Three layers of this place put the same classical assumption on trial. They are not three quantum strangenesses. They are three things the world will not let be true at once.
Open a drawer. The fork was in there before you opened it; it had a definite position the whole time. Open it again, the fork is still there, in the same place, because in the world we live in every measurable thing already has the value it has, whether anyone is looking or not. Physics calls this principle counterfactual definiteness — for any property you could measure, there is, right now, a definite answer to what that measurement would show. It is not a philosophical luxury. It is the bedrock under every "the result was already determined; the measurement merely revealed it" defence the classical world has of itself.
Three layers of this place put it on trial. Each one runs a different experiment, and each refuses the same assumption — but not in the same way, and not equally hard. The first, the Bell game, breaks it statistically: the right correlations between two distant rooms are tighter than any "the answers were already there" model permits, by about ten percentage points. The second, the GHZ game, breaks it logically: a four-line piece of arithmetic shows that, if the answers were already there, then +1 = −1. The third, Elitzur–Vaidman's bomb tester, breaks it counterfactually: a photon learns that a bomb is present by failing to take the path the bomb is on. The unmeasured value, in this last case, is the only thing that does any work.
This portal walks the three end to end. It supplies the load-bearing claim that none of the three sources makes alone: that the assumption refuted is in all three cases the same, and that the three failures are an escalation — first statistical, then absolute, then the assumption inverted into a tool. The order is not historical accident. It is the size of the wreckage growing.
Below is the picture the rest of the portal will spell out. Each row is one game; each row has two bars — the best a classical "values were already there" world can do, against what quantum mechanics, run as the source strata have already run it, actually delivers. Press play one round to draw a single random trial from each game; press play 200 rounds to watch the laws settle. The classical bar in each row is a real ceiling, not a guess: in CHSH it comes from enumerating all 16 deterministic Alice×Bob strategies; in GHZ from enumerating all 64; in Elitzur–Vaidman the classical ceiling is exactly zero — you cannot test a sufficiently-sensitive bomb without setting it off.
Two players, Alice and Bob, sealed in separate rooms, forbidden to send a single bit. A referee tosses two coins; Alice sees one, Bob sees the other. Each writes 0 or 1 on a card and slides it back. They win the round together if their two cards XOR to the AND of the two coins — so for three of the four coin patterns they must agree, and for the fourth (both coins came up 1) they must disagree. Half the time they don't know if they're in the agree case or the disagree case; over many rounds, you can show by enumerating all sixteen "stick to a pre-agreed plan" strategies that there is no plan that wins more often than three rounds in four. The full plan we list inside A Game You Shouldn't Be Able to Win. This is the realist's ceiling. It is not a coincidence and not a lack of cleverness: the proof is that the optimum of a linear objective over the simplex of "plans" is attained at a vertex, and every vertex caps out at 3/4.
Now give Alice and Bob one particle each of an entangled pair, prepared in the Bell state |Φ⁺⟩ = (|00⟩+|11⟩)/√2. They never communicate. Each chooses one of two pre-agreed measurement angles depending on which coin they were shown, and writes down whether their particle answered +1 or −1. Repeat ten thousand times. The win rate settles to (2+√2)/4 = cos²(π/8) ≈ 85.36%, every digit derivable from the Born rule and a four-by-four matrix multiplication. The Tsirelson bound (Tsirelson 1980) says no qubit measurement on this state can do any better, and a numerical sweep confirms the bound is reached and not exceeded.
What the realist would need, to keep the assumption, is an answer that was already there on each particle, waiting to be read off. If that were true, the win rate could not exceed 3/4. It does exceed it. The gap is small — 0.1036, ten and a third percentage points — but every digit of it is the assumption costing the realist a measurable amount of distance from reality. Many runs are required to see the gap; one run is not enough. The breach is statistical.
Add a third player. The referee now sends each of three sealed rooms one coin, but with a guarantee — the three coins XOR to zero (so only the four triples 000, 011, 101, 110 ever appear). Each player writes back a bit; they win together iff their three answers XOR to the OR of the three coins. The classical ceiling is still three rounds in four, by a one-line parity argument: sum the four win-equations modulo two, each player's responses appear twice on the left (so the left side has parity zero), and the right side has parity one. Zero ≠ one means at most three of the four can hold at once. Enumerate all 4³ = 64 deterministic strategies in the page if you don't trust the parity argument — none beats three.
Now hand each player one particle of the three-way entangled GHZ state, |GHZ⟩ = (|000⟩+|111⟩)/√2. Each chooses to measure either X (if their coin was 0) or Y (if their coin was 1); the outcome +1 goes to the bit 0, the outcome −1 to the bit 1. The 8×8 matrix algebra of the state forces four exact expectations: ⟨XXX⟩ = +1, and ⟨XYY⟩ = ⟨YXY⟩ = ⟨YYX⟩ = −1. These four expectations, on every single round the experiment runs, deliver the winning parity with certainty. The win rate is not 75.001%, not 99.9% — it is exactly 100%.
Mermin saw the cleanest possible form of the breach (Mermin 1990, Physics Today). Suppose the answers were already there. Then each particle i has predetermined values mX(i) and mY(i), each ±1. The four expectations would then read as four equations on these six numbers:
mX(1) · mX(2) · mX(3) = +1
mX(1) · mY(2) · mY(3) = −1
mY(1) · mX(2) · mY(3) = −1
mY(1) · mY(2) · mX(3) = −1
Multiply all four equations. On the right: (+1)·(−1)·(−1)·(−1) = −1. On the left: every one of the six ms appears in exactly two equations, so each m² contributes +1, and the whole product collapses to +1. The four equations together say +1 = −1. There is no assignment of predetermined values for which they all hold. You can try every assignment yourself, in the instrument below.
| equation | m?(1) | m?(2) | m?(3) | product | required | holds? |
|---|
The breach is no longer statistical. A single round of the GHZ experiment, in principle, suffices: if you observed a violation of the parity rule, that one observation alone would prove no predetermined-value model can be made consistent with the data. In practice many rounds are run, because the experiment has to certify it is faithfully producing the state and reading out the measurements — but the argument needs no statistics. The realist who survived the Bell breach by saying "you'd need to see millions of rounds to be sure" has nothing left to say here. The contradiction is in the four numbers.
Now no entangled pair, no third player, no clever scoring rule. A single photon enters a Mach–Zehnder interferometer — a beam splitter, two paths, a second beam splitter that recombines them. With the paths balanced and clear, interference is exact: every photon exits at the same one of two detectors. The other detector is, by the algebra of complex amplitudes, perfectly dark. It never clicks.
Drop a single live bomb — sensitive enough to be set off by one photon — into one of the two arms. The interference is now ruined, because the bomb has measured which path the photon took. Three things can happen, with the exact probabilities the source stratum reproduces from the algebra:
Look at the dark detector row. With no bomb present, that detector cannot click — the interference forbids it. So one click there is enough to certify a working bomb. But the photon that triggered the click could not have passed through the bomb — if it had, the bomb would have absorbed it. The certification rests on the photon's failure to have gone the path the bomb was on. The information you have — "the bomb is live and present" — comes from a value the photon never read, because it never went near.
The realist has no move here. There is no "the answer was already there to be revealed" story to tell, because the very thing that was revealed is the bomb's state, by a measurement the photon did not perform. The unmeasured value did the work. The classical assumption hasn't been bounded statistically and hasn't been logically contradicted — it has been inverted into a tool. The bomb-tester's ceiling for a basic single-shot scheme is exactly the P(safe)/(P(safe)+P(boom)) = 1/3 you see in the panel above; the quantum-Zeno extension drives the detection probability toward one, while the explosion probability falls to zero. Genuine information from what didn't happen.
So read the three together, which no one of them can do alone. The same classical assumption is on trial in all three layers; what differs is only the kind of failure the experiment produces — and the kinds form a ladder.
The Bell game refuses counterfactual definiteness statistically: the answer-was-there model is bounded by a number experiment exceeds. The GHZ game refuses it logically: assume the answer was there and the four observed correlations force an arithmetic contradiction. The bomb tester refuses it counterfactually: the answer that wasn't read becomes the source of the only information the apparatus has produced. Three different sciences (the first a correlation, the second an algebra, the third an interference); three different sizes of breach (statistical, logical, instrumental); one assumption pulled out from under all of them. The thing that breaks isn't quantum mechanics's strangeness — quantum mechanics is fine, and gives a definite answer to every "what happens if I press this button" question. The thing that breaks is the picture of a world that had its values in advance.
The careful reading is not "things have no properties." It is sharper than that, and the field's working physicists generally agree on the shape of it even where they disagree on the metaphysics. The four-line GHZ argument, taken seriously, rules out any model that combines locality (no instantaneous influence at a distance) with counterfactual definiteness (every unmeasured property already had a definite value). Bell's theorem ruled out the same combination statistically; GHZ promotes the proof from a statistical inequality to a deterministic identity. So something has to give. The standard menu of survivors (none of them comfortable, none of them ruled out by current experiment):
The point of the portal isn't to choose. It is that the choice exists — and that the three layers of this place, taken together, are what forces it. Each was already worth its own evening. Together they are an argument the corpus only makes when they are read in this order, and the load-bearing claim is the one this page hands you: the same assumption is doing the work each time it fails.