fix(context): compute require_citations gate after the max_chars budget

[seekmistar01]

Summary

Closes #174

build_context_pack computed the uncited claim list (and thus the require_citations gate failure and the uncited_items reported in quality) over the full item list, then the max_chars budget popped tail items. So a ContextPack could come back ok=False with uncited_items=[...] naming claims that aren't present in pack["items"] — failing the pack for citation reasons driven by items the consumer never received.

Change

• src/vouch/context.py — move the require_citations / uncited computation to after the max_chars budget step, so it only considers the items actually returned.
• tests/test_context.py — test_require_citations_only_considers_returned_items: with max_chars + require_citations, asserts every id in quality.uncited_items is in the returned items.

Why it's safe

• Behavior is unchanged when there is no max_chars budget (nothing is popped, so pre- vs post-budget computation is identical).
• When a budget drops items, the gate now correctly reflects only the returned pack.

Verification

• pytest tests/test_context.py — 6 passed (existing + the new regression test). ruff check clean.

[coderabbitai]

Warning

Review limit reached

@seekmistar01, we couldn't start this review because you've reached your PR review rate limit.

More reviews will be available in 55 minutes and 4 seconds. Learn how PR review limits work.

Your organization has run out of usage credits. Purchase more in the billing tab.

⌛ How to resolve this issue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

We recommend that you space out your commits to avoid hitting the rate limit.

🚦 How do rate limits work?

CodeRabbit enforces hourly rate limits for each developer per organization.

Our paid plans include higher PR review limits than trial, open-source, and free plans. In all cases, reviews become available again over time. During sustained high-volume PR review activity, CodeRabbit may temporarily slow when the next review becomes available.

Please see our Fair Usage Limits Policy for further information.

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 4035a033-67a4-47b9-8d48-0cf27feb8ad7

📥 Commits

Reviewing files that changed from the base of the PR and between 3beb821 and 4b5f8cc.

📒 Files selected for processing (2)

• src/vouch/context.py
• tests/test_context.py

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[plind-junior]

Review

Summary: The PR correctly moves the require_citations gate to after the max_chars budget step, fixing the root cause described in issue #174, and adds a focused regression test. The change is minimal, safe, and well-reasoned.

What works

• src/vouch/context.py:119–121 — Moving the uncited computation to a list-comprehension after items.pop() is the right fix and the comment explains the "why" clearly.
• src/vouch/context.py:130–133 — The downstream if uncited: / failed.append("require_citations") block is untouched and works correctly with the new placement; uncited is now always populated after budget trimming.
• tests/test_context.py:41–58 — The regression test directly asserts the invariant (uncited_items ⊆ returned), pins a max_chars=80 budget tight enough to force drops, and guards against a vacuous pass with the assert pack["quality"]["uncited_items"] pre-condition check.

Suggestions

• [non-blocking] src/vouch/context.py:104 — The uncited: list[str] = [] initialisation on line 104 is now dead (it gets replaced unconditionally on line 121 whenever require_citations is True, and the variable is never read before that point). Consider removing the initialiser or inlining it as uncited: list[str] to avoid the misleading empty-list default. It doesn't affect correctness, but it leaves a stale artefact from the old flow.
• [non-blocking] tests/test_context.py:53–54 — max_chars=80 with 20 claims of ~45-char summaries reliably forces drops today, but it depends on the MockEmbedder returning all 20 claims in the result set. If the embedding mock or the default limit=10 changes, the test could silently become vacuous (the pre-condition assert on line 57 would catch it, but only at runtime). Consider passing limit=20 explicitly so the intent is self-documenting and the test stays robust to future limit default changes.

Verdict

Approve — Minimal, correct fix for the exact issue described; regression test is in place; no out-of-scope changes.

[plind-junior]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 🚀

Reviewed commit: 4b5f8ccd7e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".