# Adoption limits

Trusted-host pure in-memory reference, one new receipt per second, at most 10000 records. No IAM, digital signatures, source fetching, claim evaluation, connector availability checks, durable concurrency, partial disclosure, numerical confidence or retention/erasure enforcement. Integrity/author/origin/method values are declarations. Complete imported snapshots are not authenticated and cannot establish current-root completeness. Native V3 validates only the outer value type; the companion and history-extension checks are mandatory. Startup/group/AI examples are synthetic. Prior parent source/legal/domain-profile holds remain. Independent implementation audit results appear in review.md; no whole-parent re-ratification is claimed.

Serialized JSON is limited to 8 MiB, with separate host byte limits required before parsing. Full revalidation on admission is intentionally simple and can be expensive; 10000 is a structural ceiling, not a measured capacity. A trusted host must map every write failure/diagnostic to a generic rejection for a write-only caller and return only a receipt. Distinct conflicting external digest pins are separate query targets; no alias conflict discovery or automatic claim reconciliation is implemented.

CPU cost can be quadratic. Rollover/segmentation are not implemented; plan migration before structural limits. Activity pins are retrospective recorder declarations, not proof of actor knowledge at execution. Independence uses pinned historical origin declarations; changed origins require reassessment. First snapshot history is host-approved bootstrap/import; full native envelope and trusted predecessor remain mandatory.

Review freshness is enforced per assessment identity. Parallel or successor assessment IDs can cite a review first registered no earlier than their explicit basis; the view preserves these assessments side by side and selects no truth winner. A newly minted review ID is a declaration, not proof that someone repeated a real-world review. Conservative consequences: an assessment metadata correction resets the previous-revision timing boundary; a corrected basis requires a later first-registered review; re-pinning the same review ID for a non-insufficient assessment is rejected. These are bounds of this reference, not automatic fraud or assessment-quality detection.