Someone printed each claim below. Your browser checks it against the files it cites, in exact arithmetic, and answers in one word. There is no server: the checker is in this page.
Each claim has small changes you can make. Make one and the claim is decided again, here, as you watch: the word can turn, and it turns back when you reset. The strictest word wins, so one false part refutes a claim however many parts hold.
The checker is loading.
A claim that a quantity is exactly some number. The checker computes the quantity from the file the claim cites, in whole numbers with no rounding, and compares.
A two-number test file, written to try the checker on a sum too large for ordinary arithmetic.
The two whole numbers in the file add up to exactly 9223372036854775808.
The same claim, damaged on purpose in one place.
The same sum, with 9223372036854775808 written bare, without quotes.
A claim that a constant sits below a bar. Nobody knows the constant exactly, so a computation traps the bound between two exact endpoints, and the whole interval has to clear the bar.
Printed by the public registry of optimization constants, from a certificate by R. Pinho.
A constant from number theory, C82, is below 0.2534449.
A claim I decided before and keep in a public record. Your tab decides it again, and the claim, the record and the new answer all have to agree.
Printed by the public registry of optimization constants, from a step function with 300 steps.
A constant called C1d is at most 1.463071582607559.
Printed in the claimant’s own README, from a certificate of 3,584 whole numbers.
A constant called C1c is at least 0.41030211724.
The checker is the same program I run on my own machine, carried into this page unchanged, with the 20 files these 5 claims make it read: 258,918 bytes in all. It hashes every file before it trusts one, it does its arithmetic in exact whole numbers and fractions, and it refuses a number it would have to round.
A browser cannot rebuild the long computations behind the bounds, so for those it replays my published record of each one and holds the claim, the record and its own answer to each other. Where they disagree it refuses. The textbook machine shows why checking stays cheap while finding does not.