Add before-launch.md: a founding lottery testing whether serious
pre-launch help (a founder's stipend, a feasibility study, legal setup)
creates more co-ops that last, measured independently of the bank, and
a 5% founder stake payable in verified unpaid work. Fund it from the
loan budget, and correct the pilot's growth arithmetic.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01QWPs69B8nQZjtPEUXMKzZx
Add pay.md: pay bands open to every worker, a ratio each co-op must
choose and measure honestly, a double majority for high pay and for
raises to jobs most members hold, and no one voting alone on their own
pay. Stop the reserve rule from charging co-ops for raising the bottom,
and close its gaps for perks, non-members and related companies.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01QWPs69B8nQZjtPEUXMKzZx
Add high-turnover.md: in home care, a 12-month probation and the
three-quarters rule could fail a co-op doing everything right. Make
membership automatic after a short probation, take the buy-in from
profit shares rather than pay, count agency hours and push-outs, and
phase the rule in. Align the lock, the pilot's safeguards and measures,
worker ownership, the index and open problems.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01QWPs69B8nQZjtPEUXMKzZx
Add conservice.md: the company as it is (from public sources only),
the same business founded as a locked co-op, what the profit split
could mean at its scale under labeled assumptions, whether it would
have grown, and who would have received less. Link it from the
thesis, worker ownership and the index.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01QWPs69B8nQZjtPEUXMKzZx
Add buyouts.md: owners who choose to sell a healthy firm to its workers
can borrow on a separate $100M track with its own test (money plus a
pay floor), a capped independent price, a seller note that shares the
risk, a vote of the whole workforce and the lock. Restate the total
worst case as $450M and answer open problem #12.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01QWPs69B8nQZjtPEUXMKzZx
Add "Making It Last": how public programs get killed (Ex-Im, CDFI Fund,
GGRF, NCCB) and survive (Bank of North Dakota, electric co-ops, Farm
Credit), and rules that assume any president can replace the board
after Trump v. Slaughter. Lending standards fixed in law, a "not tested"
verdict if good applicants are turned away, no freeze from board
vacancies, honest limits on where the money sits, and federal test with
state copies.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01QWPs69B8nQZjtPEUXMKzZx
Add savings-guarantee.md: a co-op-funded insurance pool covering final
unpaid wages, former members' payouts and a protected slice of current
accounts, priced for near-zero recovery by an independent actuary, with
a credit-union-style deposit, phased coverage, pro-rata payment when
short, and rules against phantom profits, early quitting and fraud
rings. Co-ops may no longer take member deposits or sell members bonds.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01QWPs69B8nQZjtPEUXMKzZx
Explain why a permanent lock has precedent (UK CICs, steward ownership,
US nonprofit conversions) and what it prevents (building-society
demutualisation). Enforce it in layers: an independent veto-only lock
share, articles, statute, and a backstop that sends the value to the
federation fund if the lock is ever broken. Treat piecemeal asset sales,
hire-backs and subsidiaries as sales; adapt Argentina's bankruptcy rule,
noting it needs a Bankruptcy Code amendment; state the lock's costs.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01QWPs69B8nQZjtPEUXMKzZx
Add precedents.md: Mondragon's Caja Laboral, the US National Consumer
Cooperative Bank, Italy's Coopfond and Marcora Law, each with what went
wrong. Add a support lottery to the pilot (intensive help for a random
half, advisers walled off from lending, all help counted against the
bank), count distressed mergers as closures, and replace unsourced
precedent claims on the bank and argument pages.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01QWPs69B8nQZjtPEUXMKzZx
Add the-long-run.md: the Furubotn-Pejovich worry, the French and
Italian evidence, the plywood "failure of success", and four rules:
the reserve takes its share of pay above the going rate, asset-sale
gains go to the reserve, qualifying development can be spread over
five years with abandonment charged to those who chose it, and
accounts earn Treasury+2 only while at risk. Update capital accounts,
worker ownership, the pilot's measures, open problems and the index.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01QWPs69B8nQZjtPEUXMKzZx
Add capital-accounts.md (profit split, cash share, formula interest,
payout rules, solvency test, savings protections, worked outcomes).
Rewrite founder-problem.md with a discounted co-op vs conventional
comparison, the risk evidence (Hall-Woodward, Hamilton), and bounded
founder tools. Add founder-related measures and safeguards to the
pilot pages, and renumber open problems.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01QWPs69B8nQZjtPEUXMKzZx
Add "The Pilot, in Numbers": a one-time $350M, ~400-co-op pilot with
a power-calculated sample size, loan limits, a checked definition of
market rate, two pre-registered deciding measures (fair survival with
a pay and membership guardrail, and a Treasury-rate hurdle for the
loan book), a matched SBA comparison plus a lottery at the margin, a
year-12 review, a fast-track vote on success and a self-executing
wind-down on failure. Built around the FONDES and Venezuela failures.
Add "Guarding the Pilot": rules against hidden-boss shells, capital
claim drains, split borrowers, relabeled rescues, zombie survival,
sabotage, a rigged review, and political favoritism.
Link both from the index, bank, test, capital and lock pages, and add
the pilot's weak points and the federal-vs-state question to Open
Problems.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01QWPs69B8nQZjtPEUXMKzZx
Replaces the chapters with the v3 foundation: new chapters for the bank,
the lock, the founder problem, the test and intellectual property; removes
conversion, conversion-tax and valuation. Rewrites the overview to lead
with "nobody is forced" and rewords draft-history references for
first-time readers.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Adds The Conversion Tax chapter. The tax is withheld in cash from each
buyout payment, the rate schedule is locked on conversion day, brackets
run on cash actually received, pre-bill gifts are added back to the
lifetime ledger, and above-market side payments to former owners count
as proceeds. It explains why taking the tax in kind was rejected
(Koontz, Horne, Tyler). It also corrects the 1950s precedent (capital
gains were capped at 25%) and states plainly how the recycled tax
lowers what workers pay. Open problems now cover proceeds vs gains,
treaty overrides and recession cyclicality.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01GsSJ98117VXm9t7rBra5NX
Adds a Valuation chapter. It covers market price, control-size arm's-length
sales, and a formula whose industry multiples come from real acquisitions
and whose profit figures are cleaned of related-party games. Disputes stay
in court (Monongahela), but each side files a final number, costs follow
reasonableness, inflated claims draw a penalty, and workers get an
independent public valuation counsel. Conversion brackets now apply to
lifetime proceeds and look through foundations and trusts. Open problems
are updated for the anticipation discount and conversion-tax durability.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01GsSJ98117VXm9t7rBra5NX
Renders the markdown documents in Content/Socialism: a landing overview at
/socialism and chapters at /socialism/{slug}, with a sectioned sidebar,
on-page table of contents, prev/next paging and cross-document links.
Pages are noindex and nothing on the site links to them. Underscore-prefixed
files (founding draft, progress log) are never served, deployed or committed.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On a fresh table the button is on seat 1, and seat 1 is where the first player
goes. Tapping it did nothing.
The dealer chip is a 22px disc, and a 22px disc is half the 44px a fingertip
needs, so it carried an invisible pad to catch the tap. The pad was moved twice
looking for somewhere to grow that wasn't already spoken for: outward first,
where it hung over the seat next door and gave that seat the button; then
inward, over "its own seat, which is a large target that can spare the corner".
The seat could not spare it. Measured on a 390px phone, a 72x46 puck: the pad
ran from the puck's own centre to its right edge and from its top to 3px off its
bottom, and the chip's handler calls stopPropagation. The seat holding the
button kept 52% of its face — 130 of 276 sampled points went to the chip, its
exact middle among them. Every other seat kept 97%. The button moves every hand,
so the dead puck went round the table with it.
And the chip it went to had nothing to say. Its only answer is "give the button
to this seat", and this is the seat that has it. The seat menu already knew that
and hides "Give the button" there; mid-hand the same no-op would have taken the
blinds off the felt and re-posted the same chips to announce that nothing
happened.
So the pad is gone rather than moved again — a fixed pad cannot be safe at every
puck size, and no finger reaches any chip now: the ghosts are pointer-events:
none without hover, and the one live chip is this one, which is now inert to the
pointer and still there to read. Giving the button is the long-press menu on
touch and the revealed ghost on a mouse, both unchanged. makeDealer returns
early on the seat that already has it, so the keyboard can't re-post blinds to
move the button nowhere either.
The sweep had a check for exactly this — "a seat's centre belongs to that seat",
written when the pad was reaching outward — and it ran clean through all of it,
because `seat.contains(hit)` waves through anything inside the seat and the chip
is inside the seat. The check written for this bug could not see the second half
of it. A control inside a seat is a rival for the tap now, and against the old
stylesheet it fails seven scenarios.
982 unit tests, 6 of them new; the layout sweep is 320 audits with the one
pre-existing failure (a $1,300 chip behind seat 11 on a twelve-handed 360x640
all-in).
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
The header stood 120px tall on a 390x844 phone and only 46 of that was a
control. The other 74 were a 22px page title with a gold rule under it, a back
link, and a gold "Hand #1" pill — three things read once, above a screen whose
whole subject is the green table underneath. Sixteen percent of the window,
spent before the felt started.
A phone held sideways had already solved this. The landscape block clips the
title and puts everything on one row, because at 390px tall there was no other
option. Upright never got the same answer, and upright is how this tool is
held. It is 44px now, the table starts at 78px instead of 152, and the felt
went from 496px to 570 — about fifteen percent more table on the same screen.
Nothing left the page:
- The title is still the document's <h1>, clipped rather than removed, so it is
still the heading a screen reader lands on and still what the browser tab and
the installed app are called.
- The back link keeps a 44px box and loses only the word "Casino", which is what
every phone does with a back link. The name a screen reader gets is on the <a>
either way, and it contains the visible word on the sizes that still show it.
- The hand number moved to the action bar's status row, in front of the street.
That row already says what hand you are in — the street, the pot, the player
count — and this is the fourth fact of the same kind; a counter read after the
fact had no business holding a line of the header all night. It costs the bar
nothing: the five audit scenarios that report the bar's height report the same
249, 227, 203, 227 and 227 pixels as before.
The three buttons take the width the title gave up rather than shrink-wrapping
to the right of it, which is 26px of extra target each. `flex: 1 1 0` was wrong
here — it divides the row into equal thirds and squeezes "Next Hand ⟳", 105px of
text, into 99, and these are nowrap, so what gives is the label. An `auto` basis
makes the label the floor and the spare width the bonus. On the narrowest phone
in the sweep the arrow and all three buttons come to 323px of 336.
The arrow was 32px on the first pass and the layout sweep caught it at 31x43 on
every touch viewport there is — 56 scenarios, one control. The word had been
carrying the width; with the word gone the arrow is the whole target. 44 square.
The sweep is down from five failures to one. Four of the five were seats and bet
chips colliding on a 360px Galaxy S8, and 76px more felt was all they needed.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
The seat sheet is the second-most-used surface here — every buy-in, every
rename, every correction goes through it — and it was the least finished one.
Seven things were stacked flat at one weight and all open at once. Measured on
a 390px phone it came to 594px and scrolled, and the three controls a dealer
touches on nearly every visit were 132px of that. The chip counter alone was
174px, the tallest single thing in the sheet, for a job most tables do once a
night or never; the two hole-card slots were 62px of dashed grey on every table
that never asked to be shown a card.
Cards, the contribution and the custom fields are behind a fold now, and which
of the three stand open is a question about the seat rather than a setting: a
seat holding cards opens its cards, a seat with chips counted opens its count,
a seat with a note on it opens its notes. A shut section says the answer on its
own row — the cards, the total, how many notes — so nothing already entered is
further away than it was. An empty chair is 419px and does not scroll; a seat
with everything filled in is 625 and opens all three.
The fields have the width the sheet always had. `.fieldRow input` caps at 140px,
which is right in a 320px sidebar of settings and wrong in a sheet holding a
whole phone: "Name" sat against the left edge with 203px of nothing between it
and a field a third of the row wide. Same 44px control, 300px instead of 140.
The chip counter's values are text. They were a `title` attribute, which is a
tooltip, which is nothing on the device this is used on — five identical number
boxes told apart by five coloured dots. It also left the wrapping <label> with
no text in it, so they had no accessible name either.
And the sheet has a scrim. It was the only one of the tool's seven overlays
without one, while covering the felt from 250px down and the action bar with
it — modal in everything but appearance. What stayed lit above it was the
header, with Misdeal, Next Hand and Resolve Pot live under a thumb reaching for
a name field. It dismisses on a tap and on Escape now, like the other six; a
card picker opened from inside still takes Escape first. In landscape the scrim
stops where the sheet stops, so Fold, Call and Raise keep the rail.
It is put away with visibility rather than opacity, which is the same lesson the
sheet above it already learned about transforms — and the layout sweep caught it
the same way. At `opacity: 0` a fixed, full-window box is invisible and still
entirely there: bottomFurniture() read it as furniture standing 640px tall at
the bottom of a 640px phone, in every state and on every screen, and asked
scroll-padding to reserve the whole window for it. Ninety-eight scenarios failed
on it; hidden is hidden, and the sweep is back to the five it started with.
The toast steps over the sheet the way it already steps over the action bar.
Both are fixed to the bottom of a phone, and a "Keeping the screen on" toast
was landing squarely on the seat form with its Undo and its × live above the
fields. The height is measured and published, because it changes the moment a
section is unfolded.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
The sweep has checked WCAG AA contrast for a long time and the page passed. That
result was worth less than it looked.
WCAG 2's ratio is a luminance gap, and a luminance gap overstates readability once
one colour is near black — which is every reading on a page whose ground is #141414.
Tuning a colour until it clears 4.5:1 there lands on text that passes the audit and
still reads as a smudge across a table in a dim room. Several rules in this
stylesheet said as much in their own comments, one of them "measured 4.5:1 at
#a5a5a5 exactly, so it sits above that": the right instinct through the wrong lens.
The check also could not see half the backgrounds it scored against. It reads
background-color and takes the best contrast of any candidate ancestor, so text on
the felt was scored against the dark page behind the felt. Two plain AA failures sat
in a clean report for as long as the check has existed — the chair number on the
brown rim at 2.65:1, and the pot's own label, gold on green, at 2.66:1.
runContrastAudit.js scores with APCA and gets the background by screenshotting the
page with its text made transparent and reading the pixels, which no gradient or
shadow can fool. The first version walked the DOM and called the gold dealer button
black-on-black at Lc 0.4; that false positive is why it works this way.
Found: 65 of 92 pieces of text below their level, 54 passing AA at the same moment,
worst Lc 28. Now nothing is below Lc 60, nothing fails AA, and the worst is Lc 60.4.
Three fixes worth naming, none of them "make the grey lighter":
- A player's stack was secondary text at Lc 44, under a name in --text at a heavier
weight. That says a dealer needs the name more than the number; during a hand the
opposite is true. Hierarchy comes from size and weight now, which is free, rather
than from fading, which on a dark ground is the most expensive way to say it.
- An empty seat had no background of its own, so the felt decided how readable its
number was: Lc 28 on the rim, 30 on the green, 33 further in. It has a ground now,
at 0.62 and not the 0.82 first written — contrast is nearly flat across that range
and the heavier wash put four dark rectangles on the felt like holes.
- The pot's label sat on the one green in the design. The disc is deeper rather than
the label bleached, because gold on dark green is the pairing this page is built
on. Deepening only its outer stop moved it two points: the highlight is at 32%/28%
and that is exactly where the label sits.
apca.js is checked against the eight published reference pairs, png.js against
round-tripped fixtures for every filter. Four fault injections caught; a fifth
turned out to be a no-op — the Paeth predictor is symmetric in its first two
arguments over all 16,777,216 byte triples, so that swap is untestable, which is
written down rather than papered over with a fixture that appears to catch it.
Still below the strict APCA level: 55, all of them 10-12px labels that would need
white type or a bigger seat puck. Reported, not chased, and said out loud in the
sweep's own notes.
Layout sweep unchanged: 5 FAILs, 14 offender lines, no regressions. 950 tests pass.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
A focused <input type="number"> is a spinbutton, and a browser lets the mouse
wheel drive it. Measured on the running page at 1280x900 with the sidebar
open: focus the small blind and scroll, $5 becomes $5.50. Focus a seat's stack
and scroll, $500 becomes $501. Both go straight through to localStorage,
because these fields save on `input` like everything else here.
Nothing is announced and nothing looks wrong. The tool has invented a chip, on
a table whose whole purpose is that the chip count is right, and the gesture
that did it was a dealer reaching for another panel.
The rule is narrow on purpose. Every other way of driving a spinbutton is
deliberate -- the arrow keys, the stepper arrows, typing -- so scrolling is
the only one taken away. Verified after: the value holds, one ArrowUp still
moves it, typing still works, and the page still scrolls.
Blur rather than preventDefault. Cancelling the event stops the value moving
and stops the page moving with it, which trades a silent bug for a sidebar
that mysteriously will not scroll -- a problem the dealer can see and cannot
explain, instead of one they cannot see at all.
The decision lives in Money as a predicate beside selectsOnFocus, so the rule
is testable without a browser and main.js only does the blurring.
Two things checked rather than assumed. The tagName guard in the predicate is
decoration: deleting it passes every test, because nothing but an <input>
carries type="number". That is now said in the comment instead of being left
to look load-bearing. And the page's one <select> -- which way the blinds move
-- is not vulnerable: focused, hovered and scrolled 300px in Chromium it stays
where it was, so it is left out of the rule rather than defended against.
916 tests pass, 6 of them new. Four fault injections into the predicate are
caught; a fifth was applied, not caught, and turned out not to be a fault.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
The stylesheet had honoured prefers-reduced-motion for a while, in the three
places it mattered least: a 2px press-down on a button, a checkbox tick, and
the acting seat's pulse. What none of them covered was motion that moves
something across the screen, which for anyone with a vestibular disorder is
the part that matters. A button dipping under a thumb is not what makes people
ill.
Four things were doing it, and the three largest were in JavaScript where a
stylesheet cannot reach: the seat editor sheet, which slides the full height
of itself up from off-screen every time a dealer taps a chair; the toast; and
three scrollIntoView({ behavior: "smooth" }) calls, the most provocative of
the lot because what moves is the whole viewport rather than one control.
The scrolls go through Render.bringIntoView now, which reads the setting per
call rather than at page load -- a dealer who turns it on mid-session should
be taken seriously without reloading a table they cannot afford to reload. The
toast keeps its fade and loses its travel, because something that appears
already in place reads as a rendering glitch rather than as news. The sheet's
visibility delay flips at once, or it would leave a sheet that has already
jumped away still painted for a quarter of a second.
The press-down block held nine :active selectors and looked deliberate. It was
written to stop a stuck press state when a tap turns into a scroll, and had
reached for the reduced-motion query as the nearest proxy -- so it covered
.dealerChip and .raiseQuickBtn while missing .actionBtn, the class on Fold,
Call, Raise and Post blinds & deal. Read as its own rule the membership is not
a judgement call: every :hover and :active transform in the file belongs
there. There are sixteen. Nine were listed.
The scan that found the other seven took two goes. A line-by-line version
walked straight past a deliberately broken build, because the injected rule
was written on one line and neither "opens a rule" nor "is a transform"
matched. Parsing blocks caught it -- and then failed on the clean file, which
is how .toastAction, .toastClose and .potWizardChoice came to light: a
multi-line group had been truncated to its last line, so nothing had ever
looked at them.
910 tests pass, 16 of them new. Six script and stylesheet fault injections are
each caught, and the reduced/no-preference split is verified in a real browser
against both context settings.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
315 of 320 audits pass. The five that do not are one root cause on the two
smallest screens: twelve seats want about 60px of footprint each and the ring
has 44px of rim to give them, in the two states that make the bar tallest.
Written down rather than left to be rediscovered, because closing it is a
design decision -- drop the bet chips at extreme crowding the way the tight
tier already drops the hole cards, cap the seat count by viewport, or let the
felt scroll -- and each trades something a dealer can see.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
The sweep had carried the same residual for rounds: on the two smallest
phones with twelve players, seats painting over each other's names and
stacks. It read like a geometry problem. It was a wrapped label.
"ALL IN" is the only thing that ever lands on the stack line with a space in
it. A stack figure has no spaces, so it can only overflow, and the
puckAmountLong/Huge sizes exist to stop that; nothing had asked what the line
does when it can break. Measured at 360x640 with twelve players all-in, it
wanted 36px and had 34, broke onto a second line, and an 11px line became 22.
That is not a typography complaint, which is why it survived: it looks fine.
The puck went from 55px tall to 66, and puck height is what the ring is short
of -- at the ends of an ellipse consecutive seats are separated vertically, so
a puck 11px taller is a puck 11px further into its neighbour. The layout
search then did exactly what it is written to do: tried every width it has,
found none that cleared, and fell back to the narrowest, which is also the one
that wraps hardest.
So the rule every design system gives a badge applies here too: one line,
always. Not wrapping is only half of it -- a line that cannot wrap and still
does not fit paints over the seat next door instead -- so on the tight tier
the mark gives up its letter-spacing, 3.6px of the 36 it wanted, and keeps the
warm colour and the bold weight that do the same job.
Halving the overlap did not clear it, and the rest of the shortage was not on
the felt. The sweep had been saying so in a warning rather than a failure: at
360x640 the bar was 267px of a 640px window, 42%, against its own "persistent
chrome under 30%". Two things in it were being shown when they could not be
used. The side-pot list is decision support, and during a run-out there is no
decision to support -- it collapses to the main pot and one line for the rest.
The status row said the player count twice in the same breath, as a chip and
again inside the run-out sentence beside it.
Measured: bar 267 -> 249 on a Galaxy S8 and 267 -> 227 on an iPhone SE, felt
243 -> 283 there, worst seat overlap 34px -> 4px. Across the sweep, overlap
offender lines 33 -> 9, chips behind seats 30 -> 5, failing audits 6 -> 5, and
the iPhone SE's seat-text overlap gone entirely. No new failures.
894 tests pass, 9 of them new. Seven fault injections are each caught.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
Three failures that only ever happen at a table, none of which show up on a
desk with a good connection and a plugged-in laptop.
The offline promise depended on how the URL was spelled. Razor Pages routing
is case-insensitive, so /Casino/TexasHoldem serves the identical page; a
service worker's scope is matched as a plain string prefix, so a worker scoped
to /casino/ does not control it. The registration succeeded, the worker
activated, nothing was logged, and the cache stayed completely empty --
because the precache list was handed to navigator.serviceWorker.ready, which
resolves with the registration controlling the page and so never resolves at
all outside its scope. A dealer who opened that URL to load the tool for the
night had loaded nothing.
One spelling wins now, via a 301 scoped to /casino rather than applied
site-wide: every file under wwwroot/casino and wwwroot/images/casino is
already lowercase, so lowercasing there cannot miss a real file on a
case-sensitive filesystem. That guarantee does not hold for the rest of the
site, and /chessHub would break. The handshake is fixed rather than left to be
saved by the redirect -- it takes the worker off the registration itself.
The manifest icons were never cached, because the shell list is read off the
DOM and the DOM never mentions them. They are read out of the manifest now,
cache-first so an offline launch stops logging ERR_INTERNET_DISCONNECTED, and
the link sweep picks up apple-touch-icon, which is the only one iOS reads.
The screen slept mid-hand. A wake lock is held while betting is open and
dropped the moment it is not: through a whole street of renders it takes
exactly one, two syncs in a tick take exactly one, a hand that ends inside the
request leaves nothing held, and the browser taking it back leaves us able to
take it again. Announced once with a Turn off button rather than a permanent
header indicator; a setting after that.
And localStorage is evictable, which the Save & reset panel now says out loud.
persist() is asked for on Post blinds & deal, because every browser wants that
request to come from something the dealer did.
885 tests pass, 18 of them new. Five fault injections into the wake lock are
each caught. One of the new tests was green for the wrong reason first:
globalThis.navigator = fake does nothing, because Node has defined navigator
as a getter since v21.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
The card feature had two halves and only one of them ever asked a question. Turn the
setting on and every street opens a sheet for the cards coming off the deck — that
worked, on a phone and on a desk, in a normal hand and in an all-in run-out. The
other half never asked at all. A dealer who entered all five community cards, pressed
"Showdown — resolve pot", and got a wizard saying
Christopher, Mike from work, Alexandra and Bob haven't turned over, so who
won is still yours to say.
with nothing in that sheet to turn anybody over with. The only route in was to close
the wizard, tap a seat puck, find the two card slots in the seat editor, do it four
times and re-open the wizard, with the pot on the table and four people waiting. The
one moment the tool is certain to know the hands is the moment it stopped asking.
So the showdown button asks. The sheet opens on whoever the bar has just named as
showing first, takes their two cards, and moves round the table from there; hands
already entered are skipped, because a hand entered is a question already answered.
Four sheets later the wizard opens reading "Chris wins $40 with three of a kind,
nines" and its button says "Pay Chris" instead of "Confirm Winner(s)".
Leaving the sheet stops the asking rather than skipping a player: closing it says the
tool has the wrong end of the stick, and the answer to that is to get out of the way.
It isn't a dead end — the wizard still names whoever didn't turn over, the bar still
says "Showdown — resolve pot", and pressing it again resumes at the player the dealer
stopped on, because the list is read off the cards on the table rather than remembered.
CardPicker.close() takes a reason so the chain can tell "that sheet filled its last
slot" from "the dealer left". Two callers pass nothing and don't care.
The setting kept its behaviour and lost its jargon: "Track the board and read the
showdown" was the feature's name rather than what turning it on does to your evening.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
The green is a promise: this is the step. Measured over a whole hand — seventeen
steps — it was on two controls at thirteen of them, the bar's Call and the header's
Resolve Pot painting the identical gradient at opposite ends of a phone, and on the
header alone at the other four while the bar held the only button there was to press.
At showdown the most consequential tap in the hand, "Showdown — resolve pot", was
blue-grey while the corner shortcut was green.
So the accent moves to .actionAdvance, which is always the next step by definition,
and comes off the header shortcut, which never is. Twenty-two measured steps of a
real hand now show exactly one.
The sweep checks it: oneNextStep() counts the accent-coloured controls that are
actually on screen, scoped to an open sheet when there is one. Six broken versions
were run at it, and the two that took three goes to catch were both the same mistake
in the injection rather than a hole in the check — a button painted green with the
panel shut isn't rendered, and one painted green below the fold is deliberately
ignored. Scrolled into view it is named alongside the bar's Call, and goes clean
again when the green comes off.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
The sidebar is eight <details>, and every one measured exactly 60px shut with
nothing on it but its own label — 480px of a phone's screen, and 480px of a 660px
desktop column, spent saying "Standings", "Hands", "Waiting list". Whether the box
balances, whether anyone is waiting for a seat, how many hands have been played:
the page had all three worked out already, and charged a tap for each of them, on
rows two thirds of which were empty.
That is the disclosure pattern's named failure — a header you have to open to find
out whether it was worth opening — and the answer the design systems give is the
same one: put the status on the header. The panel keeps the sentence. "Off by
$300" is the alarm; which seat is holding chips nobody wrote down is a paragraph,
and that is the half that belongs behind the tap.
Two rules, and panelGlance.test.js checks both. A glance appears only when it has
something a dealer would act on — eight rows that always say something are eight
rows that say nothing again, and a row reading "Nobody waiting" all night is the
clutter this went in to remove. And a glance never says something the panel under
it would contradict: it is read off the same state, never off the panel's DOM,
which for a shut <details> isn't rendered at all.
Absolutely positioned rather than floated. A <summary> only draws its disclosure
triangle while it is display: list-item, so it cannot be a flex row; the float
that replaced it kept the row at 60px but aligned to the label's line box, and a
12px string in a box sized for 15px text rode 2.5px low on all six panels. Out of
flow is also the smaller promise — an absolutely positioned box cannot make its
row taller at all.
Three paths deliberately skip renderAll, because it would take the field out from
under whoever is typing in it, and all three move something a glance reports. The
notes textarea, the chip legend's value fields and commitSeatChange each call
renderPanelGlances on their own instead; without that the note you had just typed
was summarised by the note before it until the next tap landed. A space between
the label and the glance, too: the summary is announced as one button, and without
it a screen reader was given "Table Setup9 seats".
Costs 0.08ms of a 4.79ms renderAll on a twelve-handed phone. Four injected faults
in the wording are caught one test each.
glancesFitTheirRows took three goes. The first asked only whether a row had grown
taller than a bare one, and a glance floated back into flow and allowed to wrap
sailed straight through it, because a float does not expand the box it sits in.
The second was caught by the sweep on its first clean run — 94 failing configs, 91
of the offenders this check's own: it takes the shortest row as its idea of a bare
one, and the Keyboard panel is display: none under pointer: coarse, so on every
viewport visited with a thumb the bar was a row that wasn't drawn, and all seven
real rows read as grown. Rows that aren't rendered aren't rows. The four broken
layouts are re-run with that panel hidden and each is still caught.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
The felt grows with the window and the pucks did not: 96px was the only
ceiling there was. Nine-handed, the seats covered 19% of the felt on a phone,
11% on a 1280px desktop and 8% at 1440px and beyond — and the stack figure a
dealer reads sat at 10px on a 2560px monitor, the same 10px as on a 360px
phone, well under every published floor for text meant to be read.
Two things stop this costing more than it buys.
Every candidate width is a forced layout. Simply raising the ceiling took the
layout pass from 24ms to 73ms twelve-handed on a desktop, so a screen that
cannot use the wider widths must not measure them: the extra pass is offered
only when the arithmetic already says the seats have room to spare, and a phone
measures exactly the ten widths, in the same order, that it measured before.
Timed with the order alternated so neither carries the warm-up: 10.7ms against
10.4ms on a phone, and +3 to +6ms where the widths are actually used.
And a wide puck on a narrow felt is worse than a small one. The first version
of this let phones through, and a 144px puck on an iPhone's 278px ring is half
the width of the table: six-handed, all six seats sat over the rail by 15px
against 5px before, reading as a stack of cards rather than a table with people
around it. So the ceiling is a fifth of the felt's shorter span, capped at 144.
Five is measured rather than chosen — the shorter span is 233-303px on every
phone in the sweep including sideways and 617-714px on tablets and desktops, so
any divisor between 3.2 and 6.4 separates them — and the test pins both ends.
Above the ordinary ceiling the puck takes the larger type as well, which is the
point of the extra width. Nine-handed: an iPad goes 96px/12px to 134px/15px, a
1440px desktop 96px/12px to 140px/15px, and the seats go from 8% of the felt to
12%. Every phone, at every seat count, is untouched.
The sweep is byte-identical to before this change: same six configurations,
same 63 offenders, empty diff.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
A seat puck carries the name, the hole cards, the amount that player has put
in, and their stack. Two of those are empty most of the time and both held
their space anyway.
The amount pill was worse than empty space. It has a background, a pill radius
and side padding, so with nothing in it, it painted an 8x11px green lozenge —
measured on seven of nine seats at an ordinary table. One of those is a
component holding its place; nine is a rash, and a dealer has to learn to
ignore a colour that on the other two seats means "this player has money out".
The space was not free either. seatRingRadii sizes the ring from the tallest
puck on it, so two rows nobody had filled were setting how far apart the whole
table had to sit. Across 21 tables — seven viewports, six, nine and twelve
handed — every single puck comes back shorter and six come back wider: an
iPhone 14 nine-handed 78x88 -> 78x63, twelve-handed 44x65 -> 66x55, which is the
one name in the set that was clipped and now is not.
The collapse is only safe because the layout cache moved with it.
resolveSeatLayout measures ten candidate widths, each a forced layout, against
seatShapeSignature() — which listed the names and the dealer seat, everything
that could wrap a name onto a second line. It did not mention these two rows,
because while they always held their space they could never change a puck's
height. Now they can, so the signature names them: booleans over the whole
table, because measuredSeatBox takes the tallest puck on the ring, and because
a per-seat key would be correct and would throw the cache away on every tap.
The ring settles by 24-26px when the blinds go in — and it settled by 23-25px
there before this round too, since the action bar changes height at that moment
and the felt is measured against it. The one place it moves is a place it
already moved.
uiAudit gains emptyDecoration(): nothing on the felt may paint a background or
a border with no content in it. Its one exemption is on the list because the
check reported it rather than because it was predicted — .betChipDisc is a
circle filled with the denomination colour of the chip beside it, so the colour
is the content. Three others were written from memory first and matched no
element on the page at all.
The sweep is unchanged at 6 failing configurations and 63 offenders. All four
table failures are the all-in shape, where every pill is filled and the
collapse buys nothing; the ordinary tables this helps were already passing, so
what moved is room a name has rather than whether anything was covered.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
The bar is fixed over the table, so its height is the felt's budget — and it
carried a pot breakdown that drew one row per all-in with no ceiling. On a
360x640 phone: five players idle put the bar at 151px, six all-in at 303px,
twelve all-in at 411px. That is 64% of the screen, leaving 96px of felt to draw
twelve seats in, and at that size the seats stop being a table: five pairs
collapse onto each other's centres, ten hang past the rail and a dozen labels
sit completely covered.
Past four layers the bar now draws the main pot, one line for everything
between, and the newest side pot. The two ends are the live ones for the same
reason — side pots are paid from the newest back down to the main pot, so the
top layer is the next thing that happens and the main pot is the last, and
those are the two a dealer says out loud while the chips are stacked. The
middle is the rest of that count-down, wanted at showdown, which is the pot
wizard: same builder, no cap, in a dialog that scrolls. The elided line carries
the count and the money, so the visible rows still add to the announced pot.
Nothing in the sweep could see any of this, and the reason is worth keeping.
Every check asks whether an element is covered or clipped; none asked how much
room the covering element had taken. But the deeper miss was the sweep itself —
all 294 audits drove their table with Call and Check, so no table it ever
measured had a player all-in, and the one thing on the bar whose height depends
on the table was empty in every measurement. There is now a fourth table shape,
twelve players shoved in for twelve different amounts, and furnitureBudget()
measures the fixed band against the window: failing over half of it, warning
over the 30% that persistent chrome is usually held to.
Galaxy S8, twelve all-in: bar 411px -> 267px, felt 96px -> 192px, 91 layout
offenders -> 25. iPhone SE: 73 -> 10. The rest is the seat ring on a short
phone, which the sweep can now see.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
WCAG 2.2 SC 2.4.11 asks that a focused control not be completely hidden
by something the page has put on top of it. The bar is fixed over the
bottom and is not in the flow, so the browser does not know it is there:
Tab to a control near the bottom and it is scrolled flush against the
edge of the window, under the bar, and the browser considers it done.
Tabbing the whole page measured 16 stops resting completely covered on a
390x844 phone and 27 more partly, plus four under a toast on the desktop.
It is now 0 and 0 on desktop, phone and a phone on its side — clean
against 2.4.12 (AAA) as well, which allows no overlap at all.
Three things were wrong and they needed different answers.
scroll-padding-bottom on the scroll container, sized from the bar's own
published height and the toast's, handles the case where the browser has
to scroll to reach the control. That is half of it, and the half that
looks like all of it: a control the browser already considers visible is
one it will not scroll to, and a fixed overlay on top does not enter into
that decision. Measured — tabbing to Common rulings with a toast up left
it 100% covered and scrollIntoView({block:"nearest"}) moved nothing. So a
focusin handler measures the overlap and scrolls by it, and when the page
has no room left (26px against 94px wanted) the toast gives way instead.
Sizing the page padding for the toast would have fixed that by making the
page grow and shrink 64px whenever anything was announced.
The page's room to scroll past the bar was 9rem on the sidebar alone,
which left the table's column — most of a desktop — with none. It is now
on the whole app and taken from the bar's real height, so it is also 0 in
the layout where the bar is a rail down the side.
And the seat editor's bottom sheet was put away with a transform, which
moves a thing without putting it away: its Done button sat at y=850 on an
844px screen, focusable, announced, and impossible to scroll to, because
a fixed element does not move when the page does. visibility:hidden puts
it away, transitioned so the slide-out still runs.
uiAudit.js gained focusHasRoom() and focusablesReachable() for the two
static halves. Audit unchanged at 2 failing configurations and 3
offenders, both pre-existing. 805 unit tests pass.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
Deleting a chip colour was one click of an unlabelled 24px button, and it
was never a change to a legend. computeContribution() prices a seat by
running the legend over that seat's chip counts, so a colour that leaves
takes every chip counted in it off the table.
Measured in a browser on a three-handed table: deleting the $500 row took
$22,500 of $26,874 with it. No confirmation, nothing said afterwards, and
no way back — a colour added again is a new colour with a new id, while
the counts stay keyed to the one that left. The counts were left behind
too, so the saved file carried money nothing could price and nothing
could show.
Now: a colour nobody has counted still goes on one click, and one that is
holding money asks first, naming the amount, the chips and the seats, and
saying that Undo will put it back while adding the colour again will not.
Conditional on purpose — a confirmation that fires every time is one
nobody reads by the third night, and four others already compete for that
attention here. Removal goes through runUndoable, and the counts go with
the colour so the file matches the screen.
The number in that warning is the difference between pricing every seat
with the colour and without it, not chips times value. The obvious
arithmetic is wrong both ways: a seat with a typed amount is never priced
from its counts, and a seat whose chips all leave falls through to the
table's standing ante rather than to zero. A randomised check over 2,000
tables asserts the warning equals what the table then does, and it found
a third case — fourteen $1 chips against a $200 ante, where removing the
colour puts the table UP $186. The wording says so in that direction too.
The × also had no accessible name, so uiAudit.js gained controlNames():
every control needs a name containing a letter or a digit, resolved the
way the accessibility tree resolves it. A symbol is not a name. Taking
the new aria-label off again makes it report all five buttons.
Audit unchanged at 2 failing configurations and 3 offenders, both
pre-existing. 805 unit tests pass.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
"Session" was a container word rather than a name. It held the standings,
a waiting list, the hand log and the save/load/reset buttons — a topic, a
task, a log and three utilities under one label, which is a navigation
layer with more than one logic in it.
It measured 1206px and 103 visible elements, against 555 and 29 for the
next biggest, and "The table itself" sat 1052px down inside it. Reaching
Reset table meant opening a panel called Session and scrolling past the
money and the log to get to it.
It is now Standings, Waiting list, Hands and Save & reset, each named for
the thing a dealer came looking for, in the order of a night: set the
table up, watch the money, fill the seats, look back at a hand — and,
last and apart, the two buttons that end it. Reset table goes from 1052px
inside a panel called Session to 62px inside one whose label says what it
does. The biggest panel drops from 1206px to 607px. The four subheads
that used to carry those names are gone, because the summary says it now.
Every panel name is also a heading. A <summary> is announced as a button,
so the sidebar's names — its whole navigation — were unreachable by
heading, and splitting Session would have removed the only four headings
down there. The outline is now h1 page, h2 per panel, h3 for the two
subheads that remain. Visually identical: the h2 gives up everything a
heading brings with it.
The cost is 216px more collapsed summaries at rest, three more one-line
labels. Audit unchanged at 2 failing configurations and 3 offenders, both
pre-existing. 795 unit tests pass.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
The pot breakdown listed every eligible player for every pot. Pot
eligibility is nested — the main pot is everyone still live and each side
pot above it drops whoever was all-in below — so the main pot's line was
every name at the table and each side pot repeated it minus one.
The lines too long to fit were exactly the ones nobody needed spelled
out, and the short ones at the bottom were the side pots a dealer
actually announces by name. Measured on an iPhone 14: a single all-in
five-handed already clipped the main pot, and twelve-handed the first row
wanted 521px of a 254px slot. There was no count and no title, so a
dealer reading "Alexandra…" could not tell whether three people or eleven
were eligible — the number the showdown turns on.
Each line now names whichever of "who is in" and "who is out" is shorter,
which nesting guarantees is never more than half the table, and leads
with the count because the line is right-aligned and it is the tail that
gets cut off. "all 9", "8 in, not Mike from work". On a phone this takes
one, two and three all-ins from every main-pot line clipped to nothing
clipped at all; twelve-handed goes 6 clipped rows to 5, and even a cut
row still opens with "9 in".
The wizard's copy of the same announcement wraps instead — it is a dialog
that scrolls, and it is what the dealer reads out before anybody shows a
hand — so nothing there is ever cut off. The full list is also on the
title for anyone with a pointer, which is the extra rather than the
answer, since a tooltip is not a mechanism a keyboard reaches.
potCalculator.test.js now proves the nesting across 20,000 random tables
and asserts on its own generator, so the loop cannot go vacuous. Audit
unchanged at 2 failing configurations and 3 offenders, both pre-existing.
787 unit tests pass.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
A portrait phone got a square felt, so the vertical room the layout had
already measured was published and then dropped: 522px of room under a
366px table on an iPhone 14, 638 under 406 on a Pro Max. The cost was
seat names — 44px pucks with five of twelve clipped.
--feltAspect: 1.5 spends it. Twelve seats at 390x844 go from a 44px puck
with five names clipped to 66px with one; five seats go from three
clipped to none; the Pixel 7's twelve-seat ring goes to zero. 1.6 and
1.7 measure byte-identical, so 1.5 is where the sweep stops paying. The
three smaller phones are room-bound at any cap and measured identical
from 1.15 to 1.7.
The first version of this was a no-op. The rule went into
@media (max-width: 700px), and @media (max-width: 900px) below it set
the same property back to 1 — both match a phone, same specificity,
later one wins. It measured as 1 on every phone while reading perfectly.
The 900px rule is now a 701-900px range, and feltShape.test.js resolves
--feltAspect through the media queries the way the cascade does so the
next one of these is caught.
Also tried and removed: a floor under the max-height, so a tall bar
couldn't press the table flat. Twelve all in with nine side pots takes
the bar to 341px, leaving 142px of room at 360x640; floored to 242 the
felt keeps its shape by putting five of twelve seats behind the bar,
unfloored it is flat with all twelve visible. The slit it defended
against came from forcing the custom property by hand. A test asserts
its absence with the numbers.
Audit: 4 failing configurations and 5 offenders down to 2 and 3, both
remaining ones pre-existing. 776 unit tests pass.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
Two things that were wrong at every desktop width, both of them the same shape: a
number that was right about one thing being used for another.
The action bar was centred on the window. The table is centred on its own column, and
the sidebar takes 320px out of the right of that — so the two things a dealer is
actually looking at sat 172px apart, at 960px wide and at 1600px and everywhere in
between, because the offset is half the sidebar and the sidebar is a constant. The bar
also finished 80px inside the sidebar's column at 1280, painting over whichever panel
happened to be at that height. It leans away from the sidebar now, by the sidebar's own
width, which comes out exact above and below the 1400px cap — measured on the running
page at four widths, 0px off at all of them. No 100vw in it, which is the point: the
bar's own centring comment warns that on a page with a scrollbar 100vw is wider than
the viewport, and the offsets are exact without it.
The dealer button's chip sits in the puck's top-right corner and the name kept out from
under it with `padding-right: 22px`. That takes the chip's width off *every* line of the
name, and the chip is on one of them. Measured at 1280x900: the name box is 78px,
"Christopher" is 66 and fits, and on the seat holding the button it was cut to 56 and
broke as "Christop / her" — mid-word, no hyphen. The button moves every hand, so that is
a player whose name reads properly for four hands and comes apart on the fifth.
A float is what indents only the lines it sits beside. Line one gets 56px and the rest
get 78, so a name too long to fit beside the chip drops below it whole instead of being
broken to fit next to something that isn't there.
Not on a crowded ring, though, and finding that out cost a sweep. A tight puck gives the
name one line on purpose, and one line has nowhere to move to: with two lines granted
the name dropped onto a second line that the chip's lower half was still sitting on, and
the audit called it at every viewport in the list. So the float is for the ring with room
for two lines and the padding for the one without — two rules, matching two regimes the
puck already had.
The sweep also learned two things from this:
- Text is measured where it was painted, not where its box is. The seat-label check
modelled "text kept out of somewhere" as padding, which is one way of doing it and not
the only one; a float does it per line, and the check called every seat in the sweep an
offender while the text sat exactly where it was meant to. Range rects ask the question
directly, so padding, floats, centring, wrapping and a short label in a wide box all
come out right without any of them being modelled.
- Offenders say how much is covered, not that something is. "Is under a badge" reads the
same for a stack a dealer cannot read and for a badge's rounded corner grazing the last
digit. Probing the whole line and counting turns the residual this sweep has carried for
three rounds into "1/7 under SB" — one probe of seven, one configuration, thumb only.
294 audits, 4 failing configurations, 5 offenders: twelve players on a 360px phone with
the raise pad open (3/7 and 2/7, pucks genuinely overlapping — the search reports it
rather than papering over it) and the 1/7 hairline above. 770 unit tests.
Fault-injected, all restored and re-verified: writing the sidebar's width out again
instead of the token, growing the float to cover the chip's whole height, and taking the
padding off the crowded ring are each caught by a test; and with both the float and the
padding removed the rewritten label check still reports the name 6px under the chip,
which is the fault it exists for.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
A phone held sideways is 640x360: short of height and swimming in width. The layout
spent the whole of that shortage on one column — 70px of header, 85px of felt, 205px
of action bar. Two rows of seat pucks 55px tall have to fit inside 85px of felt, which
they cannot, so seven of nine finished past the rail with up to 17px of puck floating
on the page. That single fault was 96 of the 98 offenders the layout audit could find
anywhere.
No amount of shrinking fixes it, because the shortage was never in the table. The bar
is 205px tall and 529 wide to hold three buttons and a status line, and it was taking
that height from the one element that needed height. So in landscape it moves: the bar
becomes a rail down the right-hand side, the table takes the column beside it, and the
felt goes from 529x85 to 409x205 — an actual oval, with room for two rows of pucks and
a gap between them. It also puts Fold, Call and Raise under the right thumb rather than
along the bottom edge, where a landscape grip has the palm.
Which layout the bar is in is declared by the stylesheet in a custom property and read
back by actionBar.js, because the one thing that must not exist is a second copy of
that media query written in JavaScript.
Then the ring stopped predicting itself. seatRingRadii works out the widest ring whose
outermost puck still finishes inside the rail, and it is right about a puck it measured
before the width search settled and before a name wrapped or was shortened — each worth
up to 15px of height arriving after the answer. Four viewports finished with the top
seat 5-6px over. render.js now checks the drawn result the way the audit judges it: if
the worst puck hangs 6px past the rail, the ring comes in by 6px and they are placed
again. A layout that predicts itself has to be right about the future; one that measures
itself only has to be right about the present.
Also in this round:
- Names shorten at a word instead of being chopped mid-one. "Mike from work" on a
nine-handed phone was "Mike f…"; he is Mike. A name never ends on a joining word,
because "Mike from" looks cut off rather than shortened, and the whole name is on the
tooltip either way.
- The blind badge is counted in what a seat needs. It always points away from the table,
which was taken to mean it costs nothing between two seats — true of a straight rail
and false of a ring, where neighbours sit at different heights. Three tables had a
player's stack painted under the next player's SB or BB, and that badge outranks
everything else on the felt. It also got shorter, 15px to 12: clearing its own name at
5px of overhang instead of 9 hands every chair in the ring four pixels back, which at
640x360 is the difference between a 66px puck and a 48px one.
- Border-box, inherited, for the whole app rather than nine elements opting in one at a
time. Both elements that hadn't went wrong in the same afternoon: the rail overran the
page's padding by 25px and put Resolve Pot underneath itself, and the raise pad's
amount field pushed a horizontal scrollbar across a 640px phone.
- The raise amount field is thumb-sized. The rule that sizes fields for a finger names
the settings panels and the seat form, so the one field on the one surface built for
thumbs was 36px on every phone in the list. The sweep never saw it because it had no
step that opened the raise pad; it has one now, and that step also caught the rail
squeezing its own status line under the pad.
- mobile-web-app-capable alongside Apple's older name, which Chrome has been calling
deprecated in this page's console on every load.
The audit goes from 19 failing configurations and 98 offenders to 2 and 3, across 294
audits — thirteen viewports, three table shapes, a mouse and a thumb. What is left is
twelve players on a 360px phone with the raise pad open, where twelve pucks at the 44px
minimum genuinely do not fit on the felt; the search reports that rather than papering
over it. Each new mechanism was checked by breaking it on purpose: removing the settle
pass puts 26 offenders back, dropping the badge from the footprint 5, and un-sizing the
raise field is caught by the step that now opens it.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
An option — off by default, and off means the tool is exactly what it was.
Turn it on and the community cards go on the felt. Each street asks for the cards as
it is dealt: the picker opens on the flop the moment the flop button is pressed, and
again on the turn and the river, because the moment a dealer is holding those cards is
the only moment they know what they are without turning the board round. Three taps,
then one, then one. The picker is the same sheet as the hole cards, which now knows
about the middle of the table too — the seven of clubs cannot be on the flop and in
somebody's hand, and until now it could.
Turning the players' cards over is unchanged. That already worked the way it should.
What is new is what happens next. Resolve Pot stops asking who won and says so:
Alexandra wins $50 with three of a kind, nines.
with every hand at the table read out under its player's name, the winner already
ticked, and the button reading "Pay Alexandra". A chop comes out as "Christopher and
Mike from work split it, $22.50 each, with a pair of aces", both ticked. The dealer
still confirms — the tool reads the cards, a person still says the pot is good, and a
hand entered wrong is corrected by unticking rather than by starting over.
It declines rather than guesses. A board that isn't five cards, a player who hasn't
turned over, the same card entered in two places: each says which of those it is and
hands the question back, and the wizard is the one it always was. Who is *ahead* and
who *won* are different claims and only the second one can pay somebody.
Side pots are read per round, not once. The best hand takes the main pot and is then
out of the question; the side pot goes to the best of who is left, which is a different
answer and the one the tool has to ask for again.
handRank.js is the only module here that can pay the wrong person without anybody
noticing, so it is built to be checkable rather than clever: rank five cards, then take
the best of the twenty-one five-card hands in seven. Beside the written cases — every
category against every other, the wheel and the steel wheel, three kickers deep, the
counterfeited two pair, the board that plays — it is checked against a second evaluator
written a different way, on twenty thousand random seven-card hands and four thousand
random pairs of them. An ace that keeps its value in the wheel, a dropped two-pair
kicker and a flush read only by its top card were each introduced on purpose and each
caught.
The board takes the middle of the felt, sized from the room the seat ring actually
leaves — measured against every seat rather than the ring's widest point, which is only
the answer for a seat sitting exactly on the horizontal axis and had two pucks on top
of the board while claiming 77px of clearance either side. Where there is no middle to
take, it moves to a strip above the table: a phone on its side has a felt 528x83 with
55px pucks, so the ring closes over the centre entirely and a board in it is a board
with players painted on top.
That strip found an older bug. It costs the felt height, and a felt that changes size
had only ever been re-measured, never laid out again: the seats follow it on their own,
being placed in percentages, but a bet chip anchored inside a seat is placed in pixels
off that seat's centre. Measured at 640x360, the chips were positioned for a felt 83px
tall and painted on one of 60 — seat 3's $5 came to rest on seat 4's puck, and two
pairs of seat pucks were overlapping. The bar growing mid-hand does the same thing: a
side pot appearing is 55px off the felt. Both fixed.
The sweep grew a third table — the same five players in the same nine chairs as the
second, with the board on, so the only thing that differs between those two rows is the
board itself. It earned its place on the first run.
728 unit tests. 216 audits: what is left is the seat ring hanging past the rail on a
landscape phone, the known residual, plus five or six pixels more of it on the layouts
where the strip takes the felt's height, and a blind badge over a stack on a packed
twelve-seat portrait table.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
The chips were the last thing on the table nobody had measured, and the sweep had
never seen one: it audited a table that had never had a bet placed on it, so a third
of what is on the felt during a hand went unlooked at for the tool's whole life. It
posts blinds, deals and calls three times now before it measures anything.
What that found, and what this fixes:
- Chips accumulated. Seats are kept and updated in place rather than rebuilt, so
clearing the bets layer stopped clearing the ones anchored inside a seat and
nothing else ever did. An iPhone 14 had one chip per seat after the deal and five
by the third render; a dealer's every tap left another disc on the felt.
- The ring-or-cling decision asked how wide the window was. A phone held sideways is
844px across and gives the felt 117px: it passed the width test, took the ring, and
every chip on it was painted under a puck — the seats carry z-index 1 and the bets
layer carries none, so a chip that lands under one is gone, and with it the only
number saying what that player has in. Measured now, and only measured.
- Getting clear of a box was `edgeReach(puck) + edgeReach(chip)`, which is right only
straight up, down or sideways. Two axis-aligned boxes are clear when they are apart
on one axis, so the travel is the reach of the box you get by adding their widths
and their heights. At 768x1024 a $10 leaving a seat on the diagonal wanted 88px and
got 77, and came to rest 11px inside its own player's puck — every seat off the four
compass points, which is most of them.
- Every chip was laid out at the first chip's size. A chip is as wide as the number on
it: "$5" is 21px and "$10" is 27px on a phone, so each $10 was placed as if it were
6px narrower than it is. The widest is no better: at 768x1024 one 47px chip made the
whole table give up a ring the other four had room for.
- Anchored chips settled where they landed. At 640x360 the felt is 528x83 and a puck
is 55 of those pixels tall: clearing it takes 41px, the players facing each other are
70px apart, and both chips set off 41px into a gap that is 70 wide. Three overlapping
pairs on a twelve-handed table. Each chip now gives ground only to something it is
actually standing on, a pixel at a time — a midline rule or a single scale factor
both punish the seats at the ends of a rail, which were never going to hit anything.
- Nothing knew the pot total was there. Seat 1's $10 was sitting on "Pot $65".
Two things off the felt, from the same sweep:
- The toast outlived its news. "Next hand — button on seat 2. [Undo]" stayed up while
the hand was dealt and three players called: an offer to move the button, wired to a
snapshot that predates the hand, so taking it would have thrown the hand away without
saying so. It was also in the way of thirty small screens' seats and controls. The
bar drops it as it takes the next action; the bar's own Undo names what it would
actually take back.
- The pot dialog cut a name off. Three players in the pot on a Galaxy S8, two rows
visible, the third below a hairline that reads as the end of the list — a dealer
ticking the two they could see would have paid two thirds of a three-way pot. The
85vh budget was still subtracting the action bar's height from a dialog that has sat
above the bar since its z-index was fixed, and a player's ceiling moves beside their
name on a short screen. All three fit now, and where a list does run on, it runs on
under a fade rather than stopping dead on a row boundary.
The pot badge goes on a phone held sideways: 86x66 of a 692x113 felt, for a number the
action bar prints three inches below it, standing in the only band a bet chip has.
The audit stopped calling one thing another. An element clipped away by a panel that
scrolls is not covered by whatever is painted at that height — the wizard's third
winner came back for four viewports as "fully covered by button Confirm Winner(s)".
190 audits: 25 failing configurations to 7, and every one of those seven is the seat
ring hanging past the rail on a landscape phone, which is the known residual from the
last round. 663 unit tests, 44 of them new.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
Six players at a nine-handed table is the most ordinary table this tool draws, and on
every phone in the sweep one of them had their stack painted under the player next to
them. Measured across nine viewports and six table sizes: seventeen of fifty-four
tables hid a name or a stack behind something belonging to another seat.
Two causes, and the layout only knew about a third of one of them.
The felt decided how big a puck could be from a single number — the straight-line
distance between the closest two *neighbouring* seats, judged against a fraction of a
puck's width. Distance has one number in it and a puck is half again as wide as it is
tall, so the same 84px was generous side by side and nowhere near enough on the
diagonal. And neighbours are the wrong pairs to ask about: on a phone held sideways the
felt is 528x87, the ring is very nearly a straight line, and the seat that lands on top
of you is the one across the table. Ten of the tables in the sweep had a collision that
a neighbours-only check cannot see — seats 1 and 7, overlapping by 68px, while every
adjacent pair had room.
seatsCollide() asks the real question instead: do any two of these boxes want the same
pixels. Two equal axis-aligned boxes overlap exactly when their centres are closer than
one box in both axes, so it is an answer rather than a proxy, and it looks at every
pair.
The second cause was the small blind badge, sitting at `left: -6px` — hanging off its
own puck's left corner onto whatever was beside it. On an iPhone Pro Max at six seats,
with the pucks not touching at all, "SB" was painted across the next player's $1,137.
All three badges now sit inside the puck's width, and in the same place as each other:
a seat only ever wears one, so two positions bought nothing except making the dealer's
eye find which of two spots this seat was using before it could read which blind it is.
Then the layout had to be able to act on the answer, and it couldn't. It chose between
two widths — 96px or 58px — and if neither fitted it widened the ring and let the seats
overlap, on the reasoning that a dealer can see an overhang but cannot see a tap landing
on the wrong player. That trade is real; it was also a false choice, because the tables
that overlapped did not want either number. At nine seats a Galaxy S8 had room for a
61px puck, an iPhone 14 for 68, a Pixel 7 for 74, an iPhone Pro Max for 79. Three media
queries declared three more of these numbers — 78px, 64px, 62px — at three breakpoints,
guessing at something a media query cannot see: the same 390px phone has room for a 90px
puck at four seats and 61 at nine.
So the width is searched for now, not chosen. render.js walks candidates from the top,
applies each one, measures what the browser actually drew, and takes the widest that
leaves every seat its own box. There was a width that worked on every table in the
sweep; the code was never asking for it. Ten candidates is ten forced layouts, so the
result is cached against the felt's size, the seat count and the names — measured 12.5ms
cold, 0.8ms warm, and renderSeats runs after every tap.
Three things this got wrong on the way, all caught by measuring rather than reading:
The badge overhang was folded into the space between two seats. That collapsed the
ring on a phone held sideways — 87px of height, and 18px of badge is most of the room
the ring has to move in — until a six-handed landscape table was unsatisfiable at any
width. It didn't belong there: the badge points *away* from the table at both ends, so
two seats facing each other across the felt wear them back to back.
box-sizing: border-box, which the puck needed so the searched number means the width
it is drawn at, quietly turned every min-height from a content height into a drawn one.
An empty chair went from a 52px tap target to a 42px one on every phone in the sweep.
The audit found it; the four min-heights have each gained the padding and border they
used to be quoted without.
The new audit check reported a name under a neighbour on a table whose seats were 9px
apart. A puck's name is a `-webkit-box` with `line-clamp`, and getClientRects() hands
back a rect for every line the text *would* have taken, including the ones the clamp
throws away — so the probe was hit-testing 30px below the puck, where nothing is
painted. It clips to what is painted now.
uiAudit gains "no seat covers another seat's writing", the complement of the check
already there for a puck's own decoration on its own name — which is why nothing caught
this: seatLabelsClear() only ever looked inside one seat. Fault-injected against the
previous version: eighteen configurations, naming both causes.
And the sweep itself now runs two table shapes. It had run one for its whole life —
five players at a nine-handed table — and how crowded the felt is depends on how many
people are sitting down far more than on anything a viewport can report. A full table
was never once looked at.
632 tests, 0 failures. 190 audits across thirteen viewports at two table shapes, with a
mouse and with a thumb. Every table from two seats to twelve now keeps its own text on
every viewport in the sweep. One configuration still reports: a dozen players on a
640x360 phone, where the felt is 87px tall and two rows of 55px pucks cannot both fit
and stay on the rail — the ring grows 8px past the edge to keep the seats off each
other, which is the trade the old code named and got right. Left reported rather than
tuned away.
Names on a crowded phone table truncate harder than they did, because the puck is now
sized to fit instead of allowed to overlap: "Christopher" is "Christ…" at nine seats on
an iPhone 14. The full name is in the DOM, in the seat's form and in the sidebar. The
next lever on that is the seat layer's inset, which is still a fixed 44px of "half a
puck" for a puck that is no longer 96px wide — worth 54px to 60px where it was measured.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
Opened the pot wizard — the one dialog in this tool that divides money — and its
Confirm button was below a scroll. Not on a tall table, not on one screen: at
every size I measured, with six players and a side pot.
iPhone 14 390x844 244px of the dialog below the fold
Galaxy S8 360x640 417px
844x390 landscape 583px
desktop 1280x900 172px
Nothing on screen said there was anything under the list of names, so a dealer
with the table waiting had to scroll a list of players to find the only control
that finishes the hand.
Two goes at this, and the first was wrong. A sticky footer is the obvious
answer and it cannot work here: the actions were the last child of the
scrolling content, so their containing block ended exactly where they did and
`bottom` had no room to lift them. Sticky on the *header* did work, for the
mirror-image reason — its container extends below it — which is what made the
failure confusing to read. Capping the player list wasn't enough either; the
pot breakdown and the log still overflowed on a short screen.
So the dialog is a proper column now: a title that doesn't move, a body that
scrolls, and a footer outside the scroll. potWizard.js appends its buttons to
the footer instead of into the body, and clears both together so a step can't
inherit the previous step's buttons. Measured after: Confirm on screen and
hit-testable at all six configurations, twelve-handed on a Galaxy S8 included,
and the dialog itself no longer scrolls at all.
The checkboxes are painted now, too. `accent-color` only ever coloured the tick
— the box around it stayed the platform's own bright white square, which on a
near-black dialog was the loudest thing on screen and the last raw control in a
tool where everything else had been drawn. Same surface ramp, same corner
radius, and a tick drawn with clip-path so it can't pick up a substituted font.
uiAudit gains "an open dialog's buttons are on screen", which hit-tests rather
than trusting geometry — a button can be inside the viewport and still be under
something. Fault-injected: putting the footer back inside the scrolling body
reports Confirm 105px below the fold.
Two things checked and left alone, because they turned out to be right: the
standings table hides three columns on narrow screens through a container query
keyed to the sidebar rather than the viewport, which is why a desktop shows
fewer columns than an iPad — the sidebar is the narrower box there. And In/Won/
Net read $0 for hands settled outside the wizard, which the comment above them
already explains.
Caught by my own guard on the way past: the token test read `scroll-padding` as
`padding` and failed the spacing scale. Correct by its rules, wrong in substance
— those values are a sticky bar's measured height, not a step in the rhythm — so
the lookbehind now rejects a hyphen as well as a word character. Re-fault-
injected to confirm it still catches a real ad-hoc gap.
619 tests, 0 failures. 65 layout audits across 13 viewports x 5 table sizes and
35 across the five overlays. The whole wizard walked end to end: side pot
resolved, hand logged, $1,560 on the table before and after.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
Looked at the felt at three times scale and the seats were falling off it. Not
sitting at a rail — hanging past the outer edge of the table into the page,
half a puck on green and half on nothing. Measured:
1280x900, nine-handed five of nine seats past the rail, worst 49px
844x390, six-handed all six, worst 20px
430x932, six-handed two, worst 12px
360x640, nine-handed none
The ring sat at a flat 44% by 42% of the seat layer at every size, two numbers
picked against a phone. A fixed pair can only be right for one ratio of puck to
felt, and the felt's ratio changes with every viewport — so it was a different
amount of wrong on each one, which is exactly what makes something look like a
mistake rather than a choice.
The radius is derived now: the widest ring whose outermost seat still finishes
inside the table, worked out from the measured table, seat layer and puck. It
can only ever pull seats in — the old pair is kept as the ceiling — and it falls
back to that pair whenever there is nothing to measure, which is every call from
the unit tests and the first render before the felt is laid out.
Three things this turned up on the way, each caught by the sweep:
* The seat layer is not centred on the table. Its insets differ top from
bottom, so the first version — which assumed symmetry and worked from half
widths — left a phone with 12px still over the rail. It works in absolute
pixels against the nearer edge now.
* The radius and the puck size depend on each other: the ring is limited by
how big a puck is, the puck is sized by the `seatsTight` class, and that
class was decided from the gaps between seats, measured at a radius that
hadn't been worked out yet. Fixing the gap measurement then made a
nine-handed phone overhang that had been fine, because the two were chasing
each other. It is settled by trying the candidates in preference order
instead: the roomy layout if everything fits, the tight one if not, and if
neither fits, tight pucks on a ring widened only as far as it takes to stop
seats colliding — because two seats fighting over a tap is a worse failure
than a puck over the rail. A dealer can see an overhang; they cannot see
that the seat they pressed wasn't the one that answered.
* The spacing threshold was a flat 84px, the figure a 96px puck was judged
against. Asking a 68px tight puck to keep a 96px puck's distance is a test
it can never pass, so every landscape table fell through to the fallback and
took an overhang it didn't need. It is a ratio now, not a number.
Result: worst overhang anywhere goes 49px to 9px, and 59 of 65 configurations
are completely clean. Desktop, tablet and every phone held upright: nothing over
the rail at all.
Residual, reported rather than tuned away: six configurations still overhang by
1-9px, all of them landscape at 640x360 and 667x375, where the felt is about
110-125px tall and no ring puts readable pucks inside that. The audit check is
left strict so they stay visible.
uiAudit gains "seats and chips stay on the table", which runs at every viewport
in the sweep. Fault-injected: putting the flat 44/42 back reproduces the original
failure exactly, including seat 3 at 49px.
Faster, too, because the measuring got hoisted out of the per-seat loops while I
was in there:
renderSeats 7.3ms -> 1.8ms
renderAll 12.4ms -> 6.8ms
619 tests, 0 failures. 65 layout audits across 13 viewports x 5 table sizes.
Offline re-verified: a whole hand with the network cut, $800 conserved.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
The first screen a dealer ever sees was the one nobody had looked at. Opened
fresh, with nothing saved, the tool showed:
* the word EMPTY, in capitals, ten times across the felt
* Resolve Pot as the accent-green primary — on a table with no players and no
pot, so the loudest control on screen was the one that could do least
* Next Hand, with no button to move
* no indication anywhere of what to do first, except a grey caption under the
felt saying "tap a seat to edit"
An empty state is the one screen a new user is guaranteed to meet, and this one
described itself ten times without once saying what to do.
**The felt stops repeating itself.** An open chair is already dashed instead of
solid, faded, and has no stack under it — that reads at a glance, which is how a
dealer looks at a felt, and it reads the same on the one open chair at a full
table, which is the case the label was actually for. The word is gone from all
ten, and the seat numbers underneath it are now the thing you see. What the word
carried for anyone reading by ear moves to an aria-label on the seat, which now
says more than the sighted version ever did: "Seat 3, empty", "Seat 1, Ann".
**The bar says what the table is.** It used to hide itself when nobody was
seated — silent at exactly the moment a dealer has never used this before. It
now says "Nobody's at the table yet." with one full-width button, Seat a player,
which opens the seat editor on the first free chair. That is the same panel
tapping a seat gives them, so the button teaches the gesture rather than
replacing it.
**Two dead controls stop pretending.** Resolve Pot and Next Hand are disabled
until somebody is at the table, which takes the accent green off the button that
cannot work and leaves the only green on screen on the one that can. Disabled
rather than hidden: hiding them would move the header the moment the first
player is named, and a control that vanishes teaches nothing about what it was
for. Next Hand's own explanatory title is kept and handed back when it
re-enables.
One bug of my own, found by walking the flow rather than by reading it: after
typing a name both header buttons stayed greyed out, still captioned "nobody is
at the table yet". commitSeatChange refreshes narrowly on purpose — a full
render on every keystroke takes the name field out from under the dealer
mid-word — and the comment directly above it already warns about this exact trap
for the action bar. The header needed the same line.
609 tests, 0 failures. 65 layout audits across 13 viewports x 5 table sizes,
0 failures — the empty table is now one of those sizes, and had never been
audited before. Re-verified with the network cut: the first-run screen, the CTA
and seating a player all work offline.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
Three findings this round, all measured on the running page.
**1. The tool needed the internet to exist.**
The night already survived anything — it lives in localStorage — but the page
did not. Reload in somebody's back room with one bar of signal and a dealer
holding a phone in one hand and a deck in the other got a browser error page
with nine players waiting. Verified: with no worker, an offline reload is
net::ERR_INTERNET_DISCONNECTED.
There is now a service worker (wwwroot/casino/sw.js) and a manifest, so the
tool installs to a home screen and runs with the network switched off. Measured
end to end: 19 assets cached from one ordinary visit, then with the context
offline the page loads, draws six seats, applies its CSS, and plays a whole
hand — blinds posted, four calls, $800 still on the table.
Three decisions worth writing down:
* The precache list is read off the document rather than kept in the worker.
Every stylesheet and script is already in the DOM; a hand-written file list
stays right until somebody adds a script tag, and then the tool is quietly
offline-broken until a dealer finds out at a table.
* No skipWaiting. A new worker taking over a live page can hand it assets
from a different release than the ones it started with, and this page loads
seventeen scripts that expect to agree with each other.
* Stale-while-revalidate for assets, not cache-first — see finding 3.
Scope is /casino/ rather than /casino/texasholdem/, because a scope has to be a
prefix of the page it controls and "/casino/texasholdem" does not start with
"/casino/texasholdem/". Anything in scope the worker wasn't told about goes
straight to the network.
Program.cs needed .webmanifest added to the MIME map: an unknown extension is
not served with the wrong type, it is not served at all, and a 404 manifest is
exactly the silent failure that stops a browser ever offering "Add to Home
Screen".
**2. A screen reader read out the whole action bar after every tap.**
The bar carried aria-live on the element whose entire contents render() rebuilds
each action. Measured, after one tap:
"PRE-FLOP Pot $15 9 players Bob · $250 behind 0:00 ↶ dealing the hand U
Fold Call $10 Raise"
Nothing in it is wrong; it is unusable at the speed a hand is dealt, and it is
the documented anti-pattern — a live region has to be a stable element whose
text changes, not a container that gets replaced. There is now a small
role="status" element that exists from page load and says one sentence:
"Dee to act, $10 to call. Pot $15." Guarded on the sentence rather than on the
state, because render() runs for many reasons that don't change whose turn it
is, and a live region handed the same words twice reads them twice.
**3. Every tap rebuilt ninety DOM nodes.**
renderSeats did `innerHTML = ""` and then cloned a template, attached three sets
of listeners and set a dozen attributes per chair — on every render, and
renderAll runs after every action. Profiled on a nine-handed table:
renderSeats 7.3ms -> 4.0ms
renderAll 12.4ms -> 9.5ms
The pucks are built once and updated in place, which is what the rest of that
function was already doing to their contents. A seat's node belongs to its index
for as long as the table is that size, so the listeners still close over the
index; only a change in the number of chairs rebuilds the row. Verified that
listeners don't stack: one click after six renders fires the hook exactly once,
and resizing 9 -> 4 -> 10 renumbers correctly.
Found along the way and deliberately left alone: ViewData["cssVersion"] renders
empty on every page of the site, so the ?v= cache-busting has never worked. It
is inert — Program.cs sends Cache-Control: no-store on all static files — and
the service worker was the only reason to care, which is why it uses
stale-while-revalidate instead. Noted in TexasHoldem.cshtml.
608 tests, 0 failures, including a new announcing.test.js. 52 layout audits
across 13 viewports x 4 table sizes: 0 failures, and the action bar still
doesn't change height mid-hand. Gestures re-verified after the seat refactor:
double-tap fold, long-press menu, give-the-button, and the active dealer chip.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015bnp7xKUPFFSpwkYNFaPgb
jheapsandClaude Opus 5.5