Aelira Core v0.9.11: following the result through to the file

Why another corrective release

Our v0.9.10 article explained a problem with before-and-after reporting: some remediation paths estimated the after-score instead of measuring the saved file. That release replaced those estimates with paired checks where compatible verification is available.

The broader review found more work to do. A measured score still leaves an operator with practical questions. Has the remediation job finished? Is there a saved file to download? Can I inspect the document structure behind the result?

Aelira Core v0.9.11 addresses defects in those parts of the workflow. It corrects remediation status and outcome accounting, extends supported PDF reading-order checks, and adds a source-backed original/saved comparison in Review.

We shipped a separate corrective release because these problems affect decisions people make while using the product. The v0.9.10 release and article remain part of the public record. v0.9.11 builds on that correction; it does not replace its history.

A finished scan does not mean a fixed file exists

Scanning and remediation have different outcomes. A scan can finish while remediation is still queued, needs manual work, or cannot provide a verified output. The dashboard has to preserve those distinctions wherever someone opens the result.

Upload, scan-detail and bulk-upload pages now use persisted remediation status. History and Issues lead into the review workflow without assuming a saved file exists. Batch submission checks the queue receipt, and tracked document references survive a page reload.

When partial output is withheld, all original findings remain unresolved. Changes attempted inside a withheld file are not presented as fixes delivered to the operator. Where stored evidence cannot support per-finding attribution, that detail remains explicitly unreported.

The Issues page also uses document-level remediation eligibility instead of relying on absent per-finding capability flags. Its Issues Found count describes the original scan, including findings that may since have been fixed. It is not a remaining-issues count.

Inspect the original and saved PDF

Review now includes a collapsible, read-only comparison for supported PDFs. It displays actual page images and text from the document's tags, separately for the original and current saved file. Both reads are tied to checksum-verified bytes.

That makes it possible to inspect the tagged sequence alongside the page it describes. A location highlight appears only when the source text can be mapped unambiguously to the page. The viewer does not invent a location to fill an empty space.

This is evidence for review, not a PDF editor or a simulation of every screen reader. Unsupported content remains unavailable, and the viewer does not support every PDF. The reading-order review documentation describes the supported subset and parsing limits, including the absence of a hard process-level execution deadline.

The surrounding Review controls now wrap on narrow screens. The release also corrects upload labels and affected low-contrast controls, and stops the closed mobile navigation drawer from accepting keyboard focus.

Reading-order checks need the source content

Some tagged PDFs connect their structure to page content through marked-content identifiers and a ParentTree. The reading-order checks now resolve supported forms of those links, use stable page identity, and keep generated headings and lists bound to source content.

For supported pages containing both tables and other text, comparison now checks table placement and the surrounding text rather than skipping the whole page. This does not certify internal cell order or table semantics. Ambiguous or unsupported layouts cannot produce a complete verification.

PDF validation also distinguishes a failure already present in the original from one introduced in the saved output. Unresolved content still prevents a verified after-score. The correction improves the evidence available to the comparison without making uncertain results count as passes.

The release checks now process the files

The additional defects showed that our earlier checks had not covered enough of the workflow. We extended the release gate to exercise a real API, PostgreSQL queue and worker, using synthetic PDF, Word, PowerPoint and Excel documents.

Successful outputs are downloaded, hashed and uploaded again for an independent scanner run. The tests compare fixture text and values with the original to detect content loss. Manual-required cases must refuse download and account for the unresolved findings.

This is a real document-processing gate. It complements the existing simulated-backend browser suite; it is not itself a browser journey or an assistive-technology conformance test. Release review also requires a real-browser walkthrough of each changed route and outcome.

The release includes access-control corrections across Review and shared scan/remediation paths. Review actions and aggregate counts preserve department and course restrictions, and course identifiers are no longer treated as interchangeable across LMS platforms. The release notes describe the scope of these changes.

What operators need to do

For self-hosted installations, back up the database and retained files, then upgrade the API, worker and dashboard together to 0.9.11. No new database migration or production environment variable is required. The v0.9.8 operator requirements and v0.9.10 source-retention and rescan guidance still apply.

Historical score estimates are not retroactively verified, and old jobs are not assigned invented per-finding outcomes. Rerun remediation from the retained original to obtain new outcome records. If the original is unavailable, recover it from your source system and upload it again.

Review the saved document before publishing it. An automated score describes the implemented checks; it does not establish accessibility conformance or certify every screen-reader workflow.

The whole self-hosted product remains free and open source. Download Aelira Core v0.9.11 and read the upgrade guidance.