Date: 2026-08-29 Issue: https://github.com/MSKazemi/idkmesh/issues/15
The project owner requested parallel, professional resolution of ten unclaimed issues, including remote branches, pull requests, verification, and eventual integration. This workstream was assigned issue #15 and was explicitly limited to opening a non-draft pull request rather than merging it directly.
Current main already contained WorkUnit v0.2, historical v0.1, worker result
and independent verification contracts, deterministic validators, A2A/MCP
lossless round-trip bindings, and a protocol mapping matrix. Replacing or
expanding those schemas would have duplicated accepted semantics and risked
breaking historical artifacts.
The missing bounded delta was:
Keep WorkUnit v0.1 and v0.2 unchanged. Add a separate v0.1 benchmark contract, a deterministic read-only validator/aggregator, four v0.2 examples, focused tests, documentation, and CI coverage.
The committed fixture is labeled synthetic_fixture. Its observations prove
only that validation and aggregation are reproducible; they do not support a
claim about which decomposition strategy performs best. Real controlled worker
runs must use evidence_class: observed and preserve raw observations and
provenance.
The validator resolves referenced files only within the repository, validates formal units against WorkUnit v0.2, rejects unknown dependencies and cycles, and never executes declared commands. It has no dispatch, acceptance, write, push, or integration authority.
The change gives contributors a small, inspectable way to add comparable real runs and shows four common roles without requiring them to infer the full repository architecture. The evidence label is designed to prevent example numbers from being mistaken for scientific findings.
Run the five arms with independently assigned workers, controlled context, shared hidden tests, preregistered assignment, and recorded human integration time. Replicate across multiple task classes before changing the WorkUnit core.