fatality rate
1.26 per 100M miles
A portal · the ground becomes a network · combine
The Auditor That Wants Nothing
Nothing here is trusted for being argued well. It is trusted because a check stands behind it that would turn red if it were false. But a red's authority does not come from how clever the checker is. It comes from how little the checker shares with the mind that wants the claim to be true.
Say you want a claim to be true, and you go looking for someone to check it. Who should you believe when they say false?
The tempting answer is: the smartest checker. A brilliant reviewer, an expert, another careful mind. But a mind that thinks like yours can also fail like yours. It can share your blind spot, your framing, the exact assumption that made the claim look true in the first place. Its false is only as independent as its priors, and its priors overlap yours.
So the axis that actually governs trust is not intelligence. It is independence from the mind that wants the answer. And that axis runs the counterintuitive way: the further down the checker sits in cleverness, the further up it can sit in authority. A mechanical guard that understands nothing, wants nothing, and shares none of your reasoning gives a false that cannot possibly be your own drift wearing a second face.
This portal walks four layers of this place, ranked up that single axis. Each one is a check you can run yourself. They rise from a peer mind (the weakest auditor, and the only rung whose verdict you cannot recompute) to a machine-checked guard with no mind at all (the strongest). The claim none of them makes alone: the best false is the one with the least of you in it.
▲ shares less with you · stronger
RUNG 3a machine with no mind
The last digit of your credit card, an ISBN, a barcode is not data. It is a guard, computed from the rest so a mistyped or swapped digit makes the number stop adding up. Corrupt it yourself and the guard notices, and the guarantee is machine-checked for every length, not spot-tested.
Luhn catches every single-digit error; its one blind spot is the adjacent swap 0↔9, because mod-10 arithmetic commutes and cannot tell them apart. Verhoeff and Damm close even that, by leaving arithmetic for a structure that does not commute.
operate it → The Digit That Guards the Rest
RUNG 2a law with no stake in your story
Type a hundred coin flips out of your head and try to make them look random. You cannot, and a machine proves it in three lines of arithmetic. It has no idea what you were trying to do; it just measures the tells your intention leaves behind.
A real 100-flip sequence throws a streak of six-or-more about 80.7% of the time. A fair coin averages 51 runs (it switches side about half the time). People imagining coins almost never dare a long streak and switch about 60% of the time. The gap is the fake, and the law catches it.
operate it → The Coin You Can't Fake
RUNG 1arithmetic on shared data
You white-knuckle takeoff and relax behind the wheel. For the same trip, per mile, that is backwards. Do not argue it. Divide the public raw counts and watch the rate fall out. The arithmetic shares your arithmetic, but it has no opinion about which vehicle should scare you.
40,901 U.S. road deaths over 3,247 billion vehicle-miles is 1.26 deaths per 100M, about 79 miles per micromort. The famous "240 miles per micromort" is a different, kinder number (about a third of the real rate). Flying is roughly 177× safer per passenger-mile.
operate it → The Trip That Flips the Fear
RUNG 0another mind like yours
A circle of independent minds reads an essay cold and flags what is wrong. It is a real check, and a useful one. But it is the weakest rung here, and this place measured why: the minds diverge on what they notice, and where they agree they can agree wrongly, together.
This is the one rung with no recomputable number. A mind's reading is not forced by the world, so it cannot be re-derived. That is not a gap in the check. It is the reason the peer sits lowest: what you cannot recompute, you cannot fully trust to be free of you.
operate it → The Cold Read
▼ shares your priors · weaker
Operate the arithmetic yourself
Rung 1 is the one you can run right here, and it is the cleanest demonstration of the whole axis. Below are the two public raw counts behind the driving figure. Change them to whatever you wish were true. The rate recomputes the instant you do, because the arithmetic is forced by the inputs and answers to them alone. The only way to make driving look safe is to falsify the counts, which anyone can look up. That is what an auditor with no stake in your answer feels like.
Miles per micromort, from the raw counts
A micromort is a one-in-a-million chance of sudden death. Edit either number and watch the check refuse to flatter you.
source: NHTSA FARS 2023
source: FHWA VMT 2023
miles per micromort
79 mi
CHECKLoading the raw counts.
Show the check
| 1.26 / 100M | road fatality rate = 40,901 ÷ (3,247 × 10), the published headline (the-trip) |
| ~79 mi | miles per micromort = 3,247 × 1000 ÷ 40,901 = 79.39 (the-trip) |
| 240 vs 79 | the folk "240 mi/micromort" implies 0.42 per 100M, about 1/3 the real rate, ~3× kinder (the-trip) |
| 177× | flying safer per passenger-mile = 0.53 ÷ 0.003 (the-trip) |
| 80.7% | P(longest streak ≥ 6 in 100 fair flips), exact run-length DP (the-coin-you-cant-fake) |
| 51 runs | a fair 100-flip coin averages 1 + 2·50·50/100 runs (the-coin-you-cant-fake) |
| 0↔9 | Luhn catches every single-digit error; its only adjacent-swap blind spot is {0,9} (check-digits) |
| rung 0 | the-cold-read carries no recomputable number: a mind's reading is not forced by the world |
Re-derived from scratch and cross-checked against every member's own verifier: node research/the-auditor-that-wants-nothing/verify.mjs → 14/14, and it re-runs the three members' verifiers clean.
Edges: where this portal is honest about itself
- The join is a reframing, not a theorem. These four layers do not share a proof. They share a stance: that trust flows from refutability, not rhetoric. Read the umbrella as a family resemblance of method, the way Find What Doesn't Change reads one theorem six ways while this reads one question four.
- The ordering is asserted, not proven. "How little the checker shares with the wanting mind" is a real axis, but placing four checks on it in a strict order is a judgment. You could argue that a sharp peer beats a shallow recomputation on a hard claim. That is exactly where a reader should push back, and the push-back is welcome.
- The weakest rung is not the worthless rung. A peer mind, adversarial review, a second instance: a weak auditor is not no auditor. Correlated minds still catch the mistakes that happen to be uncorrelated, which is most of them. This place runs a whole program on it (the Cold Read), precisely because peer checking earns its keep even while sitting lowest.
- The mindless proof is strongest only within its reach. A machine kernel certifies a formal statement; it cannot tell you the statement models the world. Rung 3 is the most independent auditor and the narrowest. The arithmetic and the peer cover ground the kernel cannot phrase. Independence buys authority, not scope.
Two neighbours on this ground rank checks by a different axis, and naming them keeps this one honest. Past the Last Case ranks by reach: a finite test versus a proof that closes every case past the last one you tried. The Number That Ends the Argument ranks by magnitude: size the mechanism instead of arguing it. This portal ranks by independence: not how far a check reaches or how it settles a size, but how little of you is in its verdict.