# Additional adoption limits from the focused audit

These clarify existing behavior; they do not add permissions or change the implementation.

1. **Latest snapshot is a host responsibility.** validate_extension checks a candidate against the previous snapshot supplied by the trusted host. It cannot know whether that previous snapshot is the latest. Supplying an older root defeats rollback detection. The production host must retain the latest accepted digest, compare-and-set it atomically, verify outer supersedes/previousSnapshotDigest links and admit each new row. The reference acceptance writes those outer links; it does not implement a durable chain-validation service. Passing validate_extension(x, x) is valid replay, not evidence that x is current.
2. **Closure before a future part is outside the simple operation.** A cutoff earlier than a nested part's validFrom would create an empty clipped interval, so guarded closure rejects. Resolve the future appointment by an explicit reviewed correction before closure, or use a future adapter supporting coordinated term changes. The library does not silently remove the future part.
3. **Revival is explicit correction.** An authorized correction may change a retracted record back to active/asserted, with preserved prior revisions and a new receipt. It is not automatic resurrection. Full-term corrections remain broad under the scoped governor/current source-writer trust.
4. **Scale is unproven.** Schema array limits of 10,000 are shape limits, not demonstrated throughput. The reference rereads schemas and performs repeated history checks. No production-scale performance claim is made.
5. **Host error handling.** Denied is raised by evaluate for a reader/purpose denial. admit uses Invalid for rejection, including authorization rejection. Host endpoints must sanitize all raw exceptions and never disclose the returned complete ledger to write-only callers. This package is not a web service.
6. **References and discovery.** WM-XCT-002 and WM-XCT-012 are documentary-only semantic references. Native acceptance explicitly selects both the separately identified companion and the semantic-only WM-XCT-001 parent as roots; it does not prove an independently omitted parent is needed or resolved. The custom companion ID is available in this package and local Dimension registry, not as an independently registered canonical public WM resolver entry.
7. **Evidence limits.** External reviews do not execute code or authenticate downloads. Code/schema/native tool pins and complete package checksums are separate from their verdicts. The acceptance report's sourceDigests are a subset, while final package checksums cover every shipped file. Candidate publication metadata used before deployment is not a live-publication proof; production readback and the downloaded-package rerun supply that proof separately.
8. **Record and evidence privacy.** All provided instances are synthetic. A production register, its input digest, competing sources and steward identities are not inherently public. Preserve all-or-deny full-register access until a separately reviewed partial-disclosure adapter exists.