Skip to content

test(provider): centralize catalog model pins, add a delisting tripwire#91

Merged
Aayam Bansal (aayambansal) merged 1 commit into
mainfrom
test-catalog-pins
Jul 5, 2026
Merged

test(provider): centralize catalog model pins, add a delisting tripwire#91
Aayam Bansal (aayambansal) merged 1 commit into
mainfrom
test-catalog-pins

Conversation

@aayambansal

Copy link
Copy Markdown
Member

models.dev delisted claude-sonnet-4-20250514 this afternoon and 10 provider tests hardcoding that id failed at once (this broke main's Test check — the failures predate #90 and reproduce on earlier commits).

  • All pinned anthropic ids collapse into two constants (SONNET = claude-sonnet-4-6, OPUS = claude-opus-4-5) used across whitelist/blacklist/getModel/defaultModel/small_model/variant tests — next drift is a one-line fix.
  • New tripwire test runs first: if the live catalog drops a pinned id, it fails alone with explicit instructions instead of scattering 10 cryptic failures.

Verified: provider file 69/69, full backend suite 853/853, typecheck clean.

models.dev delisted claude-sonnet-4-20250514 on 2026-07-05 and ten
provider tests hardcoding it failed at once with unrelated-looking
errors. The pinned ids now live in two constants (SONNET/OPUS, currently
claude-sonnet-4-6 and claude-opus-4-5) used everywhere, and a guard test
runs first: when the catalog drops a pinned id it fails alone with
'update the SONNET/OPUS pins at the top of this file' instead of
scattering failures across the suite.
@vercel

vercel Bot commented Jul 5, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
openscience Ready Ready Preview, Comment Jul 5, 2026 5:49pm

Request Review

@aayambansal Aayam Bansal (aayambansal) merged commit d7a0337 into main Jul 5, 2026
11 checks passed
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.

1 participant