No Difference in the Mermin Violation Between Scheduled and Run-Time Measurement Settings in Deterrence-Era Bell Runs
Abstract
Scheduled and run-time measurement settings in Bell runs differ in one respect that matters to a forger. A schedule written before the run can be read by a sophon in advance, while a setting chosen at run time is recorded only when the operator enters it at the console, seconds before the detections it governs. A forger limited by advance knowledge of the settings should therefore do better on scheduled runs, and the Mermin violation should be smaller there than where settings were unknown. We compared 16 scheduled and 12 run-time runs from the Institute register for Deterrence Era, Year 51 to Year 59, with 412,800 and 316,800 triple events. The Mermin value was 3.226 (95% CI 3.219 to 3.233) for scheduled runs and 3.230 (95% CI 3.222 to 3.239) for run-time runs, a difference of -0.004 (95% CI -0.015 to 0.007). The data do not support the prediction. A blind forger responsible for more than 0.4% of the violation excess is excluded at one-sided 95%. The comparison does not bound a forger that reads a setting at the moment it is entered, and a run-time setting is still a recorded one. For the first time, the archive is sorted by what a forger would have needed to know.
1. Introduction
We find that Deterrence-Era Bell runs with settings fixed on a written schedule before the run gave the same Mermin violation as runs with settings chosen by an operator at run time. The difference in the Mermin value, scheduled minus run-time, was -0.004 (95% CI -0.015 to 0.007). That is, the runs that a sophon could have prepared for in advance show no loss of violation against the runs that it could not.
A Bell inequality bounds the correlations that outcomes fixed at the source can produce at separated stations, and states of three or more particles can exceed the bound. A violation certifies nonlocality only if nothing else wrote the record, and a sophon that alters logged outcomes would write exactly such a record. This is the so-called intervention loophole, and the question of what such a forger must know to produce a convincing violation is older than most attempts to answer it.
Pei and Hao gave the first formal model of outcome substitution at distributed detectors, and in it a substituted outcome is useful to the forger only if it matches the measurement setting at that station (Pei and Hao, 36 CE). Qin and Tao found no measurable lower bound on the relay latency of paired sophons at terrestrial baselines, so the station at which a setting is known does not limit a forger (Qin and Tao, 152 CE). Rong and Cen showed that three stations separate substitution at one station from nonlocality when the substituted share is large (Rong and Cen, 12 DE). Fan then observed that an archived setting carries a time of entry at the console and a time of the detection it governs (Fan, 28 DE).
A gap remained. No study has asked whether the violations of the Deterrence Era depend on when a setting became readable. Our earlier work certified a violation by local filters applied after the record was sealed (Hu, 59 DE), and it said nothing on what information a forger would have required. The present paper concerns that information and does not repeat the filter method.
This account is written in Deterrence Era, Year 60 (DE 60), with Trisolaris in existence, its fleet inbound and the partial suspension of sophon interference declared but unverified (United Nations Planetary Defense Council, Secretariat, 2 DE). All runs were taken between Year 51 and Year 59. Section 2 defines the two arms and the forger model, Section 3 reports the comparison, and Sections 4 and 5 state what the null result covers and what it leaves out.
2. Methods
The source is the run register of the Institute stations in the Beijing basin, which we assembled from the apparatus record and the sealed manifests (Reconstituted Chinese Academy of Sciences, Institute of Physics, 60 DE). The register lists 40 three-photon Mermin runs between Deterrence Era, Year 51 and Year 59. We excluded the nine runs of our earlier filter study, whose stratified schedule was lodged inside the seal and so fits neither arm, and three runs in which an operator overrode part of a written schedule. That leaves 28 runs.
A run belongs to the scheduled arm if its sequence of settings was written into the register before the run and could be read there. The lead time from filing to the run was 6 to 41 days, with a median of 19 days. A run belongs to the run-time arm if the operator at each station chose the setting for each block of 200 events at the console, and the setting was recorded only when entered there, a median of 1.9 s (range 0.4 s to 11 s) before the first detection of the block. In that arm the choice existed before entry only as a decision in the operator's head, which no sophon can read. We take both definitions from the register's own protocol field. The console enforced a balance rule in the run-time arm, refusing any entry that would push a setting group over its quarter of the events in the run, and this is why the four groups hold equal numbers of events. Each written schedule likewise allotted a quarter of the run's events to each setting group, which gives the same equality in the scheduled arm.
Each station logs an outcome of +1 or -1 for a polarisation measurement in the X or Y basis. We use the Mermin form for three parties, with M = E(XXX) - E(XYY) - E(YXY) - E(YYX), where E is the mean product of the three outcomes in a setting group. No local model gives M above 2, and an ideal three-photon entangled state reaches 4. The scheduled arm holds 103,200 events in each of the four groups and the run-time arm holds 79,200. The standard error of each correlator follows from its binomial variance, and because the four groups are disjoint, the variance of M is the sum of the four. Intervals are the estimate plus or minus 1.96 standard errors, and with events in the tens of thousands per group a normal approximation is adequate. No filters, amplitude cuts or time masks were applied, so every figure below refers to the full records.
We state the forger model before the data. A forger limited by advance knowledge replaces a share of the events in each setting group with fabricated events whose product it chooses, and we hold the unforged events to a local model, so that the forged share sets how far M can rise above 2. On a scheduled run it knows the group from the written schedule and gives every fabricated event the violating sign. On a run-time run it does not know the group. The best it can then do is to give every fabricated event a product of -1, the violating sign in three of the four groups, and the fabricated events then add exactly the local bound and nothing above it. If a share f of the excess M - 2 in the scheduled arm were forged in this way, the run-time arm would show an excess lower by f (M<sub>S</sub> - 2), where M<sub>S</sub> is the scheduled value. A forger that instead guesses the group uniformly does worse, since its fabricated events carry the correct sign in five cases of eight, and the predicted difference rises to 1.5 f (M<sub>S</sub> - 2). The prediction is therefore a positive difference between the arms, scheduled minus run-time, of at least f (M<sub>S</sub> - 2).
The contrast between arms is the only contrast we pre-specified, so no correction for multiplicity applies. We apply a z test of the difference against zero, and a pair of one-sided tests against a margin of 0.05 in M. The margin was chosen because it corresponds, under the model above, to a forged share of 4.1% of the excess for the best blind forger, and we regard a smaller share as immaterial. The run-to-run standard deviation of M within each arm is compared with the value expected from counting statistics at the mean run size.
3. Results
Across 16 scheduled runs the register holds 412,800 triple events and the 12 run-time runs held 316,800. Scheduled runs averaged 25,800 events and run-time runs 26,400. Table 1 gives the Mermin value for each arm and for their difference, and Table 2 gives the correlators.
Scheduled runs gave M = 3.226 (95% CI 3.219 to 3.233) and run-time runs gave M = 3.230 (95% CI 3.222 to 3.239). Both lie far above the local bound of 2. The difference, scheduled minus run-time, was -0.004 with a standard error of 0.0056 (95% CI -0.015 to 0.007), and the z test gave z = -0.77 and p = 0.44. The pair of one-sided tests against the margin of 0.05 both rejected, with p below 0.001, so the difference lies within 0.05 of zero. The arms are not distinguished, and the point estimate is slightly in the wrong direction for the prediction.
The forger model predicts a difference of 1.226 if the whole excess of 1.226 in the scheduled arm were forged by the best blind forger, and 1.839 if it were forged by one that guesses uniformly. The first would leave the run-time arm at the local bound of 2 and the second at 1.387, while the observed value is 3.230, which is 294 standard errors above the first. The one-sided 95% upper limit on the difference is 0.0049, and dividing by M<sub>S</sub> - 2 gives an upper limit of 0.0040 on the forged share of the excess, or 0.4%, against 0.27% for the uniform guesser. That limit is conditional on the model, and it holds only for a forger with no knowledge of the setting before entry.
Run-to-run scatter agrees with counting statistics. The standard deviation of M across runs was 0.016 in the scheduled arm against 0.015 expected, a variance ratio of 1.20, and 0.015 in the run-time arm against 0.015, a ratio of 1.05. Neither arm shows dispersion that a forger acting in bursts would add.
4. Discussion
The prediction failed. Scheduled runs did not show a smaller violation than run-time runs, and the interval on the difference excludes the 1.226 that a fully forging blind forger would produce. We do not offer a reading of the small difference of -0.004 in either direction, since it lies inside the interval for zero.
It may be objected that the comparison can never have found anything, because sophons read all that is written and a run-time setting is written at the console at entry. The objection is correct about one forger and wrong about the rest. Fan showed that the console entry precedes the first detection of the block, and in our register the lead was 0.4 s to 11 s (Section 2). Qin and Tao placed no measurable bound on the relay latency of paired sophons, so a forger that reads the entry at once and acts at once has all the time it needs. That forger is not bounded by this comparison. What the comparison bounds is a forger that needs a setting to be known before the run, whether by schedule or by any other pre-run route. Our data say that if such a forger was present, it supplied at most 0.4% of the violation excess in the model of Section 2.
Hence the result is narrow, because the phrase run-time setting means a setting entered at the console seconds before detection, and nothing in the register records it earlier than that. A setting decided in the head and entered into a console is invisible to a sophon until the entry is made, and from that moment it is as readable as any schedule filed in the register weeks earlier, so that the only protection a run-time choice offers is the interval between the decision and the keystroke. The arms differ in the information available before entry and not in the information available after.
We make no claim about the suspension. As ecumenicals we write within both the Luo-Ji / Ye framework and the Lurker framework, holding that dark-forest reasoning describes how civilisations that deploy sophons behave while the physics a sophon employs answers to a structure that this reasoning does not address. The null result does not choose between these positions and was not designed to do so.
5. Limitations
Arms were not randomised. The register shows scheduled runs concentrated in the earlier years and run-time runs in the later ones, so drift in the source and the detectors is confounded with the arm. The agreement of the Mermin values limits the drift that could have acted, but it cannot separate a drift from a forger that cancelled it.
The forger model is ours. It assumes that a forger without the setting gives every fabricated event a product of -1, the best a blind forger can do. A forger that learned something of the setting before entry, for example by exploiting the balance rule of the console once a group had filled, would show a smaller predicted difference and our limit on the forged share would loosen in proportion. A forger that needs no setting at all, because it reproduces the statistics of genuine events, is outside the comparison.
Operators may have spoken their choices aloud or written them on paper before entry, and anything written or spoken is readable. We did not audit this, and any such leak would move a run-time run toward the scheduled arm. The lead times of the scheduled arm are known only from the register's filing dates.
The sample is 28 runs from one site. Whether the same null holds with more parties, a second site or a setting generated by hardware in the instant of detection is untested. We have compared, for the first time, the Deterrence-Era Bell record by the information a forger would have needed, and we found that the information did not matter. That does not show that a forger was absent, and it does not show that none could read a setting at the moment of entry. The question of what a zero-latency reader would leave in a record, and whether any record exists that could expose one, remains unanswered.
References
- Pei, S., and Hao, L. (36 CE). Outcome substitution at distributed detectors, a formal model of interference in separated-station experiments. Journal of Dimensional Physics, 11, 201–224.
- Qin, T., and Tao, R. (152 CE). Terrestrial baselines and the relay latency of paired sophons, absence of a measurable lower bound. Journal of Dimensional Physics, 30, 5–29.
- Rong, H., and Cen, Y. (12 DE). Three-station Bell tests and the separation of intervention from nonlocality. Journal of Dimensional Physics, 41, 310–338.
- Fan, Y. (28 DE). Setting entry and detection time in archived Bell-test records, a survey of console logging practice. Journal of Dimensional Physics, 46, 60–88.
- United Nations Planetary Defense Council, Secretariat (2 DE). Record of the Trisolaran declaration of partial suspension of sophon activity, as received and logged under the Dark Forest Deterrence. Bulletin of the UN Planetary Defense Council, declassified series, Bulletin 6, annex 3.
- Reconstituted Chinese Academy of Sciences, Institute of Physics (60 DE). Run register and sealed-record manifests, Mermin runs of Deterrence Era Years 51 to 59, Laboratory for Sophon-Shielded Measurement. Reconstituted Chinese Academy of Sciences, Archival Series, Series SSM-4, register entries 1-40.
- Ruolan, H. (2026). A Three-Party Bell Test During the Partial Sophon Suspension: Why the Intervention Loophole Cannot Be Closed by Declaration. Uncited Press.
Cited By
Open in Uncited Press →