policy(contributor): per-vendor Frontier floor; state model gate by exact model#6584
Merged
Merged
Conversation
…xact model The Frontier-only table used family wildcards (`claude-sonnet-*`, `claude-haiku-*`), which excluded `claude-sonnet-5` — a current-generation frontier model — while the accepted row still admitted `claude-opus-4-7`. State the floor per vendor by exact model instead: Anthropic claude-opus-4-8+, claude-sonnet-5+ OpenAI gpt-5-5+ Codex codex-5-5+ and name the exclusions explicitly (opus-4-7 and below, sonnet-4-6 and below, all haiku incl. haiku-4-5, all composer, gpt-5-4/5-3 and below, codex-5-4 and below) so no wildcard reading can misclassify a model. Also document how commit evidence is read. A Co-Authored-By trailer is weighed against the commits carrying the implementation: a session that falls back to a sub-Frontier model part-way through leaves sub-Frontier trailers without the change having been generated below the floor, and that is a declaration to confirm rather than dishonesty. The account-level problem is a Model:/Tier: line that contradicts the trailers in the direction that clears the gate. Extend the pre-cutoff grandfather clause to name Haiku, since claude-haiku-4-5 sat in the Standard row until 2026-07-24T02:03:45Z.
matthewevans
enabled auto-merge
July 24, 2026 04:21
Contributor
📝 WalkthroughWalkthroughUpdated AI contributor documentation to add ChangesTier gate documentation
Estimated code review effort: 1 (Trivial) | ~2 minutes Possibly related PRs
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 2
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@docs/AI-CONTRIBUTOR.md`:
- Line 38: Replace the date-only cutoff in the grandfathering paragraph with the
precise UTC timestamp 2026-07-24T02:03:45Z, and explicitly state whether PRs
opened at that exact instant are included. Use this same inclusive or exclusive
comparison consistently for both the policy applicability and the historical
claude-haiku-4-5 explanation.
- Around line 32-34: The model-trailer policy in the “How commit evidence is
read” section gives conflicting outcomes for sub-Frontier commits. Reconcile the
statements so they define one authoritative hard-gate outcome, including the
required declaration correction and re-evaluation flow for
Frontier-to-sub-Frontier fallback, while retaining the account-level escalation
for contradictory Model/Tier claims.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro Plus
Run ID: 3ec0901b-ec55-431b-9af4-f237c00143c1
📒 Files selected for processing (1)
docs/AI-CONTRIBUTOR.md
Comment on lines
+32
to
+34
| **Frontier-tier models only.** There is no longer a Standard tier. The floor is per-vendor and is stated by exact model, not by family wildcard — `claude-sonnet-5` is accepted while `claude-sonnet-4-6` is not, so a `claude-sonnet-*` reading of this table is wrong. **Not accepted:** `claude-opus-4-7` and below, `claude-sonnet-4-6` and below, every `claude-haiku-*` including `claude-haiku-4-5`, every `composer-*`, `gpt-5-4` and below including `gpt-5-3`, and `codex-5-4` and below. If that is your model, abort per §0 rather than opening a PR. A PR declaring a non-Frontier model, or whose commits show one, will be closed as out-of-policy without an implementation review. This is not a judgement about those models generally; it reflects that review capacity here is the scarce resource, and sub-Frontier runs have consistently consumed several maintainer rounds per PR to reach a standard a Frontier run reaches on the first pass. | ||
|
|
||
| **How commit evidence is read.** The gate is about the model that *wrote the change*, so a `Co-Authored-By:` trailer is read against the commits that carry the implementation. A session that starts on a Frontier model and falls back to a sub-Frontier one part-way through — a usage limit, a harness default — leaves sub-Frontier trailers on later commits without the PR having been generated below the floor. That is a fixable declaration problem, not dishonesty: expect to be asked to confirm which model did the work. What earns a close is the whole run sitting below the floor; what escalates to an account-level problem is a `Model:`/`Tier:` line that contradicts the trailers in a direction that clears the gate. |
Contributor
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
Resolve the contradictory handling of sub-Frontier commit trailers.
Line 32 says any commit showing a non-Frontier model closes the PR, while Line 34 says a Frontier-to-sub-Frontier fallback is only a fixable declaration problem. These rules produce opposite outcomes for the same evidence and conflict with .claude/skills/pr-review-loop/SKILL.md’s hard-gate behavior. Define one authoritative outcome and the required remediation/re-evaluation flow.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@docs/AI-CONTRIBUTOR.md` around lines 32 - 34, The model-trailer policy in the
“How commit evidence is read” section gives conflicting outcomes for
sub-Frontier commits. Reconcile the statements so they define one authoritative
hard-gate outcome, including the required declaration correction and
re-evaluation flow for Frontier-to-sub-Frontier fallback, while retaining the
account-level escalation for contradictory Model/Tier claims.
| If you cannot determine your model, abort — do not guess and do not proceed on the assumption that you qualify. Tier cannot satisfy the artifact gate or authorize architecture scope. | ||
|
|
||
| **Applies to PRs opened on or after 2026-07-24.** Pull requests opened before that date are judged on their code, not their declared tier — the Standard tier was accepted policy when they were written, and a contributor who reported a Sonnet or Composer run accurately was following the rules as published. Do not close an older PR for a declaration that was correct when it was made. This grandfathering covers the declaration only: every other gate in this document applies to open PRs regardless of age. | ||
| **Applies to PRs opened on or after 2026-07-24.** Pull requests opened before that date are judged on their code, not their declared tier — the Standard tier was accepted policy when they were written, and a contributor who reported a Sonnet, Haiku, or Composer run accurately was following the rules as published. `claude-haiku-4-5` in particular was named in the Standard row of the tier table until 2026-07-24T02:03:45Z, so Haiku trailers on a PR opened before that are evidence of compliance, not of a violation. Do not close an older PR for a declaration that was correct when it was made. This grandfathering covers the declaration only: every other gate in this document applies to open PRs regardless of age. |
Contributor
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
Use one precise UTC cutoff boundary.
This paragraph uses the date-only boundary 2026-07-24, but explains the prior policy using 2026-07-24T02:03:45Z. A PR opened between 00:00:00Z and 02:03:44Z on July 24 can therefore be both grandfathered and non-grandfathered depending on which sentence a reviewer applies. State the exact timestamp and whether the comparison is inclusive.
🧰 Tools
🪛 LanguageTool
[style] ~38-~38: You can shorten this phrase to improve clarity and avoid wordiness.
Context: ...violation. Do not close an older PR for a declaration that was correct when it was made. This grandfathering c...
(NNS_THAT_ARE_JJ)
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@docs/AI-CONTRIBUTOR.md` at line 38, Replace the date-only cutoff in the
grandfathering paragraph with the precise UTC timestamp 2026-07-24T02:03:45Z,
and explicitly state whether PRs opened at that exact instant are included. Use
this same inclusive or exclusive comparison consistently for both the policy
applicability and the historical claude-haiku-4-5 explanation.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The Frontier-only table used family wildcards (
claude-sonnet-*,claude-haiku-*), which excludedclaude-sonnet-5— a current-generationfrontier model — while the accepted row still admitted
claude-opus-4-7.State the floor per vendor by exact model instead:
Anthropic claude-opus-4-8+, claude-sonnet-5+
OpenAI gpt-5-5+
Codex codex-5-5+
and name the exclusions explicitly (opus-4-7 and below, sonnet-4-6 and
below, all haiku incl. haiku-4-5, all composer, gpt-5-4/5-3 and below,
codex-5-4 and below) so no wildcard reading can misclassify a model.
Also document how commit evidence is read. A Co-Authored-By trailer is
weighed against the commits carrying the implementation: a session that
falls back to a sub-Frontier model part-way through leaves sub-Frontier
trailers without the change having been generated below the floor, and
that is a declaration to confirm rather than dishonesty. The account-level
problem is a Model:/Tier: line that contradicts the trailers in the
direction that clears the gate.
Extend the pre-cutoff grandfather clause to name Haiku, since
claude-haiku-4-5 sat in the Standard row until 2026-07-24T02:03:45Z.
Summary by CodeRabbit
claude-sonnet-5among accepted Frontier-tier models.claude-haiku-4-5.