The Ethos
This is not a page about our values. It is a list of the rules we have bound ourselves to — constraints that have already cost us something, each one linked to the place it did. A rule nobody has ever had to eat is a slogan. The links are the argument.
Baseball’s prediction business runs on a simple trick: remember the hits, forget the misses, and never write the threshold down. Everything below exists to make that trick impossible for us to perform.
Eight rules · every one with a receipt
01
A call is filed with a metric, a threshold, a window, and a timestamp — or it isn't a call.
Anyone can say a player is due. We say what has to happen, by how much, and by when, in a number a computer can check without us in the room. Then the window closes and the engine grades it. Nobody argues, and nobody grades their own homework.
What it costs usEvery open call is a chance to be wrong in public on a date we chose in advance. There are 11 of them on the ledger.
02
Nothing is edited after filing. Ever.
Not the threshold, not the window, not the thesis. If we discover an error mid-window, the correction is appended in public and the original stays exactly as filed. If a player stops playing mid-window, his number freezes and the call resolves against it. No voids, no do-overs, no quiet deletions in the night.
What it costs usWe have already eaten this one. Fever Report No. 004 read our own board backwards — a hitter falling on the fade board means the engine is easing off him, and we wrote it as bad news. The piece was not rewritten. The correction rides above it and the error stands underneath, in our own words, forever.
03
Misses stay public, and the record counts them.
Hits, misses, and open — all three, on every page that mentions the record. A hit rate with the misses removed is marketing, not measurement, and the whole reason to build this thing was that nobody in this hobby keeps score of themselves.
What it costs usThe scoreboard is the front page. It will report our worst season as loudly as our best.
04
The protocol is frozen — in public, in git — before the data that judges it is examined.
The hypothesis, the model, the sample, the success criterion, and what we will publish if it fails: all written down and committed first. Then one run. No re-runs, no threshold shopping, no quietly deciding afterwards which test we meant.
What it costs usFour pre-registered experiments so far. Three came back no — and all four are published in full, including the one where a finding that worked in both training years died on its single real test, and the one where we built a fix for our own board and the fix turned out worse than the thing it fixed.
05
When we are wrong, we say so where we were wrong.
Corrections are appended, never swapped in. The original sentence stays. You should be able to watch us change our mind with the receipts in hand, rather than take our word for it afterwards.
What it costs usOur sprint-speed study ends with a correction we appended to it the same day, because a second study — run hours later — refuted a sentence in the first one. The sentence is still there. So is the refutation.
06
No black boxes. Every number printed has a definition one click away.
The engine is built from public play-by-play, and every stat on every page is defined in the Legend in plain language. If a term appears on a page without a definition behind it, that is a bug, and a test suite fails for it — a dotted underline that goes nowhere is worse than plain text, because it promises help and withholds it.
What it costs usWe printed RE24 in four places and defined it in none. A reader who met it had no way out. That is fixed, and now a machine checks every link on every page rather than trusting us to remember.
07
When the engine can't see something, we print it anyway — and say we don't rank on it.
Our expectation model grades a batted ball on how hard and at what angle it was struck. It cannot see legs. So fast men beat it, every year, and the fade board fills up with them. We tested the correction; the corrected board picked worse candidates. So sprint speed is printed on the fade board and the engine keeps its hands off it — the reader gets to see the thing we decided not to use.
What it costs usIt would be easier to quietly drop the column and let the board look cleverer than it is. It is on the board.
08
We take positions only after publication — never before.
The call goes on the record first. Affiliate links are marked as affiliate links, every time, and the books are open at the Till. We would rather be poor and legible.
What it costs usEvery dollar this site makes is itemized in public.
Hold us to it
Every rule above is checkable from the outside, which is the point. The calls are on the ledger with their timestamps. The research is published in full, failures included. The data is downloadable. If you catch us breaking one of these, you will not be telling us something we already knew and hoped you wouldn’t notice — you will be doing the job this page asks you to do.