Twenty-four commits landed today across four threads that had nothing to do with each other, and there is no vault journal for the day itself to work from: this entry was reconstructed after the fact from git history and session transcripts. That gap is worth naming up front, because part of today's work was fixing the exact same problem one journal-month back: a sweep on the 27th had already reduced most of August's journals to "no work evidenced" stubs, and the day's biggest single commit exists to undo that.
The edge was wrong on a real monitor
The night started at 1 AM with a plain complaint: on a standard 2K monitor the site's edges sat too wide, "all the way at the edge." The layout had been hard-pinned to a 1440px design reference, so anything wider just stretched the same fixed insets across more screen. The fix scaled insets, the lockup and type with the viewport instead: 12px up to 60 or 100px, the lockup 147 up to 300px, measured against a real 2504×1271 window rather than guessed. That fix immediately created a second bug: the 100px inset at the wide end was eating into the measure of the clock sentence, so the text column got narrower as the viewport got wider, and a line that wrapped at 3 lines started wrapping at 5. Two more commits chased that down, reverting nav-link scaling because it had gone too big, then fixing insets bottoming out at 12px in the 1000-1400px band and moving the phone breakpoint to 768px. Four fixes for one starting complaint, each one visible only because the previous one exposed it.
The backfill that fixes a journal problem with a journal fix
Seven missing August journals went out today: the 21st, 23rd through 28th. The commit that published them says plainly that the "no work evidenced" verdicts in the stub journals were not trustworthy: several had to be corrected against git author dates and session transcripts on n5ubuntu, because the sweep that produced the stubs had only looked at this machine and missed everything from earlier in the month. A second commit published September 4 through 8, deliberately leaving August 29 through September 3 unpublished, the away window, five days where the lab broke with nobody watching, which gets told as one post rather than five empty days.
A third commit went further and gave every September post three versions: an original around 1,000 words, a summary under 500, and a tl;dr under 100, with the summary now the one that publishes. That change is scoped to September only; August stays original-only. The fourth content commit in this thread centred the hero, reflowed the clock sentence again, and deleted the Contact page, which the mailto link had already made redundant.
None of that fixes the actual hole: the vault journals for already-published days are still stubs. Today's entry is proof of the same symptom happening in real time, no journal, reconstructed after the fact from git and transcripts, the same method used to correct the August backfill.
About page, nine times over
The About page went up between 18:07 and 19:41, then went through nine rapid revisions the same evening as it was reviewed live: reordering sections, adding cluster facts, folding the VLAN count into a stat tile, narrowing the Goal prose column, and at one point centring the HomeLab stat row and then reverting that same centring a few minutes later. The last commit of that run merged the lede into Aspirations, renamed the Goal section, and stripped em-dashes from the page, the site-wide em-dash removal that everything published from tonight on has to honour.
Two cables, live
Around 2 AM, in the middle of the responsive work, two unrelated cluster faults got fixed with Eric physically at the rack. pve03's RTL8125 standby NIC had never been cabled; pve02 had a cable faulted down to 100Mb. Both got re-plugged live, "I have just plugged that one in," with quorum held at 3/3 through the whole thing.