OCA-RECOVERY-001 · Application reliability · Repaired

Preserving review state after a failed sample load

A replacement that never arrives should not erase the review already on screen.

Order Confirmation Auditor previously cleared its file pair, comparison and notes before a replacement fictional sample finished loading. When that request failed, the in-tab review was unavailable. Original files and reports already downloaded were unaffected. [1]

This case explains an existing repaired finding from Browser Busters, Ten Thirty Two’s AI-assisted review project. It is a low-severity application reliability issue, not a new browser-engine vulnerability.

The failure depended on replacing an existing review

The controlled starting point was a completed comparison with annotations. A failed request for another sample then left no comparison to export. The failure reproduced in Microsoft Edge 157.0.4341.46 and Playwright WebKit 26.5 on Windows, with an additional independent Edge reproduction. [1]

Two controls narrowed the cause: changing only the sample selector retained a byte-identical reviewed report, while a successful explicit replacement correctly began a new review. The defect concerned when replacement happened, rather than every interaction with the selector. [1]

Tested transition boundaries after repair [1]
ActionReview behavior
Choose another sample onlyKeep the existing review.
Replacement is pendingRetain current work; block comparison, editing and exports.
Replacement failsKeep the prior pair and annotations.
Replacement succeedsReplace both sources together; begin unreviewed.

Prepare the candidate before replacing current work

The repair keeps the current workspace separate while the candidate pair is fetched and validated. Only a successful, still-current candidate replaces both sources. During that attempt, controls cannot approve or download a result whose source transition is unresolved. Failure retains the existing annotations. [1]

Cancellation matters as much as failure. A later source change, tolerance change or clear action invalidates pending work. An older completion must not resurrect a review after the visitor has moved on. Keeping the old review through a failed replacement does not mean restoring it after every deliberate change. [1]

All 23 distinct targeted local regression cases passed after repair, covering malformed samples, pending exports and canceled or superseded requests. One Edge case checked resuming an interrupted selected-file read. The repair is recorded in Order Confirmation Auditor Release 3. [1]

A useful check for another file-review workflow

These conceptual checks belong on an authorized local copy with fictional files. They are a way to inspect the same boundary, not evidence that another application has the defect.

  1. Complete and annotate a review. Save a comparison report so the original sources and notes can be checked.
  2. Change only the sample choice. Confirm that selecting an option does not silently replace current work.
  3. In the local test setup, make a replacement request fail. Check the original source hashes and annotations after failure.
  4. Compare that result with a successful replacement and with a deliberate clear while loading.

For a reviewer, the practical distinction is between an unsuccessful attempt and an intentional reset. This repair protects the tested in-tab transition; it does not create autosave, cloud backup or recovery after closing the tab.

Evidence and limits

Testing used fresh profiles and fictional CSVs. Playwright WebKit on Windows is not Apple Safari. Firefox and physical mobile devices remain untested. No customer loss or incident frequency was measured, and controlled old-handler dispatches do not establish an attacker capability. [1]

[1] Original application evidence. Browser Busters evidence file, finding OCA-RECOVERY-001 and recoveryRegressionSummary. The findings hub separates these application observations from historical vendor security advisories.