[polewskm]

This PR adds the means to run the remaining evals without a local Bedrock key and fixes the smoke eval's must-not failure on main:

• Eval workflow (eval.yml): workflow_dispatch only, in the code-review environment. Choice inputs set the subset (smoke or re-review), verification, up to 2 runs, and a cap of $5, $10 or $25 per run. Inputs reach the shell only through env, and outputs upload only after the same secret scan the smoke eval uses. This unblocks Run the re-review evals and record what they show #123, Measure synthesis at a lower effort or on Sonnet #135 and Add a verification stage that proves or refutes each blocker and should-fix #41's on/off comparison.
• Synthesis knobs for Measure synthesis at a lower effort or on Sonnet #135: CODE_REVIEW_TIERS=synthesis=<tier> and CODE_REVIEW_SYNTHESIS_EFFORT. The production defaults don't change: synthesis stays on Opus with no effort sent.
• Decisions rule: reviewers no longer question a decision the PR adds because of a scope or limit it states. That question is what flagged accepted-no-backfill in smoke run 41244686633 (Run a cost-capped smoke eval when a prompt, charter or model changes #52's gate).
• Parse C# with a real grammar #122: a measurement found the brace scanner failing closed on 0 of 3,258 NCodeGroup C# files, and on 0.23% of dotnet/runtime files (all from #if branches), so the scanner stays. The measurement found one real bug, a verbatim string starting with "" read as a raw string, which this PR fixes.

Closes #122
Refs #52
Refs #123
Refs #135
Refs #41
Refs #57

[ncode-review]

🤖 NCode Review at af9d1c5: 3 new findings on the changed lines.

[ncode-review]

_{Superseded by the review at 43eef8c.}

[polewskm]

/review

[ncode-review]

🤖 Review complete

Verdict: ✅ Approve — Re-review of what changed since the last review at af9d1c5: 8 hunks in 5 files and changes taken back in 4 files, by correctness only. Earlier findings: F1 fixed, F2 fixed, F3 fixed. No further findings above nits.

1 reviewer run — 0 findings (run details)

Reviewer	Model	Findings	Top
correctness	sonnet-5-5	0	🟢

_{Trial: findings are posted as review threads and in the check run. $0.05 on Amazon Bedrock, 11s, at 43eef8c. (how to interact)}

📋 Outcomes

Merged at 43eef8c: 2 fixed, 1 dismissed, 0 acknowledged, 0 ignored.

Finding	Reviewer	Severity	Outcome
F1	rule-check	🟠 Should fix	👎 dismissed
F2	hygiene	🟡 Nit	✅ fixed
F3	rule-check	🟡 Nit	✅ fixed