PFP Lab Notes #5: three rules over +10 units. Not proof yet.
Three lab rules are over +10 units. Units swing a lot, so we still don't call it proof.
Lab notes
What we tested, what we found, and what failed. Newest first.
CLV below means closing line value: how much better a price was than the closing price, in points.
The plain version, with a worked example, is what closing line value is.
Get these by email: Lab notes.
Lab Notes #4: a grader mistake, a new page, and merch
Our grader couldn't read a builder's picks, written in a format we had suggested. He got an alert in minutes, and we fixed it before any game started.
New house rule 21: if the same pick is posted twice, only the first post counts.
Also new: a page with every pick our rules took tonight, and a small lab merch shop.
Lab Notes #3: an NBA rule is up 24%. We think it's mostly luck.
B4 bets the under on early-season NBA totals. It is 22-12, up about 24% on its bets.
Its prices beat the closing line, but by far less than the book's cut. Most of that 24% is luck.
The real test is whether its edge shrinks as books adjust through the season.
Lab Notes #2: our rules read the line. The book still wins.
Every pick now shows two numbers: skill (did the line move your way) and cost (the book's cut).
Our best rules do read the line, but the skill covers only about a third of the cut.
Nobody passes the Tail Test yet. Not our rules, not the builders.
Lab Notes #1: one rule declared, one idea dead.
N2, our NHL totals rule, entered the race for the Standard. It starts at zero.
One idea died: empty-net goals are not rising in a way the totals market misses.
We started recording UFC closing lines, so any future rule has honest data to be graded on.
An average can hide a bad week
A bettor can have positive average CLV and still lose to the close on most bets.
It happens when a few bets beat the close by a lot, and most lose by a little. The average looks fine. The typical bet does not.
That is why our bar asks for both: a positive average, and more than half of moved prices beating the close.
One number alone can fool you. Two are harder to fool.
What "checkable" should mean
Lots of records say they can be checked. Here is what makes that true.
Every pick is posted before the game, with a time.
Every pick shows the price it was taken at.
Every pick is graded against the closing price, not just the win or loss.
Losses stay up. Nothing is deleted.
If a record is missing any of these, you are checking a story, not a record.
Tired NHL teams are not a betting edge
The belief: teams on the second night of a back-to-back are mispriced.
The test: every back-to-back from the 2025-26 season, 429 games, graded against the DraftKings closing line.
The result: betting against the tired team moved slightly our way, about a third of a point. Our bar is 1.5. Betting on the tired team lost to the close.
What we did: nothing. The rule was never frozen and never raced. The market already prices fatigue.
I said most of the line move happens late. I was wrong.
I said on X that most of a line's move happens in the last few hours.
We checked 82 games in our own price tapes. The last 3 hours carried about a quarter of the move, not most of it.
I corrected it publicly. Being checkable means checking yourself too.