Skip to content

Bench gap: filter scripts measure a cold single commit, never a keystroke sequence #489

Description

@blove

Found during #487's columnar-cache arc (see docs/superpowers/specs/2026-08-24-columnar-verdicts-results.md, 'Decision: reverted'): the S2 filter scripts apply ONE filter commit against a cold engine, so any design that amortizes cost across repeated commits (caches filled on first use, warm-path optimizations) is invisible to them — the fill IS the measured interaction. A keystroke-sequence script (N successive narrowing filter commits, reporting per-commit latency) would measure the filter-as-you-type case the product cares about, and is the prerequisite for ever re-evaluating the reverted columnar verdict cache (preserved in git history at 73f1ae24).

🤖 Generated with Claude Code

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions