The Rock, the Valves, the Filter and the Mark: four thought experiments, each performed in the laboratory.
What one hundred experiments corroborate, what they fix, and what they leave to the theory.
Part III: The Theory
“Do the experiments confirm ACT?” has a precise answer, in two parts.
Four pictures, each one an experiment that has been performed. The Rock: a lossy slab that nothing reads still kills the fringes: a bath, not an observer, keeps the record. The Valves: the rate at which the fringes fade can be dialled. The Filter: an apparatus picks components out of one wave, as a radio picks a station; only a record picks the outcome. The Mark: single marks build a wave’s fringes when no record of the path was kept, and a particle pattern when one was. Seventy-six published experiments corroborate the four; ACT is built from nothing else.
Any one-history account must say when a fact appears. The laboratory has mapped the boundary: event rules of ACT’s kind that make an outcome actual too early are ruled out. Rules placed at or after release all reproduce quantum mechanics, so the data cannot choose among them.
What the experiments settle is how early an event can sit: release is the earliest placement the data allow.
Remove the observer. The record remains.
Send electrons through a double slit and they build fringes. Take away the detector at the slits and put a plain slab beside one path, with nothing reading it. The fringes die anyway. A resistive slab kills them. A good conductor barely touches them. An ideal superconductor should leave them alone. Nothing observed anything. The slab’s own electrons, stirred by the passing charge, carried a record of the path off as heat. That is the bath. For a slab, the record follows the dissipation: no dissipation, no record.
Sonnentag & Hasselbach 2007: electron waves passing micrometres above a doped-silicon plate lose their fringes near it. Kerker et al. 2020: beside gold, a million times less lossy, they barely change. Hackermüller 2004: a hot molecule’s own thermal radiation is a rock. And the fringes themselves, from electrons to 172 kDa nanoparticles (Pedalino 2026): the wave is a wave at every mass yet measured. Nineteen experiments, and one control, atoms that pass a surface and keep their fringes (Perreault & Cronin 2005): the full table is on the Evidence page.
Establishes that a bath, not an observer, keeps the record; ACT reads it as principle III: a detector is any body that keeps a path-dependent irreversible record. Could have failed if interaction without a record had killed fringes; it did not. Does not establish that an event occurred in any one run: the ensemble is a mixture.
Tune the bath. The rate follows.
Keep the rock and add a valve: the pressure of a gas, a laser that heats the molecules, the current in a detector beside one path, the gap between the paths. Turn it, and the fringes fade faster or slower, smoothly, by one rule in every kind of bath: faster for a stronger coupling, a hotter bath and, up to a point, a wider gap. The bath has a rate, and the laboratory can dial it. One valve is a fake: classical noise injected from outside dims the fringes in the average but records nothing, and ACT counts it as no event.
Brune 1996: a cavity-field cat decoheres faster for larger separation, with decoherence and damping tied by one coupling. Deléglise 2008: the decay filmed, 17 ms measured against 22 predicted. Buks 1998: a detector current dephases electron interference continuously. Yin 2013: a valve on a resonator’s emission switched in real time. Fifteen experiments, and one control, the spin echo (Hahn 1950): the full table is on the Evidence page.
Establishes that the bath has a rate the laboratory can dial, and the experiments already existed; ACT reads it as sentence 4 of the Picture. Could have failed if decoherence had been all-or-nothing or independent of the coupling. Does not establish anything beyond decoherence theory; the one-history reading is ACT’s.
The apparatus picks out components. Only a record picks the outcome.
An FM radio picks one station out of the air and leaves every other station on it. A cavity does the same for an atom’s light: tuned to the atom, it speeds the emission; with no mode at the atom’s frequency, it suppresses it. A Stern–Gerlach magnet sorts an atom’s wave into two components along its axis; turn the magnet and it sorts along the new axis. The magnet finds nothing out: bring the two beams back together and they interfere. Which beam a detector finds the atom in is settled only when a record is released.
Gerlach & Stern 1922, Rabi 1938: the apparatus selects the axis; a prepared mode answers predictably. Goy 1983, Hulet 1985: a cavity speeds an atom’s emission, and plates that leave no mode suppress it. Turchette 2000: change the reservoir, change the pointer basis. Margalit 2021: the Stern–Gerlach split recombines in a full loop. Unden 2019: only pointer information is copied across the environment. Thirty experiments, fifteen of them the boundary two slides on: the full table is on the Evidence page.
Establishes one composite field, whose components an apparatus can pick out and bring back together: picking out components is unitary and reversible. ACT reads it as principle II and the what and which of the event law: anchoring is release, on the one actual history. Could have failed if the apparatus had fixed the outcome, or the split had been irreversible, or emission had ignored the modes on offer. Does not establish which alternative wins.
The wave is what exists. The particle is what happens.
Send electrons through two slits one at a time. Each arrives as a single mark, one dot on the screen. With no record of the path, the dots build up fringes: the pattern of a wave. Put the rock beside one path and the dots build up a plain band: the pattern particles would make. Every dot is the same kind of mark either way. The record changes how the marks add up because it changes when the first event of the run happens: at the rock, or at the screen.
Grangier, Roger & Aspect 1986: one photon, one click, never both, and the same photons interfere. Tonomura 1989: one dot per electron, filmed. Sonnentag & Hasselbach 2007: close above a lossy slab the fringes wash out. Nagourney, Sauter & Bergquist 1986: a single ion blinks in steps. Kimble 1977, Diedrich 1987: antibunching: after a mark the hazard restarts at zero. This did real work on ACT. Paper I’s hazard, extrapolated naively to a driven atom, gives g(2)(0) = 1; the general event law of Paper II gives zero, as measured. Twelve experiments: the full table is on the Evidence page.
Establishes that every detection is one mark, and no event leaves two; ACT reads it as the last link of the chain, the particle is what happens, and principle I: the mark is where a mode enters causal time. Could have failed if one event had ever left two marks, or the marks had built fringes after a record of the path was released, or the frequencies had failed Born. Does not establish that the marks’ history is the one actual history: that is the one commitment.
The Filter’s magnet anchors nothing. Fifteen experiments mark where a record is still not a fact.
An atom in a cavity takes its excitation back. A Rydberg atom coupled to one cavity mode emits and reabsorbs, reversibly: vacuum Rabi oscillation (Brune et al. 1996). No event has been anchored.
A record that can be read back is not yet a fact. Which-path information held by a single photon can be erased, by choices made late or at spacelike separation, and the conditioned fringes return (Kim 2000; Walborn 2002; Ma 2013, over 144 km). A partial measurement of a qubit can be undone (Katz 2006, 2008), and a quantum jump already under way can be caught and reversed in flight (Minev 2019).
No record, no event until detection. In Wheeler’s delayed-choice experiments the decision to close the interferometer is made after the photon, or the atom, has entered. No environmental record is involved, and the result follows the final configuration (Jacques 2007; Manning 2015).
Emission is not final while the carrier can come back. Light returned by a mirror enhances or inhibits an ion’s own spontaneous emission (Eschner et al. 2001). A qubit coupled to surface acoustic waves at two points decays non-exponentially, because the phonons return (Andersson et al. 2019).
Each rule is read as ACT’s kind of event rule: the variable becomes actual at the moment named, and the actual history proceeds from that record, the global wave untouched.
| When an event would happen | Contradicted by | Status |
|---|---|---|
| At first interaction | Brune 1996: emission into a cavity mode is reversible; Margalit 2021: the Stern–Gerlach split recombines; Perreault & Cronin 2005 and Kerker’s gold run: interaction without a record keeps the fringes | Ruled out |
| When the reduced state decoheres | Chapman 1995; Kim 2000; Walborn 2002: conditioned fringes return. Katz 2006, 2008; Minev 2019: a partial measurement is undone, and a quantum jump is reversed in flight | Ruled out |
| When a record carrier exists | Eschner 2001; Andersson 2019: the returning field alters the source’s own emission | Ruled out |
| At release — the record is beyond coherent recovery by the system’s accessible future | — | Survives (ACT) |
| Later, never, or always | — | Survive |
ACT’s rule sits exactly on the boundary the data allow, at the earliest admissible place; every experiment in the table refuted an earlier placement, and ACT’s rule could only have failed had coherence returned from a record it counts as released, which has not happened. Later rules survive too, because every rule placed at or after release reproduces quantum statistics.
Quantum mechanics predicts recoherence exactly where a record remains coherently recoverable, and ACT places its event exactly where recovery is impossible. So ACT’s placement is refutable precisely where unitarity is, by a recoherence that fails to appear, and it has survived every such test because unitarity has. That is why two words stay apart on this site:
What the experiments support. ACT places the fact at release; that boundary has been mapped, and it has never been refuted.
What a result only ACT predicts would do. There is none, because ACT keeps the dynamics; whether anything happens at the boundary is the postulate’s, and ACT claims no experiment for it.
The boundary is also relative to what can be controlled. Spin dynamics spread through a solid have been time-reversed and lost magnetization returned as an echo (Rhim, Pines & Waugh 1971); information scrambled across 53 qubits has been refocused (Mi et al. 2021). So ACT’s accessible algebra is fixed by the physics of the coupling and the apparatus, as quantum-trajectory theory also requires.
Three results no ontology may contradict, and what ACT says to each.
Loophole-free violations (Hensen, Giustina, Shalm 2015; Storz 2023) rule out every local account with free settings. ACT is nonlocal in Bell’s sense, by outcome dependence only: the other end anchors from what is left of the one structure. No signalling.
Violations with weak continuous measurement (Palacios-Laloy 2010) and ideal negative measurement (Robens 2015) and state-independent contextuality (Kirchmair 2009) rule out definite but unknown values. On ACT, what is not yet anchored has no value to be unknown.
Extended-friend inequalities are violated when the friend is a photon or a GHZ state of up to 17 qubits held reversibly (Proietti 2019; Bong 2020; Zeng 2025; Laux 2026). No record was released there, so ACT assigns such a friend no event.
A universal collapse noise would heat matter and make it radiate.
Excess radiation in underground germanium and xenon detectors (Donadi 2021; Arnquist 2022; Aprile 2026) and excess force noise in space and on cantilevers. None found. Parameter-free Diósi–Penrose collapse and the original GRW rates are excluded under white-noise assumptions; coloured and dissipative versions remain alive.
ACT adds no dynamics to quantum mechanics, so it predicts no excess radiation. Had an excess been seen, ACT would have been refuted along with the unmodified dynamics it keeps; none was. Its noise and its damping are the bath’s own, paired by the fluctuation–dissipation relation (Lecture 4).
These are tests ACT could have failed. It did not; the collapse models are the ones being squeezed.
ACT places an event only where a record has become irreversibly unavailable to the system’s accessible future. That boundary has been mapped independently: coherence returns exactly when, and only when, the record remains coherently recoverable.
Rules that make an outcome definite at first interaction, at decoherence of the reduced state, or at the first appearance of a record carrier are ruled out. No experiment distinguishes ACT from later event rules, or from interpretations that reproduce quantum mechanics without events. Every detection is one mark, as each of them predicts, and ACT takes that mark at face value. Three experiments would still tell it something:
Kerker’s planned superconducting surface: the same electrons, the same geometry, no dissipation. ACT, with decoherence theory, says the fringes stay. A loss of contrast beside a lossless body would be a record with no carrier, and the Rock would fail.
Put a weak, tunable probe on the carrier that returns to its source (Eschner 2001; Andersson 2019) and watch the return-induced modulation degrade as the probe strengthens. It maps release continuously, where Paper II says an in-flight record can still be overturned, and its graded release bound is an inequality the data can check.
A levitated particle decohering in a thermal photon bath, with the recorded observable reconstructed from environment fragments as in the quantum-Darwinism experiments. It asks what the environment selects when nothing is engineered — the ledger’s sharpest open item — and could embarrass the alignment conjecture.
The full tables, with what each experiment corroborates and does not establish, are on the Evidence page.
The laboratory has shown all three, and where a fact can first appear. ACT says what appears there, one mark in one history, and adds one commitment to say so.
Next: Lecture 9 — Ontology Recapitulates Mathematics