[mkarlesky]

Adds the real-Ceedling validation suites to the repo, so anyone can rerun the checks behind the 1.2.1 work. just validate <version> [suite] compiles the tests in the sidecar, then runs each suite in throwtheswitch/madsciencelab-plugins:<version> with the sidecar's Node added (sidecar/validate.Dockerfile). The suites drive the compiled CeedlingEngine with only settings, the logger, and the Problems panel stubbed, so every Ceedling command is real.

Suites (tests/validation/)

Suite	Checks	Versions
stress	Random overlapping runs, debug builds, discoveries, reparses, and cleans across four projects, with random cancellation; --shared-build-root gives them one build root	All
coverage	Exact gcovr and ReportGenerator outcomes, then coverage overlapping everything else	All (ReportGenerator checks skip on 1.0.0)
gated	Per-version command lines on same-named test files	1.2.0+
layout	Exact test build directories with two test roots; namesake counters kept	1.2.0+ (coverage 1.2.1+)
latency	Rediscovery time during a build	All, not in all

Probes (probe:fixes, probe:syscalls, probe:overlap) show Ceedling's own behavior behind the engine's choices.

The suites are TypeScript compiled with the tests, so they typecheck against the engine and pass lint with 0 findings. Docs: Development.md §5 (new; manual plan and sidecar renumbered), Architecture.md (evidence for lock-free discovery on 1.2.1, the Ceedling#1364 layouts, test seams), KnownIssues (Ceedling#1364), sidecar README.

Verification

• just validate <v> all 45 on 1.2.1, 1.2.0, 1.1.10, 1.0.0: 0 violations each, about 3 minutes per version.
• Probes and latency on 1.2.1 and 1.2.0: on 1.2.1, discovery writes nothing; on 1.2.0, it writes 10 times. Rediscovery takes 0.85 s idle and about 4.7 s during a build.
• Mutation check: an engine that clears every counter fails layout, and one that names gcov: tasks bare on 1.2.1 fails gated (27 violations).
• just test: 458 unit and 18 integration tests pass. just lint: 0 findings.
• An independent review's findings were fixed: layout coverage gated to 1.2.1; argument validation; suite-name validation; a racy executable check in stress replaced by the engine's own check, which runs under its lock; latency given a slow test so it measures; doc defaults corrected.

🤖 Generated with Claude Code