Improve RLCR manual-gate and stagnation handling

[xiaoxiao0406]

Summary

A completed RLCR session showed strong review-driven progress during implementation-heavy rounds, but weaker behavior around manual approval gates. The loop accumulated repeated hold rounds, and one later escape path treated lack of response as assumed approval, which proved fragile when user feedback later rejected an artifact.

Observed methodology patterns

• Productive implementation rounds had concrete deliverables, focused tests, and reviewer findings that were specific enough to fix in the next round.
• Repeated manual-gate hold rounds preserved process compliance but did not advance the underlying goal.
• Review verdicts sometimes blurred the distinction between being compliant with a gate and making actual goal progress.
• Stagnation detection eventually triggered, but only after several low-progress rounds had already accumulated.
• Assumed approval restored momentum locally, but increased downstream rework risk.
• Long repeated summaries made it harder to see what changed between hold rounds.

Suggested improvements

1. Add dual verdicts to every review: process compliance and goal progress.
2. Require each round contract to declare its type: implementation, review-fix, analysis, hold, or recovery.
3. Add a passive-hold limit for manual gates; after the limit, require explicit escalation or stop.
4. Replace assumed approval with explicit decision states: approved, rejected, needs more evidence, proceed with known risk, and blocked waiting for owner.
5. Track blocker age and repeated-round count in the working state, with a required exit condition.
6. Use compact reviews for no-change hold rounds instead of full implementation-style reviews.
7. Require reviewer findings to include concrete verification methods.
8. Label downstream work that depends on an unresolved gate as reversible, independent, or blocked.
9. Use delta-first summaries for repeated blockers: what changed, what remains blocked, owner, next options.
10. Capture process lessons for recurring methodology failures, not only technical issues.

Expected benefit

These changes should preserve the strengths of RLCR during implementation while making manual gates and stagnation easier to manage. The main goal is to prevent repeated low-progress rounds and avoid silently converting missing approval into approval.