A physics engine, and a verification tool enforced on top of it. PhysWall inverts a closed, published, non-linear physical law at a single measurement point — and refuses when the inverse is not unique. Seven laws, one engine, and the same refusal in all of them.
A physical-logic engine that decomposes a measurement gap along four axes — the machine, the law, the object, and which definition — and runs every domain through one structure.
RF matching, PCB loss, orbital conversion, photonic scattering, stream gauges, ocean waves, basketball. Same engine, same refusal: when the measurement cannot carry a conclusion, it says so instead of answering.
A unit, a correction, and a version number nobody
Averaging reciprocals and inverting the average are not the same operation, and the gap between them is Jensen's inequality. For positive numbers the arithmetic mean is never below the harmonic mean, so one of the two answers is always the larger.
Statisticians call the general case the fallacy of the average. It is not an obscure trap: it appears whenever a rate is averaged instead of a time, and the printed unit is usually the only sign that it happened.
In December 2025 a group at Fraunhofer EMFT published a result worth having: across eight superconducting qubits, ninety-five hours of measurement and ten cooldowns, the fluctuation in qubit lifetime follows one coefficient — the same one, regardless of how the qubit was made.
A unit printed beside that coefficient did not follow from their own exponent. We worked out which step had gone wrong, wrote it up, and were about to send it. Then somebody asked which version we had read.
Eq. 1 σT1 = a · ⟨T1⟩3/2 Fig. 5 a = (1.220 ± 0.052) × 10-2 s2/3
The exponent 3/2 appears three times: in Eq. 1, in the supplementary derivation, and in the summary, which calls the scaling superlinear.
A coefficient multiplying a time to the power n, returning a time, carries units of s1−n. For n = 3/2 that is s−1/2.
n = 3/2 1 − n = −1/2 ← what the exponent requires 1 / n = 2/3 ← what was printed
Two-thirds is exactly the reciprocal of the exponent, where the complement was needed. A single step, and naming which step is more useful than naming the discrepancy.
Submitted 13 July 2026, six weeks before this page was written. The Fig. 5 caption now reads:
a = (15.9 ± 0.9) √Hz √Hz = s−1/2
Exactly the correction we were about to send. Made a month and a half earlier, by the people who wrote the paper.
And the number moved too, not just the unit: they refitted. Version 1 extracted the mean and standard deviation by integrating over a Rician fit; version 2 does it empirically. The comparison set changed as well.
the dimensional argument stands which step went wrong stands "the paper is wrong" withdrawn our own number, 12.2 superseded — theirs is 15.9
The analysis was right and the target had moved. On a site whose whole premise is that every constraint carries its source, the step that was missing was checking whether the source had been revised.
The microsecond argument on this page rested on a supplementary table stating that time values are given in microseconds. That table is a Ramsey parameter table — T2*, not T1. The conclusion is probably still right, but it is an inference from a neighbouring table, and this page presented it as a direct quotation.
⚠ Both failures are the same one: a claim about a source that was not checked against the source as it now stands. Neither is a physics error, and neither would have been caught by any amount of rechecking the arithmetic.
A version check now runs before anything citing an arXiv paper is published: fetch the abstract page, compare the version and revision date against what was read, and stop if they differ. It takes one request. It would have caught this.
The finding is kept on the site rather than deleted, because a page about checking your sources that had to be rewritten for not checking its own is worth more standing than gone.
PhysWall was developed and architected by Gadi Zion.
Built on PhysWall — the same engine reads antenna bandwidth, conductor loss, bit erasure and heat limits. It answers what the measurement implies, and refuses when the measurement cannot say.⚠ Check this instead of believing it. Every number here reproduces from a source that is named, and the claims that turned out wrong are still printed next to what replaced them. The same engine runs all of these — it asks how much a measurement allows you to conclude, and refuses the same way in every field. The same engine runs all of these — it asks how much a measurement allows you to conclude, and refuses the same way in every field. How to check each one →