[est. 2–8h · ~0.2–1.5M tok · ~$2–30] Optional: richer History / operational insights (text, not BI)

[djbclark]

Estimate: 2–8h focused work · ~0.2–1.5M norm. tok · **$2–30 norm. $** (wide/rough).
Optional polish — operational text only; not an onWatch-class dashboard.

Summary

Issue #6 shipped History section + learn_from_history. Optional follow-up: louder operational insights from existing snapshots (chronic underuse, late-cycle leftovers, learned burn rates) without charts/anomaly BI.

Ideas (pick any subset)

• Stronger --full History copy when learning is on
• Optional aiuse history one-shot summary (no full collect if snapshots exist)
• JSON history already exists — ensure scripts can get “what to burn from history” cleanly
• Surface blended with history (N samples) more consistently on ladder rows

Non-goals

• Charts, sparklines, SQLite BI, anomaly product (onWatch)
• Replacing LaunchAgent densification (that’s free improvement over time)

Acceptance

• Human-visible improvement on --full and/or dedicated history path
• Tests for any new CLI/report surface
• Docs: docs/history-learning.md + link from docs/next-options.md

Context

docs/next-options.md, docs/history-learning.md