Bryan Hamilton / writing

bryanhamilton.info / writing / article

Fix It All: What a Full-Site SAGA Audit Caught on 47 Pages

I audited all 47 pages of my portfolio against the bar my own DJ site set: 32 pages missing Twitter cards, 4 stragglers with no canonical, 14 missing author links, and a count drift nobody noticed. One sweep, zero gaps.

  • November 15, 2026
  • 5 min read
  • seo · audits · saga · og
  • More writing

I run two sites. The DJ Calendar got the full treatment from day one: every artist page ships canonical, meta description, Open Graph, Twitter cards, JSON-LD, llms.txt, sitemap, and an IndexNow ping before I call it live. My portfolio - the site you are reading - grew article by article, and I believed it had the same standard. Believed is the operative word. I had never actually checked.

So I ran the audit I run on my own product, against my own portfolio: crawl every URL in the sitemap, check every legibility surface per page, count the gaps. The final instruction of the session was two words - "fix it all" - and this is the story of what the crawl found and what the sweep fixed.

What the checklist actually covers

My shorthand for the whole stack is SAGA, and it is not one thing - it is seven surfaces feeding seven different consumers:

  • Canonical tag - Google, so it knows which URL is real.
  • Meta description - the search snippet you actually get to write.
  • Open Graph tags - Facebook, LinkedIn, Discord, Slack previews.
  • Twitter card tags - X.com, which famously does not read Open Graph the same way.
  • JSON-LD - structured data for rich results and AI answer engines.
  • llms.txt and ai.txt - the front door for LLM crawlers, which I have written about before in Make Your Site Legible to AI Crawlers.
  • Sitemap plus IndexNow - discovery for the crawlers that still ping.

Each surface is cheap on its own. The failure mode is drift: the site grows, the surfaces do not follow, and three weeks later you have forty pages where thirty-five are perfect and five are quietly invisible somewhere.

The audit: script, not vibes

The audit was a crawler over my own sitemap - one request per URL, a checklist per page, a results grid at the end. No spot-checking the pages I was proud of, no "looks fine in the browser." Every one of the 47 live URLs (24 articles plus 23 tool and hub pages at the time) got the same seven questions, and the script kept score.

The point of scripting it is the same point behind self-verifying deploy workers: a checklist you run by hand is a checklist you will eventually run by memory. Memory is exactly where drift lives.

What it found

Most of the estate was already green - the articles I had published through my guarded publisher shipped with the full stack from day one, which is the whole point of that pipeline. The gaps were everywhere except the pipeline:

  • 32 pages missing Twitter card tags. Open Graph was present, so Facebook previews were fine - but X.com was left guessing. Every one got twitter:card, twitter:title, and twitter:image in one pass.
  • 4 straggler pages with no canonical, no OG, no JSON-LD. The oldest utility pages on the site, built before the standard existed. They got the full block.
  • 14 pages missing my Coder Legion author link in sameAs. Structured-data authorship that points back at a real profile - stitched in where it was absent.
  • A broken og-manifest. My og.php endpoint renders a 1200x630 PNG for any slug, but the manifest that feeds it had drifted out of sync. Repaired and recounted.
  • Count drift, the sneaky one. My llms.txt said "23 engineering articles." The live site had 24 - then 25 by the time I fixed it, because two dev.to imports used suffixed slugs I had never logged. The file was not lying on purpose; it just had no reason to stay true.

"Fix it all"

The fix was the same discipline as every deploy on this site: guarded, token-gated workers with per-file backups and verify-after-write, one surface class per pass. Twitter cards swept first - 32 pages, all verified. Then the four stragglers got canonical, OG, and JSON-LD in one block. Then the 14 sameAs stitches. Then the manifest, then the llms.txt and ai.txt recount from live data, not from the ledger.

Zero failures across the sweep. The final state: 25 articles and 23 pages, every URL green on all seven surfaces, the whole thing logged in my public deploy ledger. The audit found its own gaps, and the sweep closed them the same evening.

The drift lesson is the real product

Here is the part worth stealing. On this site, llms.txt and ai.txt are static files on disk - which means every new article can silently make them stale. On my DJ site they are dynamic rewrites served by a PHP builder, so they cannot drift; they are computed from the live data every time. Same for the og-manifest: a static index of dynamic content is a countdown to being wrong.

The rule I adopted from this audit: recount from live, never from the ledger. Any count that appears in a generated file gets recomputed at audit time, not recalled from notes. The ledger tells you what you did. Only the server tells you what is true.

Run your own

You do not need my checklist to steal the shape of this: crawl your sitemap, ask seven questions per URL, script it so it never goes soft, and fix everything in guarded passes with backups. The whole audit took one evening because the surfaces were enumerable - and the reason they were enumerable is that I had named them. That is most of the battle. Name your standard, then make the machine enforce it.

If you want the deeper version of the crawler-facing half of this - what llms.txt is for, how ai.txt differs, and what AI crawlers actually request - start with Make Your Site Legible to AI Crawlers. And if you want to see what a site looks like when every page ships this way from birth, that is The DJ Calendar.

NEXT READThe Deploy Pattern That Keeps My 1,000-Page Site Alive Without Shell Access

More from the workshop

All writingEvery article, on this domain. Open source11 production-extracted projects.