote them when reporting cuts.

Test evidence

npm run build PASS (Astro build, adapter Cloudflare). Usage-chunk paragraph and guidance bullets unchanged (still accurate).

Deploy notes

Deploys automatically on merge via Cloudflare Pages (push to main). Merge after the nan-devops watchdog PR lands so docs never describe behavior ahead of the deploy.

The published numbers (60,000 characters / ~44,000 measured ceiling) are
stale after the 2026-10-02 watchdog tune: the seconds bound is now 420s
(600s past 45,000 reasoning characters) and the character budget scales
with prompt size (150,000 + half the prompt, capped at 300,000).
@sre-helmcode
sre-helmcode merged commit eaaedf2 into main Oct 2, 2026
2 checks passed
@sre-helmcode
sre-helmcode deleted the docs/watchdog-input-aware-limits branch October 2, 2026 12:44
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants