[tdamsma]

I propose more granular contributor claims to help distribute the remaining completion work. The goal is behaviour equal to the C++ reference, or demonstrably better where the reference has a confirmed defect or undefined behaviour. This PR proposes coordination changes; formal acceptance and reporting rules should also be made more explicit but that another dicussion.

Machine-generated (OpenAI Codex coordinator and agents) on behalf of @tdamsma.

Optional workflow suggestion for the project owner; draft for discussion, with complete proposed wording in AGENTS.md and README.md.

The whole-unit claim workflow gives initial ports a clear owner. As the project moves toward completing functions, dependency wiring and reference comparisons, a unit can contain several independent tasks even while it has an active PR. The current guide directs newcomers to unclaimed units and gives the first PR precedence over the whole unit, which can make useful completion work appear unavailable.

Would scoped claims be useful for this phase? This proposal lets contributors claim a specific function group, wiring path, coverage gap or demonstrated parity bug inside an existing unit while protecting work already underway.

Proposed workflow

• Initial ports retain port: <unit id>. Completion PRs use descriptive titles and identify the unit, symbols, affected paths and evidence in their bodies.
• A short scope template records intended behavior, dependencies, validation, available follow-ups and blockers. Contributors can find work beyond the unit-status checklist.
• The earliest claim has precedence for overlapping work. Independent tasks in the same unit can proceed after linking the related PR and stating the boundary; changes to the same functions or shared interfaces require an agreed split or a documented branch dependency.
• Authors push WIP or post a concrete update daily while implementing. After 48 hours without either, a contributor proposes an exact takeover scope and allows 24 hours for a response before continuing from the published WIP with attribution.
• Awaiting review or CI is not abandonment. Paused authors release their scope and link WIP; blockers identify independent work others can take. Existing active claims remain valid.

For example, one contributor could complete a missing constructor while another extends exact runner imports for a separate operation in the same unit. If both need to change the constructor interface, they coordinate that boundary first.

Adoption and tradeoffs

This can be tried on new completion PRs without renaming existing PRs or migrating plan files. Owners of existing broad claims can add the scope and available-follow-ups fields as convenient. Explicit boundaries add a little PR-body upkeep, but aim to reduce idle contributors and accidental duplicate work. The suggested 48-hour inactivity and 24-hour response windows are open to adjustment by the owner.

This PR changes contributor guidance only. It does not change correctness tolerances, implementation acceptance, unit-status accounting or CI. The completion goal above motivates this proposal; changes to the reference gate and its reporting are handled separately.

Validation: reviewed both guides for consistent claim boundaries, overlapping-work rules, inactivity/handoff behavior and preservation of existing claims; git diff --check passes. No executable code changed.

[tdamsma]

@bherbruck What do you think?