Detect stagnation in repeated repair attempts

[iKwesi]

Problem

SpecForge currently halts after bounded attempts, but it does not yet detect repeated non-progressing retries explicitly.

Why it matters

A strong orchestration engine should be able to recognize when it is effectively trying the same fix over and over again and stop with a clear reason instead of merely exhausting a counter.

Proposed solution

Implement stagnation detection for bounded execution loops.

Detection should be driven by evidence such as:

• near-identical code diffs across attempts
• unchanged failing test sets
• repeated critic findings with no material progress
• repeated error signatures or command failures

Acceptance criteria

• The loop can classify a run as stagnating before the maximum attempt budget is exhausted.
• Stagnation detection is evidence-backed and deterministic.
• Halted results distinguish stagnation from ordinary max_attempts_exhausted behavior.
• Tests cover repeated-diff and repeated-failure scenarios.

Related phase

Future / Research

Related subsystem/component

critic and Ralph loop controller

Related artifacts or commands

criticRalphLoop; task execution results; critic results