evaluation-justification-before-score-prevents-anchoring-bias
Requiring written justification before assigning a numeric score prevents anchoring bias in evaluations — the score adjusts to match the justification rather than the justification being retrofitted to a gut-feel score. A pairwise double-pass (evaluate A vs B, then B vs A independently) catches directional bias where the first option evaluated receives a structural advantage. This applies to all scored evaluations: skill quality, MCP tool quality, PR review scores.
Related
- mcp-tool-namespace-discipline-fully-qualified-server-prefix
- mcp-tool-output-offload-threshold-2000-tokens
- openspace-mcp-global-registration-in-root-claude-json
- clawteam-openclaw-multi-agent-swarm-evaluation
- precompact-artifact-trail-must-be-verbatim-index
- evaluation-rigor-justification-before-score-prevents-anchori
- notebooklm-mcp-cli-consolidated-tool-pattern