Show whether the packet was useful, not just smaller.

Signals emitted per packet.

• Estimated input tokens saved and context reduction.
• Heuristic useful-context density, plus labeled eval scores when a gold set exists.
• Risk level, budget pressure, warnings, and redaction count.
• Included files by kind: source, spec, test, instruction, and docs.
• Packet freshness and previous/current diff information.

Manager summary.

Generate local Markdown and JSON summaries for team-level review, demos, and launch storytelling.

npm run summary
npm run metrics
npm run eval -- --workspace examples/demo-workspace --cases ../eval/gold-set.json

Org-level rollups later.

The metrics shape is intentionally aggregate-friendly: teams can collate JSONL outputs across repos without shipping source code into a central dashboard.

Observability without opacity.

The same packet that produces metrics is inspectable by the developer, so token savings and risk signals are explainable rather than mysterious.