Analysis output

Placeholder. The one output of analyze/, read by both ../05-dashboards/ (people) and ../06-optional/mcp-server/ (agents). Same numbers in both.

Summary

• Configurable per deployment: which metrics (from ../05-dashboards/metrics.md), period, grouping (team, repo, model, harness, intent).
• Each value carries its definition, period, sample size and the session ids behind it.

Outliers

• Per metric and group, compared against that group's own baseline (robust z-score or percentile; thresholds as named constants).
• Each outlier: metric, group, value, baseline, sample size, top contributing sessions.
• Examples: a session 10x the team's median spend, an MCP server whose error rate jumped, a repo where spend per merged change doubled.

Adoption and checklist

• Adoption inventory (../02-measure/adoption.md) and deployment checklist state (../01-deploy/checklist.md) are part of the output, so operator and agent see the same thing.

Pitfalls

• Small groups give noisy baselines: below a minimum sample size, report but do not flag.
• Weekly cycles and holidays: compare like with like.
• Per-person outliers follow the privacy defaults in ../02-measure/METHOD.md.