Back to Plate

Coverage Priority Map Post Package Sweep

docs/plans/2026-03-23-coverage-priority-map-post-package-sweep.md

53.0.59.7 KB
Original Source

Coverage Priority Map Post Package Sweep

Goal

Run fresh repo coverage, sync against the March 17 docs plus the March 22-23 package sweep, then rank the remaining non-React test work by actual value.

Coverage Run

  • Command:
    • bun test --coverage --coverage-reporter=lcov --coverage-dir=.coverage-repo-2026-03-23a --reporter=dots
  • Result:
    • 2486 pass
    • 443 files
    • 2.09s
  • Artifact:

Scoring Rules

  • Scope is packages/*/src/**.
  • Package score is the sum of the top 8 remaining non-React file scores.
  • File score is 0-10.
  • /react, test files, barrels, and type-only files score 0.
  • React-ish and browser-only buckets outside literal /react are also excluded for this pass:
    • udecode/react-*
    • udecode/cmdk
    • playwright
  • High scores go to deterministic transforms, queries, parsers, serializers, geometry helpers, and real plugin contracts.
  • Tiny wrappers and already-swept packages get penalties on purpose.
  • Recent-pass penalties were applied for March 17, March 22, and March 23 work so leftover uncovered crumbs do not outrank fresh central seams.

Sync Notes

These lanes are already substantially covered and were intentionally pushed down:

  • March 17: selection, focused non-React core, focused docx-io
  • March 22-23: yjs, list-classic, suggestion, docx, ai, layout, list, code-block, autoformat, csv, comment, dnd, combobox, link, date, code-drawing, emoji
  • Table still has raw uncovered surface, but it was the recent push and should not be the next package again.

What I Would Do Next

This order is the final value ranking, not a blind sort on raw package score. I used the score map, then pushed down recent-pass leftovers and weak component-ish lanes.

1. slate

2. diff

3. utils

4. core

5. floating

6. markdown

7. indent

8. basic-styles

9. math

10. toc

Second Tier

  • udecode/utils
    • real helper coverage, but lower framework leverage than the top group
  • udecode/depset
    • a few honest utility seams, not a major package lane
  • caption, tabbable
    • small deterministic slices if you want tiny wins instead of a core package

Explicitly Not Next

  • table
    • raw score is still decent, but reopening table right after the table push would be dumb
  • cursor
    • mixed with component-ish seams, so weaker value than floating if you want geometry/selection coverage
  • resizable
    • mostly utility dust plus component code
  • test-utils
    • not product behavior coverage
  • recently swept packages with leftover score-1 to score-3 files
    • do not reopen them unless a specific bug points there

Ranking Snapshot

RankPackageScoreStatus
1slate66candidate
2utils64candidate
3diff63candidate
4core56candidate
5floating51candidate
6udecode/utils49candidate
7table40candidate
8markdown40candidate
9cursor40candidate
10resizable38candidate
11basic-styles36candidate
12indent32candidate
13test-utils31candidate
14basic-nodes27candidate
15udecode/depset25candidate

Full Data