Replaces the generated isometric diagram with the supplied 2752x1536
render (engineer at a multi-monitor desk, container streams flowing into
server racks). Re-exported at 1024/1280/2048 widths; alt text and the
lightbox caption now describe what the image actually shows.
The previous hero-banner was composed as a wide banner with an empty left
40% reserved for overlay text. In the 50/50 hero column it rendered as a
shrunken half-width picture with dead space on both sides.
New illustration fills the whole 16:9 frame edge to edge (code/CI/CD left,
container + host cluster centre, observability right) so the right hero
column reads as a system diagram at its native 518px width, with the
2048w lightbox for detail reading.
Also add a 1280w WebP so DPR2 desktops do not pull the 2048w asset on
first paint.
- brand: three-colour Hoelee mark (rasterised from the brand vector,
transparent, public/images/hoelee-mark-96.png) followed by the name
- section links grouped in .nav-links so the mobile header puts them on
their own row instead of wrapping past the search/lang/theme controls
- measured 320/360/390/414/600/768/1280: scrollWidth == innerWidth,
header 91.6px two rows (<600px) / 60.3px one row, mark loads at 22/26px
Adds EN + ZH post php2shell-patching-every-php-runtime-i-manage: measured
PHP versions across the fleet, the order the fixes shipped in, the 8.5
package trap, the OLS vhost handler switch, and the silent bot-block bug
the sweep uncovered. Ships og cards (EN/ZH), banner + webp, 2 thumbs.
§4d now carries the real commit chain (f32ae90 → 04a1082 → cf4504e), the
verification that matters (Actions 292/293/295, six pages 200, artwork md5 vs
live, check-assets 162/162) and an owner queue with three items.
The task doc is rewritten around that queue: (1) confirm whether the four stage
timings (14.2 / 2.1 / 32.4 / 48.6 s) were measured — they have no trail in the
repo and the reviewer's complaint was exactly about unmeasured numbers;
(2) supply the before/after jitter clips; (3) decide which session owns edits to
these three posts, since two sessions on the same files put one session's
paragraphs into the other's commit.
Also drops a claim inherited from the parallel session's completion block: the
reviewer's meta descriptions were never adopted, the frontmatter keeps the
originals.
The benchmark table derives its figures from 218.3 s ((352-218.3)/352 = 37.98%),
while the TL;DR said 38.1% (that one came from rounding 218 exactly). Same page
carried both numbers; now the TL;DR matches the table. 1.61x throughput is
unchanged (352/218.3 = 1.6125).
Reviewer's five "highest priority" items, each checked against the source before acting:
* 38% speedup -> 38.1% less render time (352s -> 218s, 1.61x throughput); says it was
a single run each, same clip, same machine, same encoder settings.
* "microsecond-accurate" -> frame-accurate: at 30 fps one frame is 33.3 ms, and the text
now states that word-level timestamps are not microsecond-accurate.
* "Three minutes later" -> "a few minutes later", plus the one measured number (218 s
render for the 4-minute track) so the claim is checkable.
* Zero Jitter / Zero Light Flashing -> "no visible jitter" / "no dark flash at the cut",
scoped to the clip that was measured.
* ti.mode 0o666 -> 0o644. Hygiene, not the syntax error he reported: source and live HTML
both read 0o666 (the live page splits it across two highlight spans and his text
extraction inserted a space), and 0o666 is legal Python.
* "0 MB RAM spike in Chrome" now scopes itself to the Chrome container and says the
Python client still buffers the whole tar. New "What this pattern does not solve"
block: client buffering, target directory must already exist, tar member names become
container paths, Portainer key / CDP 9222 are root-equivalent.
All three posts get a TL;DR and inline cross-links (the trio had zero internal links
before); the ZH links use the real /posts/zh/<slug>/ path.
Not done, because it needs the owner's data rather than invention: end-to-end timing
table, benchmark conditions, whether 1.89 s was a single run, before/after jitter clips,
the "what still needs a human" list. Task list:
docs/consultant-review-2026-10-22-3-video-posts.md.
Artwork for the three slugs regenerated (1600x900 banner, 1200x630 share cards EN+ZH,
400/800 list thumbs); the ZH pipeline card already carries the corrected description.
The YuE2 and subtitles posts shipped text-only: render_from_art.mjs only READS
ogImage/banner, and after the 2026-10-19 bulk migration no script had to write
them, so both pages advertised /og-default.jpg and rendered no hero banner.
- new metaphors (in build_captions.py, frozen here via sync_prompts_catalog.py),
captions, art-meta evidence, banner plan and the five derived files per slug
- frontmatter: ogImage + banner in EN and ZH for both posts
- scripts/check-assets.mjs is now step 1 of npm run build, so CI fails when a
published post lacks a language-correct og card, banner, banner .webp or
400/800 list thumb. It reported 16 problems on the unfixed tree, all here.
/tags/<tag>/ passed chrome={false}, so a tag page showed a bare list — no filter
input and no chips — while /posts/ had both. Enable the chrome and add a
chipPosts prop: the tag strip is built from the whole language collection because
its chips navigate (page-scoped counts would lie), while category chips and the
row filter stay scoped to the posts on the page so their counts match the rows.
The current tag is marked is-active + aria-current instead of looking clickable.
Six photo cards in 2 columns (3 rows) at desktop widths; phones collapse to a
single full-width column and still show only the first four. Drops the 900px
two-column step (now the base) and the cramped-card type overrides that only
existed for the 2-up phone grid.
Related posts were 4 text-only cards. They now carry the post's language-specific
ogImage (/og/<slug>.jpg for EN, /og/zh-<slug>.jpg for ZH), so an English page never
shows Chinese art. Grid is 3 across above 900px, 2 across below, and phones show the
first 4 only; read-time is dropped under 600px so the date fits.
Verified in the built HTML + headless Chrome at 1264/884/500px: 3/2/2 columns,
6/6/4 visible cards, no horizontal overflow, all six images decoded.
The ::before used CSS escapes written through a Python string, which ate them
(octal escape) and shipped control bytes — the button read as mojibake.
- use a literal UTF-8 arrow instead of \2193\0020
- figcaption is now flex, so a long title wraps under the button rather than
being crowded by a float; verified at 360px and 680px
- round the shown duration down so it matches the player's clock (197 s = 3:17)
Tutorial on turning untimed AI-generated audio into LRC + SRT: align ASR word
timestamps against the KNOWN lyrics (SequenceMatcher) instead of trusting a
transcript, then a governor that rejects impossible anchors, refills runs by
unit-proportional duration, and enforces a 1.2 s minimum line gap.
Ships 18 English songs (342 subtitle lines, mean anchor coverage 0.96, 9/18
fully anchored) with synced lyrics plus downloadable .srt per track. Documents
both player bugs and the headless-Chrome numeric probe that finally caught them.
offsetTop is measured from the nearest positioned ancestor, so inside an
unpositioned .yue2-box it returned the line's distance from the page top
(thousands of px). scrollTop clamped that to maxScroll, so the box jumped to the
bottom the moment singing started and the text was unreadable.
Now: box.scrollTop += (pr.top - br.top) - (clientHeight - lineHeight)/2, plus
position:relative on the box. Verified in headless Chrome with a numeric probe:
the active line sits ~124px into a 274px box at t=48/100, nothing is highlighted
and scrollTop=0 at t=0, and the box settles at maxScroll only for the final lines.
findIndex(t > now) - 1 returned -1 before the first line started, which the
`if (i < 0) i = lines.length - 1` guard turned into the LAST line — so playback
began scrolled to the bottom. Use a last-passed-index scan instead (-1 = before
the first line), centre the active line inside the box (so upcoming lyrics are
visible), and scroll the box itself rather than scrollIntoView(), which dragged
the whole page. Also bumps the box to 250px. Fixes yue2 post + both Suno posts.
Twelve Mandarin songs came back exactly 240.000 s with the outros cut.
Documents the ffprobe tell, the BPM myth, the seed-cache trap, the
2.0-2.8 hanzi/s spread, the 300 s VRAM ceiling, and the archive-the-clipped
policy. Ships 11 finished tracks with ASR-derived synced lyrics (EN + ZH).
hoelee