StatIQPlayer projections, with the receipts

How we measure: every claim we make, the query behind it, and the result that would prove it wrong.

How we measure

Every claim we make, the query behind it, and the result that would prove it wrong, including the mistakes we have made and corrected.

In plain language

  • Every public claim we make is listed below with its scope and date, what we found, and the result that would prove it wrong.
  • Picks are logged before the game and graded in the open, and a graded pick stays on the record. Through Nov 12, 2026, we withdrew at least 87 college football picks before kickoff, so they never counted in the record. See the corrections.
  • A graded record is history, not a forecast. A forward claim waits for a bar we wrote down in advance.
  • Where we measured against ourselves and came out behind, the entry stays; our corrections are dated below.
  • Check the rows yourself: the record, the CLV audit (every locked price and its close, exportable for members), and WNBA’s pre-restart season, recovered pick by pick from our daily-board archive.

The record today

MLB · all-time logged picks2,210–1,709 · 56.4%3,919 graded picks, Jul 12, 2026 to Oct 28, 2026 · flat 1 unit a pick · return on stake +8.18% · 3,550 of 3,919 at licensed books · as logged +8.20% · 3,550 graded picks at licensed books, Jul 17, 2026 to Oct 28, 2026includes corrections through Nov 16Every MLB pick and loss
WNBA · recovered pre-restart window382–314–8 · 54.9%704 graded picks, Jul 8, 2026 − Sep 27, 2026 · flat 1 unit a pick · recovered from our daily-board archive, never added to the live recordThe recovered WNBA record
Past results, not a forecast. Record read Nov 18, 2:26 PM ET.

NFL forward test: mixed so far. Focused 202–169 (+0.6% measured return); Broad 246–211 (−0.9%).

NFL 2026 · graded weeks 5–9 · board as of Nov 18, 2026 · read Nov 18, 2026, 2:28 PM ET. Early, overlapping samples; a lasting edge is not established. The Focused rows are also in Broad, so the two are never added together.

Since Oct 24, 2026, StatIQ presents the focused test as its NFL selections; the broad test keeps running beside it as the control, logged and graded the same way. Nothing was regraded or removed. The choice was made on the record up to that day, so the games that kick off after it are the cleaner read.

Measured at closing prices on a flat one-unit stake. No bets were placed: the return is a measurement of the rule, not a realized profit and not a forecast.

Every NFL forward-test result, graded and pending

Claim 5 is an early finding that is not proven yet, and its own rule is that we say nothing public about it until it is; its number stays in the list so the gap is visible.

The claims

The rules this document follows

Every public claim we make, the query that backs it, and the result that would prove it wrong.

This exists so that when someone asks "why do you say that," the answer is a number and a method, not an opinion. Others are free to disagree, but they have to disagree with the data, and they can check it themselves.

Three rules for this document:

  1. The falsifier is written before the numbers come in, not after. "That metric doesn't apply to us" is a legitimate methodological position and also the exact sentence a losing bettor reaches for. The only thing that separates them is whether the test was specified in advance. If we ever want to drop a claim, we change it here first, in the open.
  2. The unflattering entries stay. A ledger you would not hand to a skeptic is not a ledger. The entries below where we measured against ourselves are what make the rest credible.
  3. Numbers are generated, never typed. Every figure in the tables below is read straight off the live boards by a script and written into this document mechanically. Nobody types a number into it, so it cannot quietly fall out of date and it cannot be nudged. One window is not live: WNBA's pre-restart 2026 season was wiped from the live board by a publishing bug, so it is read from our dated daily-board archive, which preserved it: recovered pick by pick and auditable, not a figure anyone re-keyed by hand.

Claim 1 · Our all-time MLB record has been profitable. We report it as history, not a forecast.

Result
the all-time official MLB tracked record (every pick logged before the game and graded at the price we locked, both model eras included) has been profitable: +320.26u over 3,851 graded plays, +8.32%, since 2026-07-12 (graded through 2026-10-12).
Scope
MLB, all-time, both model eras (the current model since 2026-07-28 and the previous model before it), never blended with another sport.
Sample
3,851 graded plays
Date
graded through 2026-10-12. A snapshot figure from the claims document, written at 3,851 graded MLB picks (the live record has 3,919); the live record is under “The record today” above.
Would be proved wrong by
We committed, in writing before the numbers matured, to make a forward profitability claim only once ROI clears 2σ from zero on n ≥ 1,500 tracked plays.
Full argument

What we say: the all-time official MLB tracked record (every pick logged before the game and graded at the price we locked, both model eras included) has been profitable: +320.26u over 3,851 graded plays, +8.32%, since 2026-07-12 (graded through 2026-10-12). It is a measured past result at a flat 1 unit per play, before any bonus, boost or fee, with pushes counted as no gain and voids excluded: a tracked return at recorded prices, not executed wagers, so not proof that any given bettor could have taken those prices. It reflects the whole selection, price-shopping and locking process, not model skill alone. It is not a per-bet +EV claim and not a promise about future picks.

Scope, stated plainly. MLB, all-time, both model eras (the current model since 2026-07-28 and the previous model before it), never blended with another sport. WNBA's tracked record is +5.62% over 696 graded plays: the all-time figure, both model eras, counted from its first graded pick on 2026-07-08. It is positive but a smaller sample, reported the same way. That earlier window was wiped from the live board by a 2026 publishing bug and recovered pick by pick from our dated daily-board archive, so it is auditable the same way the rest of the record is, not a live-regenerated number, and not a fossil we can no longer check. The live Record and Report Card pages now show only the window since the restart (a much smaller, more recent slice), so their win–loss totals are far lower. Read the two as different windows of the same record, each labelled with the window it covers, not as competing claims. NFL 2026 is a measurement-only forward test at −7.7%: a loss, published at equal weight, on which no money was ever wagered.

Why a big sample matters, and how we test significance. A single play's return has a standard deviation near one unit, so a few hundred plays cannot separate a real result from variance. The significance column below reports each sport's distance from zero using a standard error clustered by game-day. Picks on the same slate share exposure (one game runs hot or cold; a total moves a whole slate the same way), and counting them as independent overstates significance. Clustered, MLB is +4.2σ (a naive independent-play calculation would read +5.1σ); WNBA is +1.4σ.

The bar we pre-registered (2026-07-20). We committed, in writing before the numbers matured, to make a forward profitability claim only once ROI clears 2σ from zero on n ≥ 1,500 tracked plays. MLB's clustered significance is past that bar; WNBA's is not. We publish the MLB history above either way (a graded record is a fact) and hold any forward-looking claim to that pre-registered standard.

Status: ACTIVE.

Claim 2 · CLV is a diagnostic in WNBA/MLB props, not a pass/fail gate

Result
we publish our closing-line value, and we do not treat negative CLV in these markets as evidence we are wrong.
Scope
WNBA and MLB player props on our own board
Sample
12,790 graded picks, 2,194 WNBA and 10,596 MLB
Date
established 2026-07-20
Would be proved wrong by
if the closing price's information advantage over the open in these markets reaches +0.03 AUC or more at n > 5,000, the close is carrying information, and CLV returns to being a gate.
Full argument

What we say: we publish our closing-line value, and we do not treat negative CLV in these markets as evidence we are wrong.

What we found (first measured 2026-07-20 and re-run since; scope: WNBA and MLB player props on our own board; latest sample: 12,790 graded picks, 2,194 WNBA and 10,596 MLB): the closing price predicted outcomes no better than the opening price. CLV is a proxy that rests on the close being the better price, and in these markets that premise failed. ROI is the primary measure. CLV is published as a diagnostic with this caveat attached.

Scope limit: this does not travel. It applies to thin prop markets we have measured. For NBA and NFL sides the close is very likely informative, and CLV remains a gate there.

Falsifier: if the closing price's information advantage over the open in these markets reaches +0.03 AUC or more at n > 5,000, the close is carrying information, and CLV returns to being a gate.

A correction we made to this claim (2026-07-20): we had cited Pinnacle's prop margin as evidence that no sharp reference exists. We measured it: Pinnacle prices these props at about 7%, mid-field among US books. The number was right and the inference was refuted, so it is not cited as support. It is in the errata below.

Status: ACTIVE (established 2026-07-20).

Claim 3 · Line-shopping and lock discipline cut the hold we pay by ~64% (a relative reduction, MLB, measured 2026-07-20)

Result
taking the best available price across books materially lowers the edge required to break even, and this is true whether or not our projections are any good.
Scope
MLB only, per pick from the per-book prices at a single instant
Sample
1,011 picks
Date
measured 2026-07-20
Would be proved wrong by
if line-shopped hold stops beating the BEST SINGLE BOOK by at least 1 point (because we lose book access, or because the books we can reach converge), this claim dies with it.
Full argument

What we say: taking the best available price across books materially lowers the edge required to break even, and this is true whether or not our projections are any good. It is mechanical, not predictive, and it is verifiable on any single pick.

The measurement behind the headline (measured 2026-07-20; scope: MLB only, per pick from the per-book prices at a single instant; sample: 1,011 picks):

Hold paid
Median single book (the headline's baseline)7.29%
Best single book5.95%
Line-shopped across books4.29%
On the picks we actually locked2.65%

How to read "~64%". 7.29% → 2.65% is a relative reduction in the measured hold: 1 − 2.65 / 7.29 ≈ 64%. In absolute terms it is 4.64 percentage points of hold, not 64 points, and break-even edge falls from about 3.6 points to about 1.3. It is a cost we paid less of, not a return: it says nothing about whether the picks win. Against the best single book (5.95%) instead of the median one, the same comparison is about a 55% relative reduction, and line-shopping alone buys 1.67 points: that is the honest baseline, because a user who simply bets at the sharpest book they already hold captures most of the benefit with no tool at all. The 2.65% is measured on the picks we locked, so it is partly selected by the lock rule itself. We have no WNBA figure: WNBA stores per-book prices for our side only.

Verifiability, stated plainly. The arithmetic above can be checked from the four figures in this table. The figures themselves are a single dated measurement: our automatically regenerated tables do not currently carry per-book prices, so they cannot be re-derived from the regenerated block today. Until they can, read the headline as one measurement on one date, not a standing rate.

Correction on record (2026-07-20): the first version of this claim reported "3.56% → 1.75%" and was wrong twice, both times in our favour. It used our stored closing prices as "a single book's own hold" when those are already the best across several books, and it measured the shopped figure on locked picks only, selecting on the very threshold that defines locking. Real single-book holds are 7–9%, not 3.5%.

Falsifier: if line-shopped hold stops beating the BEST SINGLE BOOK by at least 1 point (because we lose book access, or because the books we can reach converge), this claim dies with it.

Status: ACTIVE.

Claim 4 · No edge we found in the prop markets we measured has persisted, and we say so

Result
in the WNBA and MLB player-prop segments we tested, no edge persisted.
Scope
WNBA and MLB player props, 2026
Sample
each apparent edge was first seen on fewer than 150 plays, and the largest re-test covered a full season of 302 games
Date
recorded 2026-07-20
Would be proved wrong by
an edge is real when it clears 2σ on a pre-registered, forward-tested sample, never on the sample that suggested it.
Full argument

What we say (a price/edge result, not an accuracy result): in the WNBA and MLB player-prop segments we tested, no edge persisted. We do not claim to have found a persistent edge in them. This is not a statement about how accurate our probabilities are against the closing price; that accuracy comparison has not been run for WNBA or MLB (the Model Report Card says so on each sport's card).

What we found (recorded 2026-07-20; scope: WNBA and MLB player props, 2026; sample: each apparent edge was first seen on fewer than 150 plays, and the largest re-test covered a full season of 302 games): every one of them dissolved when the sample grew. This is the entry we would most like to delete, which is exactly why it stays.

Falsifier / graduation: an edge is real when it clears 2σ on a pre-registered, forward-tested sample, never on the sample that suggested it.

Earlier wording (2026-07-20 to 2026-10-14): this claim was headed "The prop markets we have measured are efficient" and said the props "measured efficient against our model", while the Model Report Card said "no efficiency verdict published" about the separate accuracy comparison: one word for two different results. Renamed on 2026-10-14; neither finding changed.

Status: ACTIVE.

Claim 5 is not published; see the note.

Claim 6 · We publish what proves us, and the record stays open

Result
our record, our grading, our method and our mistakes are open and checkable, and they stay that way.
Scope
every sport's official record on this site, its grading, this ledger and its errata.
Sample
Not a sampled measurement.
Date
established 2026-07-20
Would be proved wrong by
any graded pick, loss, void or erratum that disappears from the public record or from this ledger.
Full argument

What we say: our record, our grading, our method and our mistakes are open and checkable, and they stay that way.

Scope: every sport's official record on this site, its grading, this ledger and its errata. A standing commitment, not a sampled measurement.

Why it matters. Everything we publish to prove ourselves is public: every pick logged before tip, the price we took, the closing price, CLV, ROI with error bars, the void and DNP reasons, this ledger and its errata. None of it is ever deleted, hidden or rewritten: not a loss, not a bad month, not a mistake we later caught. Hiding any of it would forfeit the one position nobody else in this category occupies: a record a skeptic can check for themselves.

Projections and the record stay free. The projections, the injury information and the full historical record are free and auditable, and they stay that way. The record is a permanent public commitment, never gated and never pruned. That is the whole point of this page: it is the one artifact a competitor cannot copy without also having done the work.

Falsifier: any graded pick, loss, void or erratum that disappears from the public record or from this ledger.

Status: ACTIVE (established 2026-07-20).

Exchange depth we measured

What we measured: how much money rests at a price level on player-prop markets, two ways: a systematic read of one exchange's market-data feed, and hand-collected observations from the live venue apps. Neither placed an order.

The systematic read: ProphetX test environment, 2026-08-14. 4,850 resting price levels across MLB player props on 15 events (hits, total bases, RBIs, home runs, pitcher strikeouts and outs recorded among them). Per price level: 25th percentile $6.58, median $15.82, 75th percentile $28.50, 90th percentile $45, 99th percentile $137.50, maximum $237.50. The venue seeds that environment's resting orders itself, so this is the shape of a book, not live liquidity. Nothing had ever traded there, so it says nothing about how often a resting order fills. A separate read of the same environment, at the top price level only, put the median at $12.89.

The live observations: ProphetX and Novig, 2026-08-17 to 2026-08-19. Twelve rows of MLB player props, read by hand from the live apps, almost all at the price we would have taken. ProphetX showed from nothing usable (a price that had already moved) and $5 up to about $216 resting at our price; across eight ProphetX rows one morning the usable depth averaged about $71 a row, roughly five times the test environment's median. Novig was read only overnight on 2026-08-17 and showed about $1 to $4.25 at our price; on the same row at the same moment, ProphetX held about $200. ProphetX's book was fuller in the morning than overnight. Novig was not read in the morning, so we have no morning comparison between the two.

What was not measured. NFL and other sports were in neither sample. Sportsbook stake limits were not measured; they are a different question.

How to read it for the Record's bankroll simulation. A stake larger than what rests at a price moves the price against the bettor. A stake that grows with a bankroll outgrows what rested at the prices the record logged, and the simulation does not model that; the note beside its figures says so.

Falsifier: a later systematic read of the live markets showing materially different depth. The dates stay on the figures either way.

Definitions

Definitions of the terms used on this page
TermWhat it means here
Flat 1uEvery play measured at the same size, one unit, so the record reflects the picks rather than clever sizing. It’s a measurement convention for comparing honestly, not a recommendation for how you should bet.
ROIReturn per unit risked across a set of graded plays: what actually happened, not what we expect to happen next. Small samples swing hard in both directions, so read it with the number of plays sitting beside it.
Hit rateThe share of graded plays that won. It says nothing about price. You can win most of your bets at bad numbers and still be down, which is why we show it next to price measures rather than alone.
PushThe result landed exactly on the line, so the bet ties and your stake comes back. It counts as neither a win nor a loss, and it stays on the record.
Void / DNPThe player never played, or the market was pulled, so there’s nothing to grade. The row stays on the record and settles at zero. A DNP is not a play we dropped.
Append-only recordA graded play stays on the record. Plays settle (win, loss, push or DNP) and stay exactly where they were, losses included. College football picks we withdrew before kickoff never counted, and the record’s corrections list them. A record you can edit is not a record.
Model eraWhen the model that produced these picks started. Picks from an older version are kept unedited but counted separately. Blending them would misrepresent both the old model and the new one.
Sample size (n)How many graded plays a number is built on. Below a minimum we hide the rate and show the raw counts instead, because a hit rate off a handful of plays is noise wearing a number’s clothes, and that cuts against us exactly as often as it flatters us.
95% CIThe range the true number plausibly sits in, given how few plays we have. If the range crosses zero, the honest read is “we don’t know yet”, however good the middle number looks.
Closing line value (CLV)Whether the price moved your way after you bet: a sign you got a good number before the market caught up. It is evidence about the price you took, not proof of profit: you can beat the close by less than the cut you paid and still lose money.
De-vigged CLVCLV measured fair price to fair price, with the book’s cut removed from both ends. It’s a smaller sample than the raw figure but a cleaner comparison. The raw number flatters us, because part of it is just the cut shrinking as a market matures.
Closing lineThe last price before the event starts, once all the money is in. It’s the market’s best-informed number, which is why we measure our own prices against it instead of against our results.
Vig (juice)The sportsbook’s built-in cut, baked into the price on both sides. It’s why a genuine coin flip priced at −110 both ways loses money over time. You have to clear the cut before you’ve done anything.
Pre-registeredThe rule was written down and frozen before the season, then published, so it can’t be quietly adjusted once results come in. It’s the difference between a test and a story told afterwards.
BacktestRunning a rule over past seasons to see what it would have done. Useful for ruling ideas out, and very easy to fool yourself with. Almost anything can be made to look good in hindsight.
Forward testGrading a rule on games that hadn’t been played when the rule was written. A backtest can be tuned until it looks good; a forward test can’t, which is the only reason it’s worth anything.
Measurement onlyWe are measuring something in public, not recommending it. Nothing on a page marked this way is a bet we’re telling you to make. Where a return is shown, it is measured at closing prices on a flat one-unit stake with no bets placed: a measurement of the rule, not a realized profit and not a forecast.

Every other term is in the glossary.

Corrections

Kept because a ledger without its own errata is marketing.

Known measurement failures we corrected, by date
DateWhat was wrongEffect
Jul 12Record graded at a refreshed price while displaying the open priceCredited itself a number it never posted; fixed by watch-then-lock
Jul 12DFS pick'em prices used as both fair value and bet priceBoard read +0.11% EV; restricted to takeable prices it was −4.19%
Jul 20MLB graded unlocked picks at the drifted line82% of drifts favoured us; 30 results wrong, 20 fake wins
Jul 20CLV differenced prices quoted on different linesMLB record CLV read −0.16%; correctly measured it is +1.07%
Jul 20The public share card computed CLV with its own copy of the mathThe most public artifact disagreed with the audit log it invited people to check
Jul 20Claim 3's hold table used already-shopped prices as "a book's own hold", and measured the shopped column on locked picks onlyReported 3.56% → 1.75%; real single-book holds are 7–9%. Both errors flattered us
Jul 20An early draft of our publish-versus-withhold policy relied on a timing rule that would not have done what it claimedCaught and corrected before it went live; the open record was never affected
Jul 20Claim 2 cited Pinnacle's prop vig as evidence that no sharp reference existsMeasured: Pinnacle prices both sports at ~7%, mid-field vs US books (6.4–8.9%). The number was right, the inference was refuted: ~7% is just what props cost
Oct 21The NFL forward test voided a pick whenever a player had no stat line two days after his game, without evidence that he sat out18 of the 40 NFL voids were players who played and recorded nothing. One was a tracked pick: Gunnar Helm receptions under 2.5 on 2026-10-18 (33 snaps, 0 catches), which would have graded a win at −154 in both the Broad and Focused tests. That void stays in the record as written; voids now require evidence that the player did not play
Oct 23The college football record graded each pick on the first stat line our play-by-play source published after the game, before that source had finished revising the game, and that source is not the official box scoreChecked against the official box score, 9 of the 145 graded CFB picks (games 2026-08-29 to 2026-10-17) have a different result: 5 posted wins are losses and 4 posted losses are wins. The record reads 79-66, +4.15%; on the box score it is 78-67, +2.80%. Four of the nine were graded before the source finished revising the game; the other five are plays the source records differently from the box score (a quarterback credited with 1 of his 19 completions, a 5-yard loss on a catch recorded as a gain, sacks and penalty-play runs left out of rushing). The 145 rows stay as posted. A pick now settles only once its game is final and its number has held for two days; the five source differences are not fixed by that

Measured figures (snapshot)

Every number below is read from the live boards and written here mechanically; nobody types them in.

Graded through 2026-10-12 for the MLB record row (3,851 graded, +8.32%). The WNBA row is the recovered window described in Claim 1. The CLV, information and hold tables are from an earlier run and carry no date of their own.

These tables are written by script when the document is regenerated, not on every page load, so their record totals can trail the live record (this snapshot was written at 3,851 graded MLB picks (the live record has 3,919)). Where the two differ, the live record (under “The record today” and on the Record) is the current number.

The record (official graded picks, from WNBA 2026-07-08 / MLB 2026-07-12)

sportW-LnROIsignificanceplays needed for 2σ
WNBA382-314696+5.62%+1.4σ~1,356
MLB2172-16793851+8.32%+4.3σ~821

A result under ~2σ is not distinguishable from zero. Where a sport clears that bar we report its record as measured history, never a per-bet +EV claim or a forecast; where it does not, we make no profitability claim at all. Significance is clustered by game-day (see Claim 1).

CLV (diagnostic, not a gate; see Claim 2)

sportavg CLVbeat closenexcluded: line relisted
WNBA+0.26%1661/30933093891 of 4367 graded
MLB+1.27%3387/49524952211 of 11483 graded

Break-even CLV ≈ fair × hold ≈ +2.25 points. A positive CLV below that is still a losing price. Exclusions are picks where the book relisted the line: unmeasurable, not zero.