Supporting technical evidence

A connector should recover without losing the authority change.

This bounded diagnostic tests the read-only SQL and signed-event reference paths through restart, rejection, replay, and dependency impact. The result is exact accounting, not a claim about every enterprise connector.

Recorded result

Restart safely. Reject bad events. Revalidate only what changed.

The diagnostic ran from clean installed wheels at the recorded source revision. All 15 declared checks passed.

Read-only SQL

100,000 / 100,000

Every fixture row was accounted for across composite cursor pages and a forced midpoint restart. No row was skipped or duplicated, and repeated reconciliation returned the same digest.

Signed event feed

100,000 / 100,000

Every signed event was signature-verified and normalized. Invalid signatures, stale timestamps, and oversized bodies were rejected. Duplicate and older events produced no new semantic transition.

Replay and impact

Exact

All 1,001 projected authority objects were compared and the complete ordered digests matched. One source change identified the two bound workflows and one precedent; the unrelated workflow remained unaffected.

What the run establishes

The reference path converged on the same governed state.

The SQL path resumed from a persisted composite cursor. The event path failed closed on its declared invalid inputs. Replaying the projected source events into an empty database produced the same complete object count and ordered projection digest. The source-change scenario then identified the exact dependent workflows and precedent, suspended reuse, and created the corresponding review path.

Recorded resource envelope: 17.47 seconds, 116.812 MB peak process RSS, and no less than 69% system memory available on the test machine. These measurements describe this run only; they are not a production capacity or latency commitment.

The scale path scanned and normalized all 100,000 SQL rows and signature-verified and normalized all 100,000 signed events. A configured subset of 1,000 SQL records plus two adversarial event-feed updates exercised Authority State projection, replay, ordering, and impact. This is not a claim of 200,000 end-to-end governed state transitions.

The portable bundle excludes the source and projection databases. It contains the report, sampled records, reference impact report, and manifest required to verify artifact hashes and declared checks.