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.
Bring the same numbers you gave the model. This refuses when they support two different answers at once — which a model will not do, because you did not ask it to.
Somebody gave you a number and something does not sit right. Say it however you would say it out loud — there is no right way to phrase this.
Never used it? The second button fills everything in and runs a real case — start typing at any point and it gets out of the way.
Things people have actually typed:
"the total does not look right to me" "my landlord says I owe 4,200 and I get 3,900" "the report says 94 but I counted 92" "the other lab got a different answer" "an invoice where the lines do not add up" "they say the average is 4.2 and I cannot see it"
⚠ If it is a laboratory measurement — uncertainty budgets, CMC, proficiency testing — the tool for that is here. This one is for everything else.
Or go straight to a form, if you already know which you need:
⚠ Worked out in your browser. This tool is not an inversion of a physical law, so there is no server answer to sign — it is three arithmetic checks, and it says so rather than implying otherwise.
⚠ worked out in your browser
Or pick one yourself, if you already know:
1 pick one of the three buttons (or describe what you have and let it pick) 2 the boxes change to match 3 press "Check it"
Nothing appears until you press "Check it". The answer and the drawing come after that, not before — which is worth saying because the person who asked for this page could not find the button either.
Most of what people are asked to accept has no figure in it. That does not mean no number is being claimed.
"most of the team agreed" "we get very few complaints" "it has been tested extensively" "significantly better than the old one" "this rarely happens" "the majority of our users prefer it" "there was overwhelming support"
Every one of those is a count wearing a word. Four people out of seven is "most". So is nine hundred out of a thousand. They are not the same claim, and the sentence does not tell you which one you are being handed.
most → of how many
very few → few out of what, over what period
extensively → how many times, and what happened
in the ones that did not work
significantly → compared with what, and by how much
rarely → how often is rarely
overwhelming → who was asked, and who was not
You are not challenging anybody by asking. The count exists — somebody had it in front of them when they wrote the sentence. Asking for it is asking to see what they saw.
Ask a model about "most of the team agreed" and it will discuss the sentence — thoughtfully, and at length. It will not stop to ask how many people are on the team, because you did not ask it to, and the sentence reads as complete without that.
Once you have the count, come back and one of the three checks will fit. Until then there is nothing to check — and knowing that is the useful part.
One of them is arithmetic. One rests on what you typed in a box. One depends on something this page cannot see. Treating them as the same finding would be the mistake this whole tool is about.
CERTAIN an average that does not match
its own readings -- recompute it
and you get the same answer
WORTH two entries naming the same
CHECKING source -- a name matched, and
only you know whether they are
really the same thing
DEPENDS two results disagreeing -- true
only if both plus-minuses mean
the same thing, which nothing
here can check
Each answer says which it is, and what would overturn it. That is more use than a confidence score, because it tells you where to push.
This is borrowed from the adjudication tool on this site, which grades claims the same way and reaches its precision without a single threshold — a hard constraint outranks a model, and a model outranks a reading.
It will not tell you your number is right. Nothing can do that from the number alone.
What it does is narrower and more useful: it finds the cases where your number cannot mean what you were told it means — a total that does not follow from its parts, one source counted twice, or a measurement that two different situations produce equally well.
When it finds one, it says which, and what would settle it.
⚠ 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 →
What this is, and where the numbers come from →
A checking tool, not professional advice. It tells you what a measurement does and does not support; what to do about that is your decision.
Developed and architected by Gadi Zion