- Oct 3, 2026
- 4 min read
I Ran My Website Audit on My Own Sites: 71 to 99
I sell website audits. The free audit tool runs Lighthouse, technical SEO, security and a link crawl, and turns them into one score and a plain-English report (how the score is computed).
Until this week I had never run the full version, link crawl included, against my own three sites. When I did, this site scored 71.
The command
The tool has a terminal entry point that uses the same pipeline as the web app, with no lead and no email:
make report URL=https://www.jalcocertech.com \
ARGS="--links --link-method=docker --html --json=out.json"
--link-method=docker runs the link checker from a container, so the laptop
needs nothing but Docker. The HTML report is the same one a visitor gets by
email.
Before
| Site | Overall | Performance | SEO | Security | Links |
|---|---|---|---|---|---|
| www.jalcocertech.com | 71 | 86 | 76 | 70 | 40 |
| Ebooks library | 95 | 100 | 79 | 100 | 100 |
| Consulting page | 84 | 88 | 74 | 70 | 100 |
Step 1: fix the tool before judging the sites
Two of the findings were wrong, and both were the tool’s fault:
- “No sitemap” on the ebooks site. The check only tried
/sitemap.xml. Astro’s sitemap integration writes/sitemap-index.xml. The check now reads theSitemap:lines inrobots.txtfirst, then tries both names. - “No contact details” on the consulting page, which has an email in its
footer. The contact check only scanned the first 5 KB of HTML (on a modern
site that’s mostly
<head>), and it didn’t recognise Cloudflare’s email obfuscation, which replaces addresses with/cdn-cgi/l/email-protectionlinks. It now reads the whole page and counts those links.
If I had “fixed” the sites against those findings, I would have changed working sites to please a broken check. When a tool you wrote disagrees with what you can see in the browser, check the tool first.
Step 2: fix what was real
| Finding | Fix | Write-up |
|---|---|---|
| 6 broken links in one article | Linked the published copies, added the missing image | — |
| 273 broken links on ebooks, invisible until a 404 page existed | A remark plugin that rewrites chapter links | A missing 404 page hid them |
No sitemap, no Sitemap: line, no canonical | @astrojs/sitemap, robots.txt, <link rel="canonical"> | — |
| No CSP, HSTS or X-Frame-Options | _headers, checked in a real browser | CSP on Cloudflare Pages |
| LCP 3.8 s on mobile | Self-hosted fonts | LCP was render delay |
| 1.26 MB of JavaScript on the consulting page | Load the booking embed on approach | Lazy-loading the Cal.com embed |
| A tiny or wrong share picture | A new check, then real cards | Checking og:image, share cards |
After
| Site | Overall | Performance | SEO | Security | Links |
|---|---|---|---|---|---|
| www.jalcocertech.com | 99 | 98 | 100 | 100 | 100 |
| Ebooks library | 100 | 100 | 100 | 100 | 100 |
| Consulting page | 97 | 94 | 100 | 100 | 100 |
The consulting performance figure moves between 94 and 98 from run to run; that’s normal Lighthouse noise, not a regression.
What I took from it
- Run your own product on yourself, with every option on. The quick audit skips the link crawl; the full one found 279 broken links across my sites.
- Fix the measuring tool first. Two of the “problems” were bugs in the tool.
- A clean score can hide a gap in the checks. All three sites passed the share-image check while two of them shared a 248×56 logo. The check only looked for the tag. Finding that gap was worth more than the points.
- The fixes were small. One config line for the sitemap, one file for the headers, a font import, a twenty-line observer. Most sites, mine included, aren’t one big rewrite away from good; they’re a list of small things nobody had measured.
If you want the same list for your site, the audit is free and takes about twenty seconds.