# Adoption limits

Version 0.1.0 is a bounded original companion and local executable reference. Research assurance remains reviewable-draft. Startups may use source, scope and descriptive mapping governance without enterprise infrastructure. The international and AI examples are synthetic scenarios, not claims about named companies.

Executable use requires authenticated actors, verified source/steward declarations, current policy/target catalogue, trusted host clock, protected evidence storage and a current owned SQLite database. These duties are not supplied by this package. Metadata-only commits neither acquire external bytes nor apply business facts. IAM, live connectors, distributed effects, partitions, complete erasure/retention, quarantine resolution, writable archive import, automatic EFA/EAP adapters and production-scale storage are deferred.

Archive and native functions expose the full restricted register. Protect identifiers, counts, references and digests. Copying a projection does not enforce current revocation. Register-global mapping/round IDs and active-lineage reservations also expose availability on collisions; new pins do not cross acquisition scopes. Receipt IDs expose a register-wide activity sequence and a trusted coordinator supplies shared head/fence preconditions. Authorized competing writers can infer that a key is unavailable; unauthorized calls get a uniform refusal before lookup. There is no zero-knowledge or timing guarantee.

The reference permits 128 retained events, 256 descriptors per batch and a 512 KiB canonical complete archive. It replays full history and full-state digests for each operation, with potentially quadratic work under a lock. Read benchmark-results.json for one near-capacity local observation; it is not an SLA. Raw JSON transport has a separate 8 MiB limit. SQLite busy errors and interrupted bootstrap recovery are external host duties. Crash tests cover process exits, not power-loss or lying hardware. Exact build pins and per-event full-state digests prevent silent reinterpretation of history. Hashes establish byte consistency, not source authenticity or latest-head continuity. Diagnostics retain compact observations only: at most 32 total and 4 per actor/scope/epoch, with uniform no-event refusal at diagnostic, event or byte limits. Monitor capacity externally and reserve room for round closure; there is no rollover or emergency over-budget write. Native snapshots remain read-only historical projections with explicit predecessor checking.