test(repo): concentrate regression coverage

Merged
VI
22d ago
+279-3499

Summary

The prototype suite repeats numerical, schema and rendering cases across layers. Concentrate it on representative behavior, complete calculation workflows and failure recovery, removing 3,220 net lines of test code and guidance.

- TypeScript: 893 tests; Python: 120 tests
+ TypeScript: 559 tests; Python: 107 tests

Consolidate evaluator checks, reduce rendering combinations, prune report and metadata permutations, and run each declared function through the downstream pipeline once while checking every authoring spelling. Preserve independent numerical references, concurrency, publication recovery and installed-package checks. Document test ownership and criteria for adding cases.

Based on #63 (codex/html-report-workflow) so the diff contains only the test reduction.

Evidence

  • Before: npm test passed all 1,013 tests; integration tests took 26.58 seconds locally.
    After: npm test passed all 666 retained tests; integration tests took 14.21 seconds in the final local run.
  • npm run lint, npm run typecheck, a final root TypeScript check and git diff --check passed.
  • In a temporary copy, the retained numeric suite passed unchanged and caught all three deliberate faults: multiplication changed to addition, swapped atan2 operands and an exclusive tolerance boundary.
  • Review follow-up retains ordinary pi parameter semantics, ambiguous metadata rejection with binding preservation, and malformed Python-export arity in concentrated tests. The title-overflow test uses a wider viewport to remove the Ubuntu font-dependent assumption while preserving both overflow assertions.
  • Separate isolation and installed-package gates were not rerun locally.

Merge Danger

Door: two-way

The commit can be reverted. Changes are limited to tests and development guidance.

Blast Radius: verification

This intentionally reduces distinct rare numeric and legacy-metadata edge-case coverage as well as duplication. Passing the smaller suite does not establish equivalent regression detection. The three fault-injection probes are a sanity check, not a mutation-coverage score.