Add Shipwright v2 benchmark harness

[EdgeCaser]

Summary

• Deterministic benchmark harness in scripts/run-benchmarks.mjs: loads frozen fixture scenarios, runs the existing validator, computes PASS/FAIL/DNF status, normalizes blind ratings, and emits the scoring-spec result shape
• Six frozen scenarios under benchmarks/scenarios/ with fixture artifacts covering hidden scope creep, partial-data pricing, board ambiguity, churn contradictions, weak evidence, and handoff contradictions
• All fixture artifacts on schema_version: 2.0.0; each scenario exercises a detection → resolution cycle (first pass surfaces validator issues, final pass resolves them)
• Default suite produces 4 PASS / 1 FAIL / 1 DNF; blind ratings left null until human review runs

Test plan

• node --test tests/run-benchmarks.test.mjs — 6 tests: suite execution, blind-rating normalization, DNF diagnostic scoring, scenarioIds filtering, invalid scenario loading
• node --test tests/*.test.mjs — 163 tests, all passing
• node scripts/run-benchmarks.mjs --format json — expected mix confirmed
• bash scripts/validate.sh — green

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 8d64281b84

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".

[chatgpt-codex-connector]

Determine DNF from all passes, not final pass only

This marks a scenario as DNF whenever the final artifact is unusable, even if the first pass was usable. Under the scoring spec, DNF should only occur when no artifact reaches usability; a regression in a later draft should not erase an earlier usable artifact. With the current logic, scenarios that were usable on first pass can be misclassified as DNF and their time_to_first_usable_artifact_seconds is later nulled out, corrupting benchmark outcomes.

Useful? React with 👍 / 👎.