Corrections to #520: diagram below the floor, and 76 more descriptions

[pftg]

Follow-up to #520. Two commits, both corrections to work in that PR.

1. R2's diagram was below the 9px floor, and I reported it as passing

The floor formula is minFontSize * (displayedWidth / viewBoxWidth). I used the viewport width (390) as displayedWidth. The actual content column at a 390px viewport is 354px.

• reported: 9.36px, passes
• actual: 8.5px, fails

Two measurement faults behind that. chrome-devtools resize_page(390) does not give a 390px CSS viewport - it clamps at 500, so every "390px mobile" check in #520 was taken at 500. emulate with 390x844x3,mobile,touch gives a true 390.

Narrowing the labels made it worse - the viewBox went 478 then 488 - because mermaid width is driven by parallel columns, not label length, exactly as the pipeline warns. A branching decision tree cannot clear this floor. Rebuilt as a vertical ladder: 272px viewBox, 12px rendered, verified in-browser at a true 390.

R3 (257px) and the Kamal post (272px) re-checked against the 354px column; both pass.

2. "plain HTTP" is not a term

Paul flagged it: developers say raw HTTP. I introduced "plain HTTP" during #520's plain-English pass, which was backwards - plain English means cutting abstractions the reader must decode, not replacing established terminology with words I invented. Reverted.

3. The remaining 92 descriptions - 76 recovered

#520 left 92 for a human. Diagnosing them showed they were not one problem:

Count	Why the first pass skipped it
63	First sentence complete, but under my 90-char minimum (83, 69 chars)
26	First sentence longer than 158 (159, 209 chars)
4	First prose line has no sentence ending

The 63 were an artificial floor I invented - I had conflated "short" with "fragment". A complete 83-character sentence is a good description; a 158-character dangling clause is not. MIN drops to 55.

For the 26, the fix walks further into the body for a sentence that fits rather than cutting the long one at a clause boundary. Clause cutting is what produced the dangling text in #520's first attempt.

76 rebuilt, 17 remain. Site-wide, truncated descriptions go 234 -> 17. Verified across all 76: zero end in ..., zero end without terminal punctuation, and all 76 dev.to-backed files carry seo_override so the 10-minute sync cron cannot revert them.

The 17 are announcement posts and link roundups whose opening prose offers no usable sentence. They need writing; a script should not guess.

Retraction from #520

There is no theme-level table overflow bug. I reported one. single-post.css:287-296 already sets display: block; overflow-x: auto on post tables under 767px and it works - a true 390px check on fractional-cto-vs-full-time-cto shows both tables scrolling inside themselves (406 and 420 in a 354 container) with no page overflow.

My original diagnosis read getComputedStyle(t.parentElement) - the wrapper DIV - when the rule targets the table itself. Dropping R2's Raw HTTP column in #520 was therefore not required. It stays dropped on content grounds, since every cell read "you write it".

Gates

bin/hugo-build 8/8. Local macOS suite per Paul's instruction rather than CI.

Branch cut fresh from master and cherry-picked: #520 was squash-merged, so a plain rebase would replay already-merged commits and conflict with their own content. 79 files here, not the 229 a rebase showed.

🤖 Generated with Claude Code

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: a6a00b71-6946-4423-8136-4564f2b76872

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}