Files
ARCADIA-Studio/harness/experiments/diagram-readability-20260929/source-review.md
T

1.9 KiB

Independent preparatory review

Read-only review of the new Rust probe, fixed inputs/answer keys, experiment contract and production projection/picking code. No compiler, probe or GUI was run by the reviewer. No remaining source blocker was identified before execution.

The author addressed these consequential findings:

  • Port-owner mapping adds five valid relationships to the representative view: its complete inventory is twelve, rather than seven. Kind, semantic endpoints, ports, projected endpoints and trace identity are checked independently.
  • Port occlusion follows the actual square half-width max(10/zoom, 10) world units, with a corner control; association-end label rectangles also precede ordinary line picking and are included.
  • Query candidates count semantic edges separately from their matching segments. Empty ordinary labels do not occlude a point merely because bounds remain.
  • Historical reproduction requires both declared real identities, zero distance, no occluder and the recorded common span. Declared source provenance requires actual build/fixture verification before a current-router claim.

Admission, warning baselines and current routing remain execution gates. Two flat scenes cannot establish nested/folded behavior; Traceability removes visual parents. Geometric samples do not establish native selection, readability or global inaccessibility. Read-only Project equality does not verify edit transactions, Undo/Redo or save/reopen. Drawing a new relation requires a separate declared post-change answer key and Project delta.

The next workflow should trace, deliberately select, repair one label or bend, then undo in the complete scene, with unique labels/unshared spans as controls. Creating a valid relation, refusal and cancellation are separate tasks. A chooser that prints correct endpoints cannot substitute for unaided diagram readability.