roadmap(#212): engineering complete; Prove-It Gate deferred

[BaseInfinity]

Summary

Mark ROADMAP #212 (Local-Max E2E shepherd) as engineering-complete with the Prove-It Gate explicitly parked.

Maintainer call 2026-05-05: "just defer prove it who cares" — pragmatic decision after audit confirmed:

• ✅ Simulation on Max via claude --print (Lock live-session auto-setup as the default human path #228, v1.59.0)
• ✅ Evaluator on Max via EVAL_USE_CLI=1 (Lock live-session auto-setup as the default human path #228, v1.59.0)
• ✅ Orchestrator tests/e2e/local-shepherd.sh --compare-baseline (docs: GPT-5.5 preferred for cross-model review + README quickstart #230)
• ✅ Weekly+monthly workflow ports (roadmap(#212): Option 1 — de-require e2e, shepherd becomes advisory #231 phases 1–4, v1.50.0–v1.55.0)
• ✅ Docs reference local-shepherd as default path

Bonus correction: Codex cross-model review listed as ~$3-5/PR (raw-API billing). Verified via codex login status → "Logged in using ChatGPT" (subscription auth). Real cost: $0 marginal. Roadmap updated to reflect.

Net per release: $20+ (old API burn) → $0.

Why defer Prove-It: would require ~$20 one-time API spend to pair-run N=15 trials for 95% CI overlap. Engineering signal works fine in practice; reopen if local scoring drifts vs known-good baselines. Original scope archived in collapsed <details> block for history.

Test plan

• CI green (validate job)
• Diff is roadmap-only (no code changes)