# Release review — Enterprise Source Synchronization 0.1.0

Actual independent Claude Opus 5 High and Grok Heavy studies and S1/S2 research clarification were followed by separate implementation audits. Divergent R2/R3 verdicts and reproduced defects are retained. Both R4 delta audits accept with limits; both final R4a documentation/test addendum reviews accept. Author-executed 101 tests, three native compositions and one near-capacity observation are separate evidence, not reviewer execution.

Disposition: **publish bounded reviewable draft** under the owner's standing authorization. No provider grants authority.

All41 R4a frozen files retain exact bytes at release paths. Thirty complete R3 bodies were supplied in the same provider conversations, followed by complete R4 diffs/three added files and the explicit R4a text/value delta. JSON was reviewed at value level against minified baseline bodies; author patch reconstruction and shipped-schema byte tests are separate. Eight unchanged context files and external toolchain implementations were hash-listed, not body-reviewed. reviewedFiles identifies this cumulative semantic review coverage, not an assertion that each R4a whole body was resent. Generated release review, manifests, page, checksums and ZIP are post-audit bookkeeping and separately verified. Candidate-status text in frozen README and synthetic installation reports describes capture time; actual lifecycle is established only by the publication manifest and verified live runtime.

Accepted adoption limits:

- Single-host metadata reference only:128 events,512KiBcanonical archive,8MiBraw transport; no rollover. Full replay and per-event state hashing can be quadratic. One benchmark is not an SLA.
- Host must authenticate actors, supply current grants/catalogue/time, protect evidence, prove current owned store and wrap privileged exports with current disclosure.
- Attesters must review every shared page, including other writers. Intake contributors can occupy nonterminal slots or force errored incomplete closure.
- Global ID/lineage/key availability and register-wide receipt sequence remain observable. No timing or zero-knowledge guarantee.
- Only zero-byte interrupted bootstrap gets explicit refusal; nonempty partial/corrupt stores may raise raw SQLite errors. No automatic repair, clone detection or writable archive resume.
- Native outer checks alone are insufficient; installed companion replay and trusted predecessor are required. Three new synthetic Dimensions do not establish existing-Dimension migration or production readiness.
- No live connectors,IAM,source authentication,distributed exactly-once,authority adapters,automatic quarantine resolution,partitions,production-scale storage or universal standards conformance.
- Remaining test limits include nonempty hot-journal bootstrap,multiple-than-two-writer contention,intermediate recursive validation and cross-purpose round predecessor construction. Trusted clock validation precedes ordinary authorization; runtime versions recorded,not fully locked.

This release has 5 bundles,10 layers,27 findings and54 questions. EM-XCT-08 remains partially covered; deferred capabilities require separately reviewed increments.