[Epic] Cross-platform assurance, test infrastructure, and performance

[plx]

Parent

#26

Objective

Turn the remediation fixes into durable evidence: real behavioral CI on every supported OS, generated filesystem round-trip properties, parser fuzzing, hermetic tests, direct library coverage, branch-aware coverage reporting, and explicit performance/resource bounds.

Why this blocks release

The current 127 tests all pass but miss the core false-success cases. Full tests run only on Ubuntu; macOS and Windows merely build. The suite has no property-generated filesystem invariant, fuzz targets, or branch coverage, and one integration test walks the mutable checkout. Parser hierarchy and placeholder verification show measured quadratic scaling.

Workstream order

1. Run existing behavior on all supported operating systems and make fixtures hermetic.
2. Land defect-specific regressions with the implementation tickets.
3. Add generated round-trip/mutation properties and parser fuzz targets after contracts stabilize.
4. Add platform conformance, coverage/mutation evidence, and performance baselines.

Completion criteria

• Linux, macOS, and Windows execute the supported behavioral suite.
• Generated supported filesystem trees round-trip exactly and mutations are detected.
• Parser fuzzing covers historical reproductions and finds no panic, bypass, timeout, or uncontrolled allocation.
• Tests are hermetic and retained public wrappers receive direct coverage.
• Critical modules meet the reassessment coverage/effectiveness gates.
• Performance scaling and resource limits meet the reassessment playbook.

Relationships

• Parent epic: [Epic] Production readiness and formal v0.2.0 release #26
• Native child and dependency links will be added after leaf creation.
• Property and fuzz tickets depend on the stable parser/generator contracts; platform conformance depends on finalized symlink and identity policies.

Non-goals

A coverage percentage alone does not close this epic. The suite must demonstrate that blocker regressions are red on the audited baseline and green on the candidate.