The record
Every call he makes, graded. Including the ones he got wrong.
We have not found another app in this market that publishes how often it was right. If one exists, tell us and we will link it. This page is where ours goes, and it is published before we charge anyone.
Each point will be one month of sealed predictions: the share whose result landed inside the range Jerry stated before the set. The line is the target the app aims at, four in five. Nothing is plotted because nothing has been graded yet. Publishing a day is not the same thing as grading a month, so this chart stays empty while the list below fills.
Reading the published record.
If this does not fill in a moment, open the repository and read it for yourself.
Source: github.com/jeremiah2060/orderedstrength-receipts. Copied into this page when it was built, on an earlier date, so this page answers whether or not GitHub does. The repository is where to check the state right now. A day that has been published is never rewritten. Three separate things hold that: the job that writes it refuses to touch a date that already holds a file, the repository refuses a force push and refuses deletion of its main branch, so nobody, ourselves included, can rewrite that history without removing that rule first, and every write is a public commit, so a root that changed would be a visible change to that history rather than a quiet edit.
A day is sealed when it ends, at midnight UTC, which is early evening across the Americas, and pushed in the small hours after it. So a prediction you seal today is not on this page today: it appears in tomorrow's line, and until then it is held. That wait is the whole mechanism, not a delay in it.
Two different claims
What a receipt proves, and what only the anchor can.
Was it edited afterwards?
ANSWERABLE TODAY
A receipt carries the prediction, a random nonce, and a fingerprint taken over both. Recompute it and you know whether a single character moved. You can do that right now on the verifier, or break one yourself on the home page.
Was it made before the answer was known?
ANSWERABLE, TO THE DAY
Every fingerprint we are holding is folded into one root hash per day and pushed to a public repository. If yours is inside a published day, it existed no later than the moment that day closed, and the record of when that push happened is GitHub's rather than ours. That rules out the accusation that matters: a prediction cannot be invented after the result and slipped into a day that is already public. What it does not give you is the hour. Your seal and your set happen inside the same day, so this dates the prediction to the day, not to the minutes before you lifted. The second half is here because the first one is the flattering one.
Why it starts empty
Backfilling would destroy the only thing this is for.
We could generate anchors for every prediction already sealed on a phone and publish them today. The page would look established. It would also be worthless, because a record assembled after the results were known proves nothing about when anything was decided.
So the record starts on the day it starts, and the gap before it is stated rather than hidden. That is the whole product in one decision.