Compression bundle

A no-DB, report-only review bundle for note-improvement passes, run via kb/instructions/run-compression-bundle-on-note.md. Deliberately kept outside the kb/instructions/review-gates/ catalog so it never gets swept into --all-gates or writes review-DB freshness-baseline state.

Purpose: catch material that is true, coherent, or locally defensible but should still be compressed, folded, deleted, split, or rehomed because it does not earn its context cost.

This is separate from the normal prose and semantic bundles because the lens is edit-strategy oriented: it asks whether the note's current shape is worth keeping, not merely whether claims are true, grounded, clear, or consistent.

Gates

  • marginal-value-redundancy — true material that should not remain in its current form because its marginal value does not justify its size or placement.
  • branch-bloat — interesting side claims that weaken the main claim and should be deleted, demoted, or split.
  • detail-overhang — examples, caveats, mechanisms, or background details that outgrow their role.
  • core-claim-obscured — the strongest retained claim is buried under supporting apparatus.

Synthesis

After applying the four criteria, run high-impact simplification synthesis. It selects at most three changes that would materially improve the artifact's central claim, conceptual structure, or argumentative flow. The synthesis can combine several local findings into one structural revision, but it does not settle ambiguous meanings or rewrite the artifact.

Candidate future gates

  • split-candidate-threshold — material marked for rehoming without enough evidence that it deserves its own durable note.

2026-06-17 adjustment

Case 03 showed that the bundle correctly caught branch bloat, but the first draft over-compressed some rhetorically load-bearing material. detail-overhang now protects fair opponent concessions, short source setup, and vivid contrasts when they give the argument force rather than adding a second thesis. marginal-value-redundancy now distinguishes empty repetition from conclusions that restate a condition as a stance, warning, or operational bet.