pairv2-llm-driven-philosophy

name	pairv2-llm-driven-philosophy
description	Pairv2 implementation philosophy: LLM-decided outcomes, fail-soft file handling, and evidence-first promotion
type	policy
scope	project

Pairv2 LLM-Driven Philosophy

Core Rule

Pairv2 should fail only for bounded safety reasons (timeout/max cycles) or verifier judgment, not because a specific file path was missing.

Decision Hierarchy

1. Verifier LLM verdict is the primary quality signal.
2. Non-LLM checks provide notes and evidence, not hard overrides.
3. Missing optional inputs (design docs, user-provided contract paths) must degrade gracefully to generation or inference.

Anti-Brittle Rules

• Do not hard-fail solely because a provided path does not exist.
• When provided left/right contract files are missing, fall back to LLM generation.
• When design doc path is missing, continue without design doc context and add a note.
• Prefer reproducible evidence bundles (reports/artifacts/logs) over path heuristics.

Practical Checklist

When editing pairv2:

1. Ask: "Will this block execution because a file is missing?"
2. If yes, replace with note + fallback path.
3. Ensure verifier still gets enough context to decide PASS/FAIL/NEEDS_HUMAN.
4. Add tests for missing-file fallback and note emission.