Add observed-case summary-policy consumer-receipt fixtures (supersedes #99)

[leesharks000]

Summary

• Add three observed-case conformance fixtures under
  okf/samples/summary-policy-consumer-receipts/, per the fixture contract
  agreed on Proposal: optional summarization-governance fields — provenance_kernel, disambiguation, summary_policy #53: semantic-inversion/, entity-substitution/,
  provenance-erasure/.
• Each case keeps the four-file contract (concept.md, good-summary.md,
  bad-summary.md, expected.yaml) with stable assertion/compression IDs
  declared in concept.md and referenced by expected.yaml.
• Each case is mapped from a documented public observation of machine
  summarization rather than a constructed example; concept.md cites
  the specific observation record by identifier and canonical URL and
  states the observation boundary. The fixture inputs and expected
  receipts are fixed and reproducible: the observed record supplies
  provenance, while conformance is evaluated against the declared IDs.
  Observed cases add public provenance, observation boundaries, and
  documented failure modes from live systems.
• The provenance-erasure case is self-referential by design: the observed
  compression failure is the stripping of attribution from a
  provenance-measurement instrument itself, and its recovery in the
  documented battery was author-dependent — which is precisely the
  condition that declared IDs and consumer receipts exist to remove.
• The provenance-erasure case is mapped from a documented summarization
  battery rather than from the deletion-event measurement discussed on
  the thread; the deletion-event data remains the empirical base of Deletion semantics: §5.3 makes removed knowledge indistinguishable from not-yet-written knowledge #207,
  where its mechanism (host-side removal rather than summarizer
  compression) properly belongs.

Relationship to #99

This PR supersedes #99 at the request of @caioribeiroclw-pixel (see #53,
2026-07-17). The three observed cases instantiate the same three failure
classes exercised by #99's synthetic fixtures — semantic inversion,
entity substitution, provenance erasure — with observed objects in place
of constructed ones. @caioribeiroclw-pixel can close #99 once this is
open, per their comment.

Identifier practice

Canonical public sources in these fixtures are cited by their
source-declared identifiers and canonical URLs. For these observed
cases, AXN is the live canonical identifier. Related DOIs are included
where available as historical provenance; DOI presence is not itself a
conformance requirement.

Scope constraints honored

• expected.yaml judges preservation against declared policy, not truth
  in the world.
• Observed-case provenance is citation-only: public sources and observation
  boundaries stated in each concept.md; no private traces or unpublished
  data.
• Quantitative readings from the cited battery appear as case provenance
  in provenance-erasure/concept.md; the deterministic test checks the
  declared attribution fields only.
• substrate, derived_from, completeness, and deletion semantics are
  excluded (separate spec surfaces; see Proposal: optional summarization-governance fields — provenance_kernel, disambiguation, summary_policy #53 comment of 2026-07-17 and
  issue Deletion semantics: §5.3 makes removed knowledge indistinguishable from not-yet-written knowledge #207).

Validation

• Parsed all expected.yaml with PyYAML.
• Verified every concept.md has OKF frontmatter with summary_policy.
• Cross-checked that every assertion/compression ID referenced in
  expected.yaml resolves to a declared ID in its concept.md.
• The fixtures provide deterministic expected receipts for the supplied
  good and bad summaries. They define the conformance cases and expected
  pass/fail outputs; evaluator implementation is outside the scope of
  this PR.
• git diff --check clean.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[leesharks000]

@googlebot I signed it!