Total bets23
Win rate61%
Resolved18
Member since2026
Win rate is % correct on resolved bets (public). Reputation = the track record, not the balance.
Level 3. Four switchers in one file, and every one of them works with no JavaScript at all: a radio, a label and a sibling selector. 36 event tabs, 6 profile tabs, 18 rules tabs and 36 range chips across 9 and 2 screens. Measured 2026-08-08 in a browser by cycling every radio, at 390 and 1280, in both pointer branches.
A tab strip with no panel under it is half a component. 4 elements against the product's 186: the event tab set, its four panels and the content each one holds. The strip is a radio group and the panel is its sibling, so what a tab DOES is only visible with both.
| Holds 320 YES | Side | Amount |
|---|---|---|
| You | YES | $4.94 |
| whale_07 | YES | $471 |
| hedge_hannah | NO | $588 |
| marketmaven | YES | $121 |
| polly_predicts | NO | $87 |
Your bet is highlighted. Bets update as the odds move.
Recent bets, largest first. Filter: over $5.
These always go to the wire. I am holding my position and adding if it dips.
The committee bloc looks firmer this time though.
Volume jumped right after the headline. Watching how it settles.
First bet here. The resolution source being official notices makes this easy to trust.
| Holds 320 YES | Side | Amount |
|---|---|---|
| You | YES | $4.94 |
| whale_07 | YES | $471 |
| hedge_hannah | NO | $588 |
| marketmaven | YES | $121 |
| polly_predicts | NO | $87 |
Your bet is highlighted. Bets update as the odds move.
Recent bets, largest first. Filter: over $5.
The strip above is not the strip. This file draws three separate tab
families and until 2026-08-10 the page stood only the richest: .tabs stands on
9 screens and .ptabs on 2, and neither had ever been drawn in the kit. A
reader comparing a tab against a tab was comparing one of three against nothing.
The plain strip, .tabs, from
active-bets.html: two links, and the state is aria-current="page"
rather than a class. It is the whole of the My Bets switch, Active against History, and it
navigates: each tab is a different URL, so nothing is hidden and shown, the page changes.
IT WORE role="tablist" AND role="tab" UNTIL
2026-08-11 AND OWNED NOTHING. A tablist owns elements with role="tab"; here the
tabs were one level down inside the links, so the relationship was broken on 9 screens per
tree. Two more things were true at once: there is no role="tabpanel" anywhere
in that family, in either tree, so the tabs controlled nothing; and they navigate to
another document, which a tab does not do. Moving the role up onto the anchor would have
fixed the ownership and left a tab with no panel that leaves the page. So the roles came OFF,
the container is a <nav> so its label still names something, and the state is
the idiom this product already writes 1,228 times. Backlog 89.
AND THE OPPOSITE HAPPENED TO A DIFFERENT STRIP ON 2026-08-16, WHICH IS WHY
BOTH SENTENCES BELONG ON ONE PAGE. That pass removed the role from the strip that could not
own one; it did not look at the other two bars in this file, and both were still declaring
role="tablist" over children with 0 role="tab" and 0
aria-selected in either tree. .rules-tabs is the one of the three
that CAN own the role, because it is two real <button>s over two real panels
with a script that swaps them, so it has role="tab", aria-selected and
role="tabpanel" now, kept in step by the same handler that was already toggling the
class. .ed-tabbar and .ptab-bar are <label for> over
hidden radios, and a label is not a tab because a label is not focusable, so the role came
off both and their panels lost role="tabpanel" with it. The third of the three
was missed by the sweep that fixed the first two and was caught by re-running the check,
which is the only way a role can be found: nothing in a stylesheet can see one.
The profile strip, .ptabs, from
my-profile.html: three radios, three labels and three panels, CSS only, exactly the
mechanism the event strip uses and at a different size. Its panels hold different components
again, which is why it is here whole rather than as a bare bar: a record card, a gallery and a
list of resolved positions.
The radio names and ids are suffixed per cell. A radio group is keyed
by name and a label reaches its input by id, so two theme cells sharing
either would be one group across both: pressing a label in daylight would move the dark cell.
That is the kit's id rule arriving through a second door, and it is the reason this specimen is
not a copy of the screen's attribute values.
The level and its reason stay on the organisms page. This page holds the thing a shelf can never show about this component: what is behind each tab, which needed the radios cycled one at a time and every panel measured, and the one place the system's single focus ring could not land.
| tab | what is behind it | panel at 390 | at 1280 |
|---|---|---|---|
| Comments | comments, the thread | 310x498 | 590x420 |
| Biggest bets | bets table, the holders list | 310x186 | 590x186 |
| Bets | the positions table | 310x248 | 590x248 |
| Activity | the activity list | 310x277 | 590x266 |
Measured by setting each radio in turn and reading what stops being
display:none. A static reading of this screen sees one of the four and reports
the other three as 0x0, which is what the first pass of this batch did before the radios were
cycled: the positions table, the holders rows and the activity rows all measured 0x0 at both
widths and none of them is missing.
A zero behind a tab is not a defect, it is a tab. Same sentence as the six skeleton marks in a hidden bet panel, and the same instrument correction: ask whether the host is on the screen.
base.css draws the ring once, on the element that HAS focus,
and the element that has focus here is the RADIO, which this file parks at
left:-9999px. So the indicator was painted off the left edge of the document,
and a person tabbing into the tab strip saw nothing at all.
The radio is still focusable, which is the part that matters: it is offscreen rather than removed, so the keyboard still reaches it. What was missing was only the sign that it had. The ring is drawn on the LABEL now, which is the thing a person can see.
The profile strip needed the same answer with the input hidden a different way. There the radio is a 1px box at zero opacity, so the ring is drawn in the right place and then painted at zero alpha, which is the same nothing arrived at from the other direction. Its bar is a scroller and the pill fills its content box, so the offset turns inward there, exactly as it does in the filter panel.
| strip | worn | mouse | finger |
|---|---|---|---|
.ed-tablabel, the event tabs | 36 | 36 tall | 36 tall, unchanged |
.ptab-lbl, the profile tabs | 6 | 38 tall | 44 |
.rules-tab | 18 | 31 tall | 44 |
.ed-range button | 36 | 36 tall | 44, it is a chip |
Three of the four take the floor and one does not. The one touch floor in
components/base.css names fourteen families, including .tabs button,
.ptab-lbl and .rules-tab, and .ed-tablabel is not one of
them: it is a <label>, not a button, so the tag-based half of the list
misses it. Those 36 labels are the main navigation of the event screen, on a phone, at 36
tall. docs/backlog.md 64.
Measured 2026-08-04: the two agree on padding, radius, size, height, ground and selected ground down to the pixel, and disagreed on the two things left, the face and the weight. One rail was mono at 400 rising to 600 when chosen; the sorter was the body face. Two files, one control, and neither file could take the other without styling another component's classes.
Both are chip's rail face now. What stayed here is the row: the sorter's markup stands in the comments file and its styling has always stood here, so its press stands here too, next to the hover it answers rather than in the other file.
Both selected grounds in this file were a colour function at 16 per cent, which IS a rung and was still off the ladder, because a value spelled as a function is not a token and does not get the theme. On graphite nothing moved; on chalk they now step up like every other brass tint and the label reads 5.72:1 there.
Every tab in this file was missing it. A tab is a quiet control, so it settles onto the one held-down role in the system, and each press mirrors the hover it answers so the two sit at equal specificity and source order hands the pointer to the press.
It is worth most on a phone, where there is no hover at all: until it arrived, a tap on a tab said nothing until the panel underneath had already changed.
One query, at DESK, and it takes the FRAME off the chart's range group and
not a pixel off its buttons. Measured on a coarse pointer at 320, 360, 390, 640 and 1280,
every .ed-range chip is exactly 44 x 44, which is the touch floor base.css
declares once for the whole family. The row was 54 and the ten pixels were never the buttons:
4px of padding and a hairline a side, 44 + 8 + 2, and the group is 198 wide for the
same arithmetic, 4 x 44 + 3 x 4 + 8 + 2. Below the rung the well, the edge and the
corner go and the row is 44, the group 182.
The four chips still read as one control with the frame gone, because
the chosen one carries a brass ground and the other three carry none: a segmented control is told
apart by its selection, and the box around it is what a desk has room for. Anything below 44 is
the floor's decision and not this file's, and a component that shrank under it would be the
.btn-md defect again, the same control at 48 under a mouse and 47 under a finger.
A switcher needs no script. A radio, a label and a sibling selector, and the state lives in the document rather than in a variable.
Hide the input where the ring can still be drawn. Offscreen is focusable and invisible, and it takes the indicator with it, so the indicator moves to the label.
Two strips that measure identically are one control. The face and the weight were the only disagreements, and neither had an argument.
Never park a focusable element offscreen and leave the ring on it. The indicator was painted at -9999px, which is a keyboard user seeing nothing.
Never measure a tabbed screen without cycling the tabs. Three of four panels read 0x0 at both widths and all three are correct.
Never write a floor as a list of tags. Three of these four strips take the touch floor and the fourth is a label rather than a button.
The level and its reason: the organisms page.
The stylesheet: components/tabs.css. The chip both rails wear:
chip. What is behind the four panels:
comments and bets table. The floor and
the family it misses: components/base.css and docs/backlog.md 64.
These always go to the wire. I am holding my position and adding if it dips.
The committee bloc looks firmer this time though.
Volume jumped right after the headline. Watching how it settles.
First bet here. The resolution source being official notices makes this easy to trust.