Release contract needs dataset-role classes: national-default checks reject valid non-default local-area artifacts

[MaxGhenis]

Finding (from publishing the first non-default artifact)

--no-latest (#397) made the pointer safe, but the release-contract validator remains hard-wired to the national certified default: it requires the US critical-target set (premium tax credit, taxable Social Security, itemized/SALT/medical deductions, …) in calibration_diagnostics.json and enumerates fiscal_target_sources against a national family expectation. The Build L ACS local-area artifact — calibrated deliberately to state/CD population marginals + state administrative families — can never satisfy that contract, by design rather than by defect.

Publishing it therefore had to route around the tracked path (direct Hub commit + tag, no pointer, all applicable gates passing and recorded). That's the wrong long-term state: non-default artifacts deserve a contract, just not the national one.

Proposal

Add a dataset-role dimension to the contract:

• release_manifest.json already carries dataset_role (non_default_local_area on the Build L artifact) — key validation off it.
• national_default (current behavior, unchanged): full US critical-target checks, national family enumeration, pointer update allowed.
• local_area / research (non-default): require gates object + pass, calibration diagnostics shape (targets list ≡ n_targets, surface + registry refs, options), source coverage keyed to the families actually calibrated (derived from the diagnostics, not a national list), reviewed-limitations register non-empty or explicitly empty, and forbid the pointer op entirely for these roles (defense in depth beyond --no-latest).
• publish CLI: refuse --no-latest absent on non-default roles and refuse pointer updates for them even if requested.

First consumer to migrate: populace-us-2024-buildl-acs-local-36de5d9a-20260712T104640Z (published directly 2026-07-12; files under releases/<id>/, tagged, pointer untouched and verified).