Module feedback 

User-supplied feedback signals that refine stored meta-learning records.

Automated quality metrics (QualityMetric) only see the geometry of the built pipeline. They can’t tell whether actual users found query results useful. This module defines a minimal feedback primitive — one scalar signal per query — plus an aggregator that summarizes signals per corpus_id so a MetaTrainingRecord’s best_score can be blended with observed user satisfaction.

Intended flow (L3 of the metalearning ladder):

1. Deploy a tuned pipeline to users.
2. On each query result, collect a satisfaction signal (thumbs, rating, click-through, …). Map it to [0, 1] and emit a FeedbackEvent.
3. Aggregate events into a FeedbackAggregator, persisted under FeedbackAggregator::default_store_path.
4. When selecting a stored record for a new corpus, blend the record’s automated best_score with the corpus’s feedback summary via MetaTrainingRecord::adjust_score_with_feedback.

The meta-model is deliberately not retrained here — that’s a v2 concern. This module supplies the primitives; composition is up to the caller.

Structs

FeedbackAggregator

Accumulates FeedbackEvents across sessions and summarizes them by corpus_id.

FeedbackEvent

One user-supplied satisfaction signal attached to a specific query.

FeedbackSummary

Summary statistics for the feedback observed on a single corpus.