bug: session-schema adapter defaults rework_cycles to 0 for SODA sessions

[decko]

Problem

The session-schema adapter reads meta_raw.get("rework_cycles", 0) but SODA's meta.json never writes a rework_cycles field. SODA records rework as generation > 1 on per-phase metadata in meta.json.

Out of 30 SODA sessions in .soda/, 7 have implement.generation=2 — all are reported as rework_cycles=0.

This directly corrupts two key metrics:

• first_pass_success_rate — reports 1.0 when it should be ~0.77
• self_correction_rate — reports N/A instead of computing actual resolution rates

Root cause

src/raki/adapters/session_schema.py line ~100:

meta_raw.get("rework_cycles", 0)

SODA meta.json stores rework as phases.<phase>.generation (1 = first pass, 2+ = rework). Example from ticket 131:

{
  "phases": {
    "implement": {
      "status": "completed",
      "generation": 2,
      ...
    }
  }
}

Suggested fix

When rework_cycles is absent from meta.json, compute it from phase metadata:

rework_cycles = max(phase.get("generation", 1) for phase in phases.values()) - 1

Impact

All SODA sessions evaluated with raki run since v0.9.0 have incorrect first_pass_success_rate and self_correction_rate values.

Related

• Alcove adapter: hardcoded rework_cycles=0, findings=[], and single phase #176 — adapter metric improvements (shipped in 0.9.0)

Acceptance Criteria

• When rework_cycles is absent from meta.json, adapter computes it from max(generation) - 1 across all phases
• When rework_cycles is present, it takes precedence (backward compatibility)
• SODA sessions with generation=2 report rework_cycles=1
• first_pass_success_rate reflects actual first-pass failures
• Tests cover both paths: explicit rework_cycles field and computed from generation

Task 1: Write failing tests for generation-based rework computation

Files: tests/test_adapters.py

1. Write failing test: meta.json without rework_cycles but with phases.implement.generation=2 → rework_cycles should be 1
2. Write failing test: meta.json without rework_cycles and all phases generation=1 → rework_cycles should be 0
3. Write failing test: meta.json with explicit rework_cycles=3 (backward compat) → should use the explicit value regardless of generation fields

Task 2: Implement generation-based rework computation

Files: src/raki/adapters/session_schema.py

1. In the meta loading logic, after reading meta_raw:
   • If rework_cycles is present in meta_raw, use it (backward compat)
   • Otherwise, look for phases dict in meta_raw, compute max(phase.get("generation", 1) for phase in phases.values()) - 1
   • Default to 0 if no phases found
2. Verify all 3 failing tests now pass

Task 3: Verification

1. uv run pytest tests/ -v -m "not slow" — all tests pass
2. uv run ruff check src/ tests/ && uv run ruff format src/ tests/
3. uv run ty check src/raki/