The record
Every play StatIQ logged at a price before the game started, graded in the open afterwards. Wins and losses, never deleted — loading the ledger…
Every play StatIQ logged at a price before the game started, graded in the open afterwards. Wins and losses, never deleted — loading the ledger…
Every play StatIQ logged at a price before the game started, graded in the open afterwards. Wins and losses, never deleted. This page exists to be checked, not believed.
Everything below counts WNBA only, previous model (rebuilt after the Sep 2026 data loss — the earlier record is gone), in dollars at 1 unit = $10. Change any of the three and every number on the page changes with it.
Heads up: WNBA’s earlier graded record was lost in a September 2026 data-loss incident and can no longer be audited. Everything here is the record rebuilt after that — not WNBA’s complete history, and never added to the lost one. See what was lost ↓
Before a September 2026 data-loss incident, the tracked WNBA record stood at 367–305–8 (+36.58u), as last recorded Sep 13, 2026. A publishing bug destroyed the underlying picks, so unlike the rest of this page we can no longer audit those figures play by play — and they are gone from the live board. Everything shown on this page is the WNBA record rebuilt after that loss: a fresh, smaller window — not that earlier history, and the two are never added together. This is the same restart the WNBA Today page explains.
Those last-known figures carry no ROI claim: WNBA was a labeled forward test, never a proven edge, and a lost, un-auditable record cannot become one. New guards now reject a publish that would shrink the graded record and stop a failed board read from overwriting what is stored. The record below this note is live and checkable play by play, the way the rest of the site is.
Two windows, always both on screen. Once WNBA has more than one season logged, the prior ones live in the all-time row below, not in a reset scoreboard — this is a calendar year, not a model era (that split is further down, and the two are different axes over the same rows).
| Window | W−L | Graded | Hit rate | Return on stake | Net P/L |
|---|---|---|---|---|---|
| 2026 season37 rows logged this year | 12−12 | 24 | 50.0% | −6.92% | −$16.61 |
| Record since the restartthe rebuilt window — earlier WNBA history was lost | 12−12 | 24 | 50.0% | −6.92% | −$16.61 |
2026-07-08; the current model era starts 2026-07-28. Anything before the first date is kept and shown but not counted — it was graded under rules we have since disowned. The excluded stretch is itemised at the foot of the era table, including when it helped us.A model cutover starts a new era. The previous era stays published and unedited — a scoreboard you are allowed to reset is not a scoreboard. So rather than hide two of the three numbers behind the one that flatters us, all three sit in one table, and the top two add up to the bottom one. Pick a row to make it the headline.
| Window | W−L | Graded | Hit rate | Return on stake | Net P/L |
|---|---|---|---|---|---|
| Record since the restartrebuilt after the Sep 2026 data loss · the model selecting plays today | 12−12 | 24 | 50.0% | −6.92% | −$16.61 |
| Previous modelrebuilt after the Sep 2026 data loss · kept, unedited | 0−0 | 0 | — | — | $0.00 |
| Record since the restartrebuilt after the Sep 2026 data loss · both eras together | 12−12 | 24 | 50.0% | −6.92% | −$16.61 |
The whole official record on one line, with the model cutover drawn as a rule rather than as a fresh start. The shaded span is the era the headline above counts. The best and worst points are both labelled: a chart that only marks its peak is an advertisement.
Cumulative units across 2 graded days, Sep 29 to Sep 30, ending at −1.66u. One point per game day, evenly spaced — days with no games are not stretched out, so the horizontal axis is order, not calendar time. The line is the whole official record, both eras: the era the headline counts is shaded, and the cutover is drawn as a rule rather than as a restart, because restarting the chart at every recalibration is how a losing record disappears one cutover at a time. Every point here is also a row in the table below.
No WNBA plays in the previous model window. Nothing has been removed — switch the era above to see the window that does have them.
Each split is an exhaustive partition of the plays in the current scope: the rows add to the headline’s W–L and to its units, and the table prints the total so you can see that they do. Rates are withheld below n=20 — and on a small sample almost every cell is withheld, which is the honest answer rather than a table of noise.
| Market | Plays | W−L | Hit rate | Return on stake | Net P/L | Contribution |
|---|---|---|---|---|---|---|
| All 0 groups | 0 | 0−0 | — | — | $0.00 |
Both of these count a different set of rows from the headline, and that is exactly how a number stops reconciling and starts looking like a contradiction. So each one states its own n, how that n relates to the headline’s n, and what happened to the rows it leaves out.
No play in this scope had a fillable Kalshi price, so there is nothing to compare and we are not showing a comparison. All 0 plays are in the headline.
Never read closing line value as a return. Beating the close by less than the margin we paid to get there still loses money: break-even is roughly half the hold, about +2.25 points. We are not quoting a mean here at all.
This is the CLV of the plays in this record, and it is not the only CLV number we publish: the wider pool of every price we ever locked — including model-qualified picks we price purely to measure ourselves and never bet — is itemised on the CLV audit. Both are real and they answer different questions, which is why they are never averaged into one.
Rows that could not win or lose money. They are not in the W–L and not in the return, because no stake was ever at risk on them — and they are not deleted, because “we took this line” is the claim, and the claim stands whether or not the player played. Every count here is scoped to the era selected above, the same as the headline.
No voids in this scope.
Where we cannot name the reason precisely we say “no action” rather than guess “did not play”. A postponed game is not a player sitting out, and a label that is only probably true has no business on a record page.
There are no pushes in this scope — every line we took was a half-number, so nothing could land exactly on it. The row is printed anyway: a zero you can see is worth more than a category that quietly does not exist.
The 205 measurement-only picks are model-qualified numbers we take a price on and grade purely to measure ourselves faster. They have their own wins and losses and none of them count here, good or bad. They are the reason the closing-line-value sample on the audit page is bigger than this record.
“Append-only” is easy to write and hard to prove. The proof is that the worst things that have happened to us are still on this page, at full size, in the same type as the good ones.
Both directions, same type size, same panel. A record page that can tell you its best day and not its worst one is an advertisement with a table in it. Every one of these days is expandable in the ledger above — the bad ones are not archived any deeper than the good ones.
What would prove this page wrong. A play that appears here today and is gone tomorrow; a loss that changes to a void; a start date that moves after a bad week; a subset whose n cannot be reconciled with the headline’s n. The CSV below is downloadable and diffable precisely so those are checkable rather than promised.
Every identity this page relies on, re-derived from the raw rows. If a subset near a headline cannot be reconciled with the headline, that is the worst bug this product can ship — so these are printed rather than promised, and a failing one renders as a failure on the page instead of quietly disappearing.
Past results are a record of what happened, not a forecast. No ROI claim is made anywhere on this page: a return figure here describes plays already graded, and the sample is far short of the several thousand it would take to separate skill from variance. See every play in this record, day by day →
+ WNBA is ahead
Over 0 graded WNBA plays in this era we are up $0.00. That is a result, not a claim — a return figure needs something like 4,000 plays before it separates skill from variance, so we publish it exactly the way we publish the one that is behind. The other two windows are not hidden: the record since the restart is 12−12 at −$16.61, and the record since the restart is 12−12 at −$16.61. Records are never blended across sports — different models, different markets, different start dates; one combined figure would be a number we invented.
What this page deliberately does not do. It does not score the probability we published against what happened, and it does not repeat the backtests — that is the model report card, and printing the same calibration table in two places is how two surfaces start to disagree. Likewise the itemised list of every price we ever locked, including the measurement-only ones that are not in this record, is the CLV audit. This page answers exactly one question: what happened to the plays we published?