Skip to content

fix: audit backlog and second Bing scan (hero image, feed, alt text, sitemap dates) - #104

Merged
ZeeshanAdilButt merged 9 commits into
mainfrom
fix/visibility-backlog
Nov 14, 2026
Merged

ZeeshanAdilButt merged 9 commits into
mainfrom
fix/visibility-backlog

Conversation

@ZeeshanAdilButt

Copy link
Copy Markdown
Member

Why

Two lists of findings for the live site: the visibility audit backlog of 13 Nov, and the second Bing Site Scan (2 errors, 6 warnings, 1 notice). Each item was checked against the live site first; this fixes the ones that were still true.

Audit backlog

Home page largest paint (poor for 8% of real visits in Cloudflare's data). On a slow visit the largest element was the hero photo, fetched through the image service with a 4.7 second wait for the first byte. The service fetches the original from another host and converts it, once per width and format, and on a quiet site that cache is often cold. The photo is now three ready-made widths in AVIF and WebP with a JPEG fallback, served as static files, preloaded, sized, and cached for a month.

Sitemap dates. 17 of 39 addresses had a lastmod. Every page now has the date its own source last changed, read from git history (scripts/page-dates.mjs, stored in lib/page-dates.json). Never the build date. The deploy workflow refreshes the file before it pushes; that step cannot stop a deploy.

RSS. There was no feed. /blog/rss.xml only looked as if it worked, because the blog answered any unknown address with a page and status 200. /rss.xml now lists every post and guide, every page links to it from its head, and /feed.xml and /blog/rss.xml redirect to it.

Soft 404s (found while checking the above). Unknown addresses under /blog/ and /guides/ now answer 404.

security.txt. /.well-known/security.txt with the contact SECURITY.md already gives and an expiry date that moves forward with every deploy.

Structured data. BreadcrumbList on posts, guides and DSOC project pages. Organization already had sameAs; it gains the Facebook page the site links.

Second Bing Site Scan

Images with an empty alt, counted by Bing as missing: 51 on six pages. Each gets a real alt from data the page already has: the story's name for video thumbnails, the person's name for the hero faces, the organisation's name for icon-only logos, the post title for blog covers. Nothing is read twice: where the same words are already next to the image (a button's label, a printed name, the title in the same link), the duplicate is hidden from screen readers.

/brand/session-template.html. Still flagged despite the noindex header, so the file itself now has a title, a description, its own noindex tag and one h1. The three later slide titles are h2 with the same look.

/api/feed/export blocked by robots.txt. The block stays. The export is now a button, so crawlers have no address to find.

Not done, and why

  • Event markup on sessions. Sessions are loaded in the browser and their dates are free text, so valid Event data cannot be produced without changing how sessions are stored.
  • Thumbnails on /links keep an empty alt: each sits beside its own community title inside one link, and Bing did not flag that page.

Checked on a local production build

  • Empty alts on the six pages: 51 on the live site, 0 on this build. Accessibility 100 on home, blog, team and partners.
  • Hero: one AVIF request at phone and desktop sizes, no request to the image service, renders as before.
  • Feed, redirects, 404s, security.txt, sitemap dates (35 of 35 locally; the database projects join on the live build) and breadcrumbs, all read from the HTML a crawler gets.

Cloudflare's real-visitor data rates the home page's largest paint poor for 8%
of visits. On a slow visit the largest element was the hero photo, fetched
through the image service with a 4.7 second wait for the first byte: the
service had to fetch the original from another host and convert it, and on a
quiet site that happens again for each of its eight widths and formats.

The photo is now three ready-made widths in AVIF and WebP, with a JPEG
fallback, served as static files from the edge cache and cached for a month.
It is cropped to the 5:4 shape it is shown in, preloaded in the format the
browser prefers, and has its width and height.
There was no feed. /blog/rss.xml only looked as if it worked: the blog
answered any unknown address with a page and a 200.

/rss.xml lists every post and guide, newest first, and every page links to it
from its head. /feed.xml and /blog/rss.xml redirect to it.
Any address under /blog/ or /guides/ returned a "not found" page with status
200. Search engines call that a soft 404 and may index it. Only the posts and
guides that exist are served now; anything else is a real 404.
/.well-known/security.txt tells researchers where to report a problem
(RFC 9116). The contact is the address SECURITY.md already gives, and the
expiry date moves forward with every deploy.
Only posts, guides and DSOC projects had a lastmod. Each page now carries the
date its own source last changed, read from git history by
scripts/page-dates.mjs into lib/page-dates.json. Never the build date: saying
every page changed on each deploy tells search engines nothing.

The deploy workflow refreshes the file from the full history before it pushes.
That step cannot stop a deploy: if it fails, the committed file is used.
Each of these pages now tells search engines where it sits: Home, its section,
then the page. The organisation's list of profiles gains its Facebook page,
which the site already links.
Bing's Site Scan counts an empty alt as a missing one, and 51 images on six
pages had one although they carry meaning.

- Video thumbnails are named after the story they open.
- The faces in the hero carry each person's name.
- Icon-only logos carry the organisation's name.
- Blog covers carry the post's title.

Nothing is read twice: where the same words are already next to the image (a
button's label, a printed name, the post title in the same link), the
duplicate is hidden from screen readers.
Bing's scanner reads the file even though it answers noindex, and reported no
description and four h1 tags. It now has a title, a description and a noindex
tag of its own, and the three later slide titles are h2 with the same look.
The export address is blocked in robots.txt on purpose, and Bing kept
reporting the link to it even with nofollow. A button gives crawlers no
address to find.
@ZeeshanAdilButt
ZeeshanAdilButt merged commit 278d356 into main Nov 14, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant