Synergy Score Audit

[Doberjohn]

Description

Targeted audit of 2-3 synergy rules where scoring confidence is weakest. Validate that assigned scores are backed by solid reasoning, not arbitrary.

Scope

This is NOT an exhaustive audit of all rules. Identify the 2-3 rules where scores feel least justified and audit those.

Process

• Review all current rules and identify 2-3 with weakest scoring rationale
• For each audited rule:
  • Research game mechanics and meta relevance
  • Compare score assignments against actual gameplay value
  • Check if score tiers (Perfect/Strong/Moderate/Weak) align with player experience
  • Document scoring rationale (why this score, not higher/lower)
• Adjust scores if audit reveals misalignment
• Update CLAUDE.md score tables and rule documentation
• Rebuild precomputed synergies if scores change

Phase

Phase 2 (weeks 3-4) — benefits from Singers/Ramp being done first

Notes

• Community voting will provide ongoing calibration post-launch
• This audit catches the most egregious misscores before users see them
• Focus on rules where the scoring feels most "made up"

[Doberjohn]

Note from #210

The pair_scores view now includes accuracy_sentiment — a -1.0 to +1.0 signal from community votes indicating whether the engine's score is too low (positive) or too high (negative).

Once voting data accumulates (post #211-#214), this field can inform the score audit: rules where accuracy_sentiment consistently trends negative across many pairs may be systematically over-scoring.