Parent

Epic #426, Original issue #434, Plan: .local/plan/m26.1-output-filtering-improvements.md
Priority: P0 (development feedback)

Problem

No data during Phase 1-2 filter development. Cannot validate claimed 60-90% token savings or detect regressions until Phase 3.

Design

Phase 1 (P0): basic in-memory metrics with periodic logging.

pub struct FilterMetrics {
    pub total_commands: usize,
    pub filtered_commands: usize,
    pub raw_chars_total: usize,
    pub filtered_chars_total: usize,
    pub confidence_distribution: HashMap<FilterConfidence, usize>,
}

• Store in registry as Arc<Mutex<FilterMetrics>>
• Record in apply() after each filter result
• Log every 100 commands
• CLI /metrics command displays stats

Dependencies

• Requires: FilterConfidence enum (feat: FilterConfidence enum for output filter results #440)
• Blocks: test/git/clippy filters

Acceptance Criteria

• FilterMetrics struct with record() and savings_percent()
• Registry stores and records metrics
• Periodic log every 100 commands
• Unit tests for metrics recording
• Integration test: verify metrics after filtering 10 commands