Qualify parity claims against the returned macOS reference

[coneilen]

Summary

Stop treating the thirty previously unassessed rows as current visual parity. Adopt the returned macOS reference, preserve historical Windows evidence, and make the remaining qualification and publication limits explicit.

Changes

• Keep all 98 ledger identities and functional evidence. Reclassify 30 unsupported visual validations as Partial qualification gaps, not newly discovered product failures.
• Classify the source-confirmed sidebar-only Needs-you extra outside parity; actual removal needs a separate approved slice and reachable equivalent actions. The totals become 6 Validated / 89 Partial / 3 Out of scope.
• Record the exact thirty-row reference join, the project-canvas/90% labelling correction, and Windows beta26 as pre-wave-1 evidence rather than a current-head capture.
• Make the new pinned source, privacy report, local publication-hold notice and individual image clearance the canonical reference procedure. No bundle, image, hold notice or private operational data is included.
• Keep final A-E merged-slice reconciliation separate. No whole issue is closed, no product behavior changes, and no screenshot/golden/pixel-baseline CI is introduced.

Test plan

The checker below is a retained session-local documentation contract, not a new repository test. Its command is shortened to the script basename to omit local operational paths; the original RED and GREEN logs are retained separately. It checks the actual documentation and pinned baseline, not product runtime.

RED: Test-GroundTruthIntake.ps1 against the baseline checkout -> INTAKE: primary macOS source SHA missing from kickoff
GREEN: Test-GroundTruthIntake.ps1 against this patch -> 98 preserved status rows, exactly 30 joined targets, exactly 3 changed documents; 6 Validated / 89 Partial / 3 Out of scope
REGRESSION: pwsh -NoProfile -File Tools\windows\validate.ps1 -Task privacy; git diff --check; Test-GroundTruthIntake.ps1 -> privacy passed, whitespace passed, unrelated rows/statuses and all 98 ordered identities preserved

Evidence limits: the different-vendor reference spot-check opened 12 primary and 3 comparison images and checked 5 metadata negative controls; this is not a blanket privacy or parity verdict. The retained Windows bundle is the installed beta26 package, not current main. Missing current pairs remain qualification gaps. No native app, physical input, persistence, macOS build, make test or make check was run for this documentation change.

Checklist

• I have read the Contributing Guidelines
• I have signed off my commits (git commit -s) per the DCO
• Tests pass locally (make test)
• Code follows the existing style (make check)
• I added the test/contract before the implementation and observed the intended RED failure