- README.md: what this is, how to run and build it, the folder map, how light and dark mode work (@hanzo/appearance, @hanzo/design), where every price and checkout link lives, scripts, and known differences from live - docs/build-notes/: status files, request notes, per-page checklists and reviews, plans - Removed: screenshot galleries, captures of third-party sites used as references, the generated status page and its script, editor config, a personal working-rules file - Code comments point at the new docs paths (no code changes) Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01P5YBdwj33XByFVTvu9bfci
9.3 KiB
/s3 — James-eye review
Ready: 2026-11-07, slice D — black room has one moment ("4 can be lost, and it still reads" at 64, rest at 26), phone object row no longer shifts per phase (CLS 0), readout/fan verified contained; peel gives the lower half its own object
Verdict: 7/10 First complaint: "The top is the best thing on the cloud pages, but look: two seconds in, 'ten data · four parity' falls out of its white box onto the grey, and little line ends poke out between every tile. Then I get ten and four again in the chapter, again in four big equal numbers on black, and the bottom half is the same pin, black room and close as /pubsub and /realtime."
Walked 7 Nov 2026, about 16:58–17:02 UTC, on main @ 2b836213 (S3 last touched by d6ba0f17, "survivors stay bright… shorter pinned section, chapter 3 plays the loss, pill on one line"), with Playwright Chromium. I walked it moving at 1470×801 and at 375×812 (isMobile, hasTouch, DPR 2):
- each size had a video and frames over the first ~7 s with no input, phase sampled every 250–650 ms;
- desktop got a slow wheel scroll and the phone got smooth half-screen steps, with a 1.1 s dwell on every screen;
- then hover on a volume and on the ask, four taps, a refused fifth tap, Replay;
- then reduced motion, JS off, 320 px and 375×667.
Non-localhost POSTs were blocked.
Stills: s3/desktop-sharded-overflow.jpg (phase 1: readout spilling out of the object card, fan stubs in the gaps), s3/desktop-hero-reads.jpg, s3/phone-hero.jpg. s3/hand-four-survivors-grey.jpg is from the previous walk and shows the bug that is now fixed.
Not held against the page:
- The red "1 Issue" badge is a hydration mismatch in the shared
FooterHTML(data-x-shell="footer"__htmldiffers server and client). It appears intermittently (2 of 5 loads), and the trace names no page file. - The phone header's solid "Try Hanzo" makes two solids on the phone's first screen.
- The cookie card covers the bottom of the phone stage on first visit.
- The header stays black for a moment over the white close after the black room.
LineMaskgives the h1 textContent "Hanzo ObjectStorage" with no space between its lines. That component is shared.
What works
- The first screen sells and moves without input. The 52px title sits left. The lede, the one page solid "API reference" and the "Source" text link sit right. The stage ends at y 650 of 801.
- The object breaks up about 1.3 s after navigation commits, roughly 0.9 s after first paint on this loaded box.
- Phase timeline:
0 → 1 (1.3 s) → 2 (2.8 s) → 3 (4.1 s). It runs twice, then rests on "✓ still reads". - Shards visibly fly on eased tweens (
d-t3: shards in flight between d3 and p1).
- The fixes from the last review landed:
- Hand-picking is honest now. After four taps all fourteen volumes stay at opacity 1. The ten survivors are white with ink shards, the four are dashed #ededed with ✕. A fifth tap adds
.no(3px nudge), the readout says "4 / 4 dark · still reads", and Replay hands the loop back. - The pin is shorter. Desktop steps change at y≈1800, 2160 and 2520, and the pin is about 2.7 screens, down from about 5.
- Chapter 3 plays the loss. The cells go dark one after another, then "✓ still reads" lands.
- Chapter 4 draws the dashed protocol lines. The KMS chapter lights the volume role and the lock snaps in.
- The phone pill stays on one line.
- Hand-picking is honest now. After four taps all fourteen volumes stay at opacity 1. The ten survivors are white with ink shards, the four are dashed #ededed with ✕. A fifth tap adds
- Hover and press answer. A volume fills #e4e4e7, its name inks, and the toolbar names it ("d4 · data"). The ask goes #0a0a0a → #262626 at opacity 1, so it does not fade. Volumes and Replay press .97, and text links press 1px.
- Engineering:
- Desktop CLS is 0.0009.
- No horizontal scroll at 1470, 375 or 320.
- Reduced motion lands on phase 3 with
getAnimations()at 0. - JS off gives a 4,871px page with 0 hidden text.
- The ask is at y 248–296 at 375×667.
parity.pyOK (28/28, 3/3),taste.pyOK,unique.pyOK.
The 11 answers
- Sells in 5 s, moves within 1 s? Yes on both sizes. The title, lede and ask are readable at once, and the break-up has begun by about 1 s after paint.
- One in, one out, or a document? Better than last time, still half. The hero is a real scene, and the pin now changes per step with motion in chapters 3 and 4. Below that come the black room's four figures plus the prompt, then the shared close. The lower half reads as a sequence of rooms, not one idea handing to the next.
- A real visual object in every section? Yes: the stage, the roles diagram, figures plus prompt, and the close's Sigil. The roles diagram is still four rounded boxes around a pill, the weakest object.
- One ask, nothing floating, scale rhythm?
- One ask: yes. On desktop the header is outlined until the hero leaves, then carries the solid "API reference".
- Nothing floating: nothing page-owned floats.
- Scale: 52 title, 44 intro, 32 chapters, 21 roles, 64 figures, 13 captions. The object shrinks to shards and fills back, and "Get Object Storage" grows to 64.
- The flat moment: the black room still sets four equal 64px numbers side by side (10 / 4 / 4 / 1), with no one big thing.
- Every hover and press answers, no jumps? Yes. One weakness: on hover the ink line for d4 runs under d1–d3, so you only see ink stubs in the 8px gaps, not a line to the tile.
- Overlap, font swap, layout shift?
- Desktop overlap: in phase 1 the readout "ten data · four parity" is
nowrapand ends at x 390 while the object card ends at 355. It spills 35px out of the white card onto the stage, on every run and on the first screen (desktop-sharded-overflow.jpg). This is a regression from the one-line pill fix. - Fan stubs: they still show in every gap between volume tiles, because the paths end at tile centres beneath the tiles.
- Layout shift: desktop CLS 0.0009. Phone CLS 0.024, all from
.s3-filereflowing as the object row changes phase. That is under the 0.1 fail line but over the 0.02 lab target. - Font swap: none seen.
- Phone diagram: the "one binary" pill sits over the inner corners of the role tiles.
- Desktop overlap: in phase 1 the readout "ten data · four parity" is
- Which reference, and is its motion there?
- Apple: one idea per screen and a pinned scene that now changes per step. The pin is there and is no longer dead.
- Self-running demo: the hero's demo is the authored moment (idle + pointer + state).
- Synchronized: no card open, and none expected.
- Different from its neighbours? The hero is its own. Below the hero,
PinnedChapters→ClipWipedark room →LaunchCloseis still shared with /mq, /node, /pubsub, /realtime, /network, /o11y and /registry (8 of the cloud files importPinnedChapters, 10 useLaunchClose). That is the "it's just a template" risk. - All 18 parts covered? The card has 18 headings, each with a filled
Done here:(see the card check). - Banned tells? None: no hue, no gradient text, no eyebrow above the h1 ("Storage" sits in the stage toolbar), no Inter, no cursor label, no fading buttons, no icon boxes.
- Copy untouched? Parity is 28/28 text and 3/3 links. The numerals 10/4/4/1 stand for Hanzo's counted words, and the card lists that as an open question.
Card check
docs/build-notes/coverage/s3.md has 18 ## N. headings with every Done here: filled, plus the three lines above part 1.
Claims that check out on the walk:
- one page solid;
data-ctas3.hero/s3.close;data-thumbused by the header;- survivors at full contrast after a hand-pick (parts 8, 12, 17);
- the shorter pin;
- chapter 3 playing the loss;
- reduced motion reading;
- JS off legible;
- no
setInterval.
Corrections:
- Part 11 / part 3 say "no overlap" at 1470×801. The phase-1 readout overflows the object card by 35px.
- Part 11 says the phone ask is at y≈230. I measured 248, still well above the fold.
- Part 18 says the review's fixes "are in". Three of six are in: the hand-pick, the pin length with chapter-3 motion, and the pill. The fan stubs, the 10/4/4 repetition and the template close are not.
To reach 8
- Keep the readout inside its card. Either let
.s3-readwrap inside the object card on desktop (dropnowrapabove 900px, or narrow the copy box's padding), or widen.s3-objto fit "ten data · four parity" (about 200px). It must never cross the white edge. - Fix the fan. End each path at its tile's left edge (or draw it over the tiles to the slot), so no stubs show in the gaps and the hovered ink line is actually visible end to end.
- Stop telling 10/4/4 four times, and break the equal row. The hero says it best. In the black room make one figure the moment ("4 can be lost, and it still reads" at 64) and set the others at 26. Or let the close's object be something other than a Sigil seeded [10, 4, 4].
- Give /s3 a lower half its neighbours don't share. For example, the agent prompt becomes the close's object, or the black room holds the hero's object again at the moment it reads back, instead of the generic
ClipWipe→LaunchClosesequence used on 7 other cloud pages. - Phone polish.
- Reserve the object row's height so
.s3-filedoes not reflow per phase (CLS 0.024 → under 0.02). - Give the shards a visible drop from the object row into the 5×3 grid, since the fan is hidden there.
- Move the "one binary" pill off the role tiles' corners.
- Reserve the object row's height so