The in-place client half of the anti-slop interaction work (guidelines
G1, G2, G4, G13). It still never re-POSTs, still serializes saves, and a
batch still never reloads.
- Focus: the focused element is recorded by identity (its region, its
key, which same-key element it was) and the fresh one is focused
without scrolling. If an answered pick's form folds into a closed
<details>, focus goes to its summary; if nothing is left, to the region
(tabindex=-1, set by the script). Focus outside the swapped regions
is not touched.
- One status line per page (_status.html). It floats at the bottom
centre, above the fixed review stage, so it moves nothing and is in view
wherever the reader is. Wider than 900px, the letterhead and footer the
review covers leave the Tab order (visibility:hidden, CSS only).
- The line is never hidden: empty, it takes no space and stays displayed.
"Saving…" at the press, "Still saving…" on a repeat press, "Saved." when
the swap lands (cleared after 2s if still the same write), and warnings
with data-tone="warn". Every write sets or clears the tone. The form in
flight carries aria-busy until its save settles.
- The client never reloads over a draft: both of its reloads run only
when every in-place form is clean except the one just sent, unchanged
since its press, asked again at the reload beat; otherwise it says so
and stays. A beforeunload guard asks when an in-place form is dirty (its
own reload does not ask). The embed asks when one of our answers is
unsent, and skips the pressed form on its own one-form submit.
- Six booth-dev browser tests read the line's hidden state; they read
its words and tone instead. Two r2_submit_all.toml rows are re-anchored
to the same failure in the moved code.
Folded from the heid bug-hunt (panel 4/4, thread 01M3MRTNTWEPJHTN4APRR81KH4):
- aria-busy mirrors which forms are in flight on the LIVE page. It is set
at the press and re-synced whenever a save settles, so it ends on every
path (a stale tile the swap never replaced included), and a queued form
replaced by an earlier swap is marked busy again.
- A press inside the reload beat cancels the reload.
- A failure that stayed is said again after an unrelated save, rather
than buried under "Saved.".
- A 204 followed by a failed page GET is "Saved.", never "could not save".
- An edit made while its save flew is said to be unsaved.
- A focused <summary> has a key.
- The queue settles on rejection.
- The embed's skip covers the one navigation its submit starts; a
cancelled submit, or one that leaves the page in place, is guarded
again.
- Pinned: no in-place form holds a control dirty() cannot read, and no
region nests in another.
Folded from this slice's gate: the status line floats (fixed, bottom
centre, above the review stage) instead of sitting at the top of <main>
or under the viewer's bar. In the flow, every save's "Saving…" moved the
page; booth-dev's test_a_flag_lands_in_place_and_every_region_catches_up
caught a 50px jump. The viewers' grids are back as they were.
Contract: as_antislop S5b (heid contract review and bug-hunt folded).
Falsifiers: antislop.toml S5b sections.
From booth-dev's hulda bug-hunt with heid's second voice (BRINGA, thread
01M3MVGQ7QSCCK8WT59TQ4J469):
- The booth page's "kept — release" asks by name, as the Desk's does.
- WORDS has no prototype: data-confirm="__proto__" or "constructor" is an
unknown word, and asks, instead of throwing before preventDefault.
- The confirm helper moved into <head>: its capture listener exists
before any form, so a click during load is asked too (the inline
confirm() it replaced had that property).
- shown() also marks U+2028/U+2029 and the zero-width characters.
- Derived ids take ':' (bk-ask-<id>-<key>:prompt, bk-ask-<id>:title), which
no id or key can contain; '-prompt' and '-title' collided with valid
keys. booth-dev's chip test now looks its fragment up by [id=...].
- human_dur says "—" for a value that is not finite, instead of raising.
- The tile's copy of a note drops its id (booth-dev: mark-<id> is the
panel's article).
- Four guards that asserted source patterns now also hold on computed
effects: embed rings, rings inside clipping containers, the withdraw ×
on both axes, and question-level notes fields.
Contract: as_antislop S5a (fixup). Falsifiers: antislop.toml 102/102 with
r2_flow.toml 24/24 proved; the full gate follows.
The markup and CSS half of the anti-slop interaction work. The in-place
client is untouched (that is S5b).
- Glyph-only controls carry a name: the withdraw ×s, downloads, open full
page, the viewers' ✕, the board's pin, copy and remove, the bench's
remove, the 1:1 toggle ("1:1, natural pixels"). Film-strip and tray
frames carry the file's name as sr-only text instead of reading "01".
A Desk row's wipe names its booth.
- Fields are named by aria-label, not by their placeholder.
- The inline ask's options are a radiogroup labelled by the prompt; a
single-question fieldset gets an sr-only legend; a titled ask's title
takes bk-ask-<id>-title (it duplicated the question's id).
- One h1 per page (sr-only on the Desk, review and compare), a skip link
to <main id="main">, theme-color for light and dark.
- The review tape is one picture (role=img); its segments leave the tab
order (the film strip holds the same links, named).
- Wipe now uses the Desk's delegated prompt, moved to base.html: it names
the booth and asks the kept-booth question for a kept booth.
- Embed focus rings of its own; rings drawn inside clipping containers;
the withdraw × at least 24px, 44px under a coarse pointer;
touch-action:manipulation; strips contain their overscroll; a long
slug wraps on a phone.
- A truncated why carries its full text in title; a countdown of 48h or
more reads in days.
Two r2_flow.toml rows for the confirm helper now name base.html, where
the helper moved (anchors unchanged; the gate found them drifted).
Contract: as_antislop S5a. Falsifiers: antislop.toml 86/86 proved (S1-S6, S5a);
all 12 tables 366/366 proved on this tree.
Operator, 2026-09-28: "go with your recommendations".
- Tagline: 'held for review · wipes in {ttl}h unless kept', mono, muted,
12px, sentence case ("ephemeral" goes, as agreed with booth-dev).
- 'Needs you' rows lose the 3px side stripe; the '? N OPEN' stamp says it.
The flagged filmstrip frame keeps its bottom stripe.
- The brand dot is matte (glow = live power; a live bench keeps its glow).
- Wipe now and the armed bulk delete carry the hazard stripe on a 3px
::before, so the button's own background is honestly what sits under its
text; the stripe renders as before.
Contract: as_antislop S6. Falsifiers: antislop.toml 49/49 proved (S1-S4, S6);
all 12 tables 329/329 proved on this tree.
From the anti-slop run (design-dev, 2026-09-28). A rendered doc ran 110-120
characters a line and its h3 sat at 1.08x its body. Prose blocks in
.markdown-body are capped at 72ch (pre and tables keep the full width, where
they scroll), and h3/h2/h1 step at 1.2/1.44/1.73em. Measured in a real
browser: a long paragraph now reads under 76 characters across, and every
heading step is >= 1.18.
Contract: as_antislop S4. Falsifiers: antislop.toml 43/43 proved (S1-S4);
all 12 tables 323/323 proved on this tree.
From the anti-slop run (design-dev, 2026-09-28). Measured in a real browser
at 390x844 (tests/test_antislop_browser.py), because a layout claim read off
a stylesheet is a guess.
- Bench rows wrap at <=600px (state + name/URL, then who/when/actions); the
name's column was squeezed to ~53px and overprinted the owner and date.
- The board head and the benches head drop their note under the count, so
"33 links · 1 pinned" / "3 benches" keep one line.
- An inline doc's bar wraps: the name takes the full width and breaks only
where it must; it was set one word wide.
- A file tile's download link starts below the ordinal badge.
- The review and compare stages drop the tagline at <=600px (the server
marks them `page-stage` on <html>), so the header is one line; the Desk
keeps its tagline (the test's negative control).
Not changed, with reasons in the contract: `.vname` already ellipsises, and
the filmstrip's clipped edge frame is the scroller's "more" cue.
Contract: as_antislop S3. Falsifiers: antislop.toml 41/41 proved (S1-S3);
all 12 tables 321/321 proved on this tree.
From the anti-slop run (design-dev, 2026-09-28). Faded is not legible:
opacity divides whatever contrast a line had.
- Review arrows: the chip under the thin chevron is 82% dense, not 60%;
the glyph now clears 7:1 over a white stage by colour (was 3.84:1), and
reads at pixel level where the detector sampled a 2.9:1 median.
- Filmstrip numbers, the marks' state stamp and the inline ask's state tag
are labels at 11px (were 9.5 / 10.5 / 10.5px).
- The Desk's section rules, the board note and the bench note are
sentences at 12px.
- Retired benches: no opacity; the link and URL take --text-muted.
- Embed chrome: answered-ask details and the "recorded:" line inherit the
host's text colour at full strength (the embed cannot know the host's
palette); the notes placeholder inherits it at 75%, not the UA grey.
Folded from the heid bug-hunt: the embed's ask title no longer fades
either (Q9), and the legibility claims are also held on the browser's
COMPUTED style (tests/test_antislop_s2_browser.py): a stylesheet grep
cannot see a later rule in the cascade (font-size:1px, color:transparent,
filter:grayscale, a placeholder at opacity:0); the browser can.
Folded after the first gate run: the flagged tray's number, the tile's
"flagged" stamp and compare's A/B badge were still under the 11px label
floor; they take --size-micro too. Two film-number rows are re-anchored
on the .film-ord selector: the tray's line is now identical to it, and
the runner mutates the first match.
Contract: as_antislop S2. Falsifiers: antislop.toml 34/34 proved (S1+S2);
all 12 tables 314/314 proved on this tree.
The anti-slop run (design-dev, 2026-09-28; operator: "start the fix slices")
found raw ISO stamps with microseconds and offsets, the poster's IP address,
and HH:MM in board rows and <time> tooltips. Operator convention 2026-09-24:
a clock the operator reads is 24-hour local time as four digits, no colon.
- `clock` filter: ISO (any precision, any offset), epoch, or the board's
`YYYY-MM-DD HH:MM` -> `28 Sep 0848` local, year only when not this year's.
Never raises; what it cannot read is shown as given. No regex (INV-3).
- `byline` filter: a handle is shown, an IP address is not. Stored `by` and
`answered_by` are unchanged (u2 still records the client host).
- Applied to the marks' answer and memo lines, the inline ask's state tag
(so the embed chrome inherits it) and the link board's row time, each in a
<time> whose datetime= carries the stored value exactly.
- `date_stamp` (the created/updated tooltips) renders `YYYY-MM-DD HHMM`.
Folded from the heid bug-hunt (panel 4/4, thread 01M3MGPFKWBX0SJK5HFE0P3AFM):
clock converts a number inside its guard (an int past float range raised,
Q1); a date or ISO week renders no invented 0000 (Q8); byline also hides
addr:port, [v6]:port, addr/prefix and addresses behind invisible characters
(Q7); the board row's author is bylined (Q5). Refuted: Q3 (default Jinja
Undefined has length 0; the test stays as a StrictUndefined guard).
Accepted with reasons: Q4, Q6.
Contract: docs/contracts/as_antislop.contract.md S1. Falsifiers: antislop.toml 15/15 proved (S1);
all 12 tables 295/295 proved on this tree.
vh