Release Notes

v1.0.0-beta.59 August 2026

Four fixes, two of which affect whether you can trust a report you have already sent.

A suggested phone number could come back missing its area code. When a phone link carried an extension after a comma - tel:+16306270316,405, a common way to write a direct line - Vesper correctly flagged the link as broken and then offered a corrected address to paste in. That correction silently dropped three digits from the front of the number. It looked entirely plausible, it was a valid phone address, and a page "fixed" with it would have passed the check on the next scan while no longer reaching the library. Extensions are now offered in the standard ;ext= form with every digit preserved, and Vesper refuses to print a corrected number at all unless it can account for every digit you wrote. The same fault affected numbers outside North America: a London number written with spaces was being rewritten into an unrelated US number. If you have sent remediation guidance containing a corrected tel: value, it is worth re-checking against the number on the site.

An audit that could not read a single page reported success and produced a report. If the address you gave Vesper could not be reached - a redirect it declined to follow, a block, a firewall, a typo - the scan finished green, said "0 pages scanned", and wrote a full report bundle including a PDF titled with the site's name. Every count in it was zero, and zero findings looks exactly like a clean site. Vesper now treats this as a failure rather than a result: the scan reports an error, says what stopped it, and every format - the in-app report, the PDF, the Markdown, the CSV, the Notion page - states on its face that nothing was audited and must not be read as a pass. If you hold a report showing zero pages audited, it is not a clean bill of health and never was.

A site's address with and without www. is now understood as one site. Giving Vesper yourlibrary.org when the site redirects to www.yourlibrary.org caused it to treat its own destination as somebody else's website and skip it, so the scan came back empty. Both forms are now recognized as the same site. Separate subdomains - a catalog, an events system, a hosted booking tool - are deliberately still treated as separate sites, because they usually belong to a different vendor and need a different fix.

Skipped addresses now say why, and where they went. A page Vesper declined to open used to appear with nothing but its address, leaving no way to tell a real gap in coverage from correct behavior at the edge of the site. Each now carries a reason and, for a redirect, the address it actually resolved to.

An image link's name is now credited to the image. When a link contains nothing but a picture, the name a screen reader announces comes from that picture's alt text. Vesper reported the name correctly but said it came from the link's own text, which pointed anyone reading the report at markup that does not exist.

v1.0.0-beta.58 August 2026

Corrections to how findings are evidenced, and one reliability fix worth reading.

Reports no longer miss stylesheets that arrive through another stylesheet. A theme can pull in a second stylesheet with an @import instead of a link in the page, and the link-underline check never followed those. Two things went wrong at once: any underline rule or exception inside an imported file was invisible, and the report still stated that every stylesheet had been read. A result that had skipped a file was being presented as complete. Imported stylesheets are now followed, including ones that import others in turn, and any that still cannot be read are named individually along with how they were reached. If you have an earlier report on a site whose theme uses @import, its statement about stylesheet coverage was optimistic - a rescan will say so accurately.

"Setting: On" is no longer said about a site where no link is underlined. A theme can carry a rule that underlines links and then cancel it with a later rule, so the setting is technically on and no reader ever sees an underline. The report now says "On, but not reaching any link" and explains that the rule exists but never reaches the page.

Evidence snippets are no longer cut off before the useful part. Each finding carries a piece of the element's markup, trimmed to keep reports readable, and it was always trimmed from the end. An element with a long opening tag - an inline icon, a long list of classes - lost its own text, so a button reading "View All Services" could appear in the report as though it had no text at all. Snippets now keep both ends and say how much was left out in between.

Every finding on an element now carries that element's accessible name - the text a screen reader announces - and where it came from: an aria-label, a title, the element's own text, or nothing at all. Where a rule fires precisely because an element has no accessible name, the report states that outright rather than leaving it to be worked out from a fragment of markup. It appears in the reports, the CSV and the MCP results.

Findings on repeated embeds now say which embed. When a page carries several embeds from the same source without distinct titles, every finding pointed at the same generic location. Each now names the address it embeds.

Link decoration no longer appears in the PDF or the Notion page. Those two are the client-facing report, and a table of links with contrast ratios reads as a list of defects to someone who did not commission the measurement - which, as the report itself says, it is not. It remains in the in-app report, Markdown, CSV and the MCP results, where the audit work happens. The same decision was already in force for the hover and focus measurements.

v1.0.0-beta.52 August 2026

One feature, and the fixes that made its numbers trustworthy.

Vesper Audit now reports which links are distinguished from the text around them by color alone. WCAG 1.4.1 asks that where color is the only difference between a link and its surrounding text, that difference reaches 3:1. No automated rule answered this reliably - the closest one fires on a small fraction of sites, so its silence never meant a site was clean. Vesper now reads the computed style of every link on every page and reports two things: whether the site underlines links by default, and every link that ended up without one, with the CSS rule that removed it, the link's color and weight, and the contrast against the nearest text that is not a link. It appears in the in-app report, Markdown, PDF, Notion, a new link-decoration.csv, and the MCP server's results.

Links inside menus are measured, not skipped. A dropdown or mega menu whose column headings are ordinary bold text sitting above bold link text is one of the most common places this hides, and it is invisible to any check that treats "inside the navigation" as a reason to look away. Vesper measures those links, labels them so you can filter them, and reads them even when the menu is closed - so a whole mega menu is no longer skipped because nobody clicked it open during the scan.

Every measurement says how close the compared text actually was. When a link has no neighboring text, a check that keeps looking outward will eventually find something unrelated and report a convincing ratio against it. Each row carries the distance the search traveled, so a real neighbor and a coincidence are easy to tell apart. The report explains how to read it.

Counts are distinct elements, not instances. A link in a header or footer template appears on every page, so counting instances makes a number grow with the size of the crawl rather than with the size of the site - the same theme scanned at 50 pages and at 150 would report wildly different figures. Every count is now the number of distinct elements, with the instance total shown beside it and the number of pages each one appears on.

Hover and focus are reported too, and kept apart. A link can be underlined at rest and stay underlined on hover, so hovering tells the reader nothing. Vesper reports where that happens. No success criterion requires a hover affordance, so this is a design question rather than a WCAG failure, and it is described that way. Focus is different - WCAG 2.4.7 does require a visible focus indicator - but an outline, a shadow or a background change satisfies it just as well, and those are not measured here, so a focus result is a prompt to go and look rather than a finding.

This is a measurement, not a verdict. Vesper reports what it found and leaves the judgment to you, the same way it handles accessibility-overlay detection. None of it counts toward the Critical, Serious, Moderate or Minor totals. Links presented as buttons, pagers, breadcrumbs or icons are excluded, since their shape already tells you they are interactive, and skip links are excluded because they are hidden until focused and are never findings.

Stylesheets that could not be read are named. A stylesheet served from another domain cannot be inspected, and a result that quietly skipped one is not a clean result. Where that happens the report says so and marks the finding provisional.

Stopping a scan at the page budget now builds the report. When a crawl paused at your page budget, choosing to stop discarded everything it had scanned - the button offered a review it could not produce. Stopping now finishes the crawl properly and writes the full report for the pages already visited, and the report opens by stating that it is partial and how many pages were never reached. The pause itself also reads honestly: it says how many URLs are still queued, only offers to apply exclusions when it has some to suggest, and no longer styles stopping as though it were destructive.

PDF export works from the MCP server. Asking an assistant for a full audit with a PDF returned every other format and an error for that one. The PDF is now produced the same way and to the same page setup wherever the scan was started from.

v1.0.0-beta.47 August 2026

A large release covering everything since beta.42.

An AI assistant can now run a full audit for you. Vesper Audit ships an MCP server, so any client that speaks Model Context Protocol - Claude Desktop, Claude Code, Cursor, Zed, VS Code and others - can be pointed at it and asked to audit a site. It crawls the whole thing and writes the same report bundle you would get from running the scan yourself: the JSON, CSV, Markdown and optionally PDF, in your usual output folder, with the usual folder name. The assistant gets the path and reads the complete results from there, so it works from every finding rather than a sample. Four tools are available: scan_url for a single page, scan_site for a full crawl, scan_status to follow one in progress, and check_contact_links for broken tel: and mailto: links. The workflow this is built for is the remediation loop: change a template, ask for a rescan, see whether the finding cleared. Setup lives in the app under Settings > MCP server and in full on the MCP Server page. Scans still run entirely on your own machine using your own Chrome; nothing is sent anywhere.

Long crawls do not time out, and a site scan audits the whole site. A crawl starts and returns immediately, then reports progress - how many pages so far, and which page it is on - while it runs. It uses the page budget from your Scan settings rather than stopping at an arbitrary cap. And if a scan does hit its limit, the result says it stopped and how many pages it never reached: a partial audit that looks complete is worse than one that refuses, because nothing tells you the pages you cared about were never visited.

Broken phone and email links are now detected. A tel: link whose visible text reads perfectly can still carry an href no dialer can parse - URL-encoded parentheses, literal spaces, a stray plus in the wrong place. The defect is invisible to sighted review and the link simply fails to dial. Vesper Audit now checks every tel: and mailto: link and reports the specific problem alongside a corrected href you can paste in. These are reported as functional defects rather than WCAG failures, and carry no WCAG criterion, because a broken link is broken for everyone - the report says so plainly rather than claiming a success criterion it does not fail.

Pages whose only issues are invisible no longer count as pages with issues. A hidden element in a site template - a phone link inside an inactive tab panel, say - appears on every page, which could make an entire site report zero clean pages when almost all of them were clean to a visitor. Findings on elements that were not being presented at scan time are still reported in full, but they no longer count against the clean-page total, and the Rendered column in violations.csv marks them. That column is a real determination for every finding: Yes means the element was checked and was being presented, No means it was hidden, and an empty cell means no determination could be made - never a default. Screen-reader-only content is deliberately treated as rendered, since it is in the accessibility tree and a defect on it is a real defect.

Large image warnings no longer depend on the order pages were crawled in. When the same oversized image appeared on several pages, only the first page to load it was recorded - the browser served it from its cache everywhere else, so the scan never saw it again. The Large Images Report states how many pages carry each image, and that count could differ between two scans of a site that had not changed at all. Every page using the image is now counted, without re-downloading anything.

Suppressions that match nothing are called out. A suppression with a stale selector looks exactly like one that is working - both leave no trace in the report - so a rule that silently stopped matching could hide findings you believed were being filtered deliberately. Every suppression now reports how many instances it actually removed, and any that matched nothing is flagged in the report and in the live scan log.

Pages that are not real content are surfaced separately. Soft 404s and maintenance pages scan clean simply because there is nothing on them to fail. These are listed in their own section and flagged in summary.csv, so a page that passed for the wrong reason is not mistaken for a page that passed.

The scan log is saved with the report. Skipped pages, off-domain redirects, applied exclusions, per-page errors and suppressions that matched nothing were all recorded during a scan and then discarded when it finished. They are now part of the saved results, so a scan you open months later still explains itself. Every report also records the Vesper Audit and axe-core versions that produced it - rule behavior changes between builds, so when comparing against an earlier scan that is the first thing to check.

The updater can no longer take you backwards, and no longer offers to. Two separate faults produced the same result. The app was willing to treat an older version as an available update, so it would download one and offer it; and a downloaded update installed itself on quit regardless of what was actually running, so you could quietly end up on an older version than you started with with nothing to tell you. Older versions are now never offered, and an update installs when you click Restart now in the banner, only when it is genuinely newer. Anything stale is discarded.

Optional zipped reports. Settings, Report options, "Also save a zip of each report". A report is a folder, so sharing one always meant zipping it by hand first. The zip is written alongside the folder and the folder is kept, because the report viewer reads it.

Smaller page budgets. Scan settings now offers 50 and 100 pages alongside the existing choices. The budget pauses a scan for review at that page count; the smallest option used to be 500, which put the pause and the pattern suggestions it offers out of reach on a typical library site.

Smaller fixes. Container class signatures no longer include layout utility classes such as position-relative and d-flex, so the same component matches across sites that style it differently. Sitewide rolled-up findings now name the specific element each one refers to, so two findings from the same rule are no longer indistinguishable. Dropdown controls in Settings match the rest of the interface, in both light and dark mode. Both modal dialogs trap keyboard focus correctly and return focus where it came from. Documentation and in-app text no longer read as though one particular AI assistant is required. And the scan-time guidance is honest: measured across real scans it is two to three seconds per page with link checking on, not the roughly one second the docs used to claim.

v1.0.0-beta.42 June 22, 2026

Findings without a landmark now name their nearest container. Some flagged elements sit outside every page landmark (header, nav, main, footer), so the report correctly says "Landmark: none" - but on its own that left no way to place the finding without opening the browser. Vesper Audit now adds a Container line directly below the Landmark line, naming the nearest meaningful ancestor: the nearest stable id (auto-generated ids - entity numbers, hashes, framework ids - are skipped in favor of a stable one, for example div#block-branchselect), a dialog or modal, or a meaningful component class. It appears on every element-bound finding in every format - the in-app viewer, PDF, Markdown, Notion, and a new Container column in both violations.csv and sitewide.csv - and is most useful exactly when the landmark is "none". Pure utility wrappers are skipped so the container is always something meaningful.

Color contrast issues show the exact ratio and the colors involved. A color-contrast finding could previously come through with an empty explanation - no ratio and no foreground or background color - which forced a manual check in the browser. Vesper Audit now shows the full detail on every color-contrast finding: the measured contrast ratio, the required ratio, the gap below the threshold, and the foreground and background colors (for example, "4.33:1, required 4.5:1 - 0.17 below threshold, foreground #0f7547 on background #f6d9d9"). This includes findings rolled up in the Sitewide Template Findings section, and the CSV export's sitewide.csv file has dedicated Contrast Ratio, Foreground, Background, and Expected Ratio columns. When a background can't be measured - a gradient, an image, or a semi-transparent color - the report explains that reason instead of leaving the field blank. This affects every report format: the in-app viewer, PDF, Markdown, CSV, and Notion.

v1.0.0-beta.40 June 9, 2026

PDFs served at a clean URL are no longer scored as web pages. When a site links a PDF at a tidy address with no .pdf on the end, the browser opens it in its built-in PDF viewer - a near-empty shell with no page title or language. Earlier builds ran the web-page checks against that shell and reported the missing title and language as Serious issues, even though the address is a PDF and there is no web page to fix. Vesper Audit now recognizes these during the scan and lists them under PDFs Found (marked "served directly at this URL") instead of scoring them as HTML, so they no longer inflate the severity counts or the audited-page total. A normal page that simply embeds a PDF is unaffected.

Findings that repeat across the whole site are now rolled up. A single template element - a phone link in the header, a block in the footer - appears on every page, so one element could previously show up as hundreds of separate findings and make a site look far worse than it is. Vesper Audit now recognizes when the same finding repeats across most of the site and rolls it into one entry labeled "appears sitewide (N pages)" in a new Sitewide Template Findings section. The severity counts now count each distinct finding once rather than once per page, so the headline numbers reflect the real issues to fix. Every affected page is still listed in the detailed per-page results, so nothing is lost. This appears across the in-app viewer, PDF, Markdown, CSV, and Notion reports, with a dedicated sitewide.csv file in CSV exports.

v1.0.0-beta.37 June 5, 2026

Reports now open on a new Summary page, pinned at the top of the report viewer. It shows the whole audit at a glance: two rings - the total issues split by severity, and how many of the audited pages have issues (hover any section for its exact count) - followed by the counts by severity, scan stats, the top issues ranked by how often each occurs, and expandable reports for ambiguous links, broken links, and large images. A reminder that automated checks can produce false positives sits above those reports. Accessibility overlay detection lives here too, with a short explanation of why overlays are not recommended.

v1.0.0-beta.34 June 4, 2026

Accessibility overlay detection. Vesper Audit now detects third-party accessibility overlay and widget products on the sites it scans - UserWay, accessiBe, Recite Me, ReachDeck/Browsealoud, and others - and names the vendor. Detection is informational: it is a vendor-presence signal, not a WCAG violation, and is never included in the severity totals. Each detected overlay carries a recommendation to remove it and remediate the site directly, because overlays do not fix the underlying HTML and frequently interfere with the assistive technology people already use. Results appear in the in-app viewer, PDF, Markdown, and Notion reports, and as a dedicated overlays.csv file in CSV exports.

v1.0.0-beta.33 June 3, 2026

Landmark context on every finding. Element-bound violations now include a Landmark line directly above the selector in every report format, naming the nearest ARIA landmark (header, footer, nav, main, aside, or a named section) plus the chain of meaningful wrappers down to the flagged element. This makes it immediately clear where on the page each issue lives - for example, telling apart two identical links in the header versus the footer - without opening the live page. CSV exports gain a Landmark column.

v1.0.0-beta.32 May 17, 2026

Two long-standing Pattern Detector bugs are fixed in this build. Clicking Apply on a Pattern card during a scan now removes the card and prunes the queue as intended. Scans no longer cascade into hundreds of wasted iterations when a single page hits a fatal browser frame error - the scan recovers cleanly and continues from the next URL in the queue. Pattern cards also now appear at the threshold count of 25 rather than jumping straight to a post-batch number.

v1.0.0 April 24, 2026

Vesper Audit is a macOS desktop application for automated WCAG 2.2 A/AA accessibility auditing. Provide a starting URL and Vesper Audit produces a complete site-wide report.

Scanning

Reports and exports

Settings and workflow

Platform