mix lemon.feedback (lemon_router v0.1.0)

View Source

Inspect and analyze routing feedback collected from past runs.

Routing feedback records outcome and duration per task fingerprint. Stats are read-only and do not affect live routing.

Subcommands

mix lemon.feedback stats
  Store-level summary: total records, unique fingerprints, date range,
  and the configured min_sample_size threshold.

mix lemon.feedback list
  All fingerprints with confidence annotation.

mix lemon.feedback list --workspace KEY
  Filter by workspace key (3rd segment in fingerprint key).

mix lemon.feedback list --family FAMILY
  Filter by task family: code, query, file_ops, chat, unknown.

mix lemon.feedback inspect KEY
  Full aggregate for a single fingerprint key.

Options

--workspace KEY    Filter list by workspace path
--family FAMILY    Filter list by task family
--since DAYS       Only include entries seen in the last N days
--verbose, -v      Extra detail

Confidence levels

Each fingerprint is annotated with a confidence level based on sample size and historical success rate:

HIGH         success_rate >= 0.8 and total >= min_sample_size
MEDIUM       success_rate >= 0.5 and total >= min_sample_size
LOW          success_rate < 0.5  and total >= min_sample_size
INSUFFICIENT total < min_sample_size (default: 5)

Confidence is only meaningful at or above min_sample_size. Recency matters: stale fingerprints (from expired data) are excluded when --since is used.

Notes

Routing feedback is recorded when the routing_feedback feature flag is enabled. If no data is found, confirm the flag is on and that runs have been finalized.

Exit codes

0  Success.
1  Error.