cert-machine · report · generated from the records

The impostor catalog

Published constants that agree with simple closed forms for 16 to 62 significant digits — and exact proofs that every one of them is lying. Digit agreement to any depth is not evidence, and this page is the demonstration: each of these impersonations would pass any decimal screen ever used to announce a discovery.

tl;dr
  • The finding. Twenty-one published constants agree with simple closed forms for up to 62 significant digits — and every one is provably NOT that form. Digit agreement is not evidence; an answer key validated by digit matching inherits exactly this failure class.
  • The mechanism. One BigInt integer comparison per claim at the full published digit length — no floating point anywhere — separates the impostor from the identity.
  • Check it. node tools/test-engine.js — the engine battery pins A019762 as a permanent regression control.
constants
5
Published OEIS decimal expansions, audited among 14593 — the only survivors of a 17-digit double screen.
impersonations refuted
21
Every spelling decided by one BigInt integer comparison at the full published digit length.
deepest agreement
62 digits
Before the first disagreeing digit. A 20-digit match is called a discovery elsewhere.
false refutations
0
The same instrument keeps every true closed form in the corpus — gated by the engine battery, every build.

Published, not peer-reviewed, not independently rerun. Verdicts are conditional on the OEIS published digit streams being correct — a refutation here reads "not equal, given those digits".

§1 · the method

One integer comparison, no floating point

The engine's double-precision screen tests each constant against ~14,000 closed-form spellings at 17 significant digits. For 5 constants, at least one rational value survived that screen — the constant and the form genuinely share 17 leading digits. Those survivors are re-decided at the FULL published precision, exactly:

the form p/q is consistent with the digits ⇔ D·Q ≤ P·10^(k−1) < (D+1)·Q

where D is the published digit stream read as a k-digit integer and P/Q is the form's mantissa scaled to [1, 10). Everything in that line is a BigInt; the verdict is an integer comparison. No float participates, so there is no precision to argue about: the form either fits inside the digit interval or it provably does not.

The instrument is calibrated in the other direction on every build: constants whose closed form IS the truth (sqrt 2, pi, e, phi, ln 2 …) must keep their forms — zero false refutations is a gate in tools/test-engine.js, and it once caught this engine refuting sqrt(2) as the closed form of the decimal expansion of sqrt(2).

§2 · the catalog

Who they impersonate, and for how long

entrywhat OEIS says it isimpersonatesas spelled by the vocabularydigits agree / publishedverdict
A271880Decimal expansion of the probability that a random real n…22/1 sqrt(4/1) cbrt(8/1) sqrt2^(2/1)62 / 105refuted exactly
A271880Decimal expansion of the probability that a random real n…2020/162 / 105refuted exactly
A271880Decimal expansion of the probability that a random real n…1/51/562 / 105refuted exactly
A181284Decimal expansion of sqrt(9/121 * 100^m + (112 - 44*m)/12…3/113/1158 / 105refuted exactly
A181284Decimal expansion of sqrt(9/121 * 100^m + (112 - 44*m)/12…30/1130/1158 / 105refuted exactly
A359187Decimal expansion of the real part of (-sqrt(2))^^9, wher…11/1 sqrt(1/1) cbrt(1/1)44 / 87refuted exactly
A359187Decimal expansion of the real part of (-sqrt(2))^^9, wher…1010/144 / 87refuted exactly
A359187Decimal expansion of the real part of (-sqrt(2))^^9, wher…1/101/1044 / 87refuted exactly
A226120Decimal expansion of Sum_{n>=1} n^3/(exp(2*Pi*n/7)-1).11/1 sqrt(1/1) cbrt(1/1)16 / 100refuted exactly
A226120Decimal expansion of Sum_{n>=1} n^3/(exp(2*Pi*n/7)-1).1010/116 / 100refuted exactly
A226120Decimal expansion of Sum_{n>=1} n^3/(exp(2*Pi*n/7)-1).1/101/1016 / 100refuted exactly
A266296Decimal expansion of a number close to 24, related to the…2424/116 / 105refuted exactly
A266296Decimal expansion of a number close to 24, related to the…12/512/516 / 105refuted exactly
A266296Decimal expansion of a number close to 24, related to the…6/256/2516 / 105refuted exactly

Multiple spellings of one value (sqrt(4/1), cbrt(8/1), sqrt2^(2/1) are all 2) are listed together and counted separately — the vocabulary generates forms, and every generated form that fit at 17 digits was individually put to the exact test.

§3 · why this matters

The collision-probability argument, run in reverse

Digit-matching pipelines — the Ramanujan Machine's published methodology among them — declare a hit when a computed value agrees with a closed form to some tens of digits, and argue from collision probability that the match is almost certainly an identity. This catalog is the reverse experiment on the same kind of data: here are constants that sustain the match to 62 digits and are provably NOT the number they resemble.

A271880 — the probability that a random real is "evil" — agrees with 1/5 for 62 digits before diverging; OEIS records the difference separately (A271881, about 2.17e-64). Any screen shallower than 64 digits calls this 1/5.

A266296 is literally named "a number close to 24, related to the Ramanujan constant" — the near-integer phenomenon that made e^(pi sqrt 163) famous, caught here impersonating 24 for 16 digits and refuted at 105.

The point is not that these five are surprising — OEIS contributors know them. The point is that a probability argument cannot tell these five from a real identity, and an exact decision can. A pipeline whose screening step can mint false positives is indistinguishable from one that cannot; the cure is a certificate, and each row of this catalog carries one: a single integer inequality anyone can recheck.