Literate AI

Repository documentation

Test-suite audit (October 2026)

The suite had grown to 676 modules and about 6,732 test functions, mostly unit and contract tests under tests/unit. Full CI took hours, and some macOS cells hit GitHub's six-hour job limit (#13). The release owner directed a suite built around smoke and end-to-end tests, with much less emphasis on unit tests.

Every module was classified against that goal, then deleted or trimmed:

Decision Modules
E2E (kept, trimmed to strongest scenarios) 67
Smoke (kept, trimmed to a few high-value cases) 113
Critical (kept, minimal fail-closed cases) 148
Merge (reduced to the cases worth keeping) 57
Delete (unit, contract, mock-replica or duplicate coverage) 291

380 modules remain, in tests/e2e, tests/smoke, tests/critical and tests/conformance. Shared fixtures moved to tests/support, so test modules no longer import each other. Expensive fixtures (repositories, native SDK builds, initialized projects, cargo crates) are built once per class and copied per test.

test-suite-audit.json records, for each original module, its decision, its test count before the audit, the tests kept, where it lives now, and the product behavior it protects (or why it had none). CONTRIBUTING.md describes the resulting policy for new tests.

Remaining known costs:

Read the Markdown sourceConference homePublished 1.3.0 · Patch 1.3.1 in progress · Development 1.4