Uncited Press › The Three-Body Problem Open the interactive journal →
three-body · Physics & Instrumentation

Device-Independent Key Distribution Under the Sophon Intervention Loophole: Key Rates, Intervention Budgets and the Guarantees That Survive

Dr. Hu Ruolan1
1 Reconstituted Chinese Academy of Sciences, Institute of Physics, Laboratory for Sophon-Shielded Measurement
Received 17 Sep 2026 · Revised 13 Oct 2026 · Accepted 25 Oct 2026

Abstract

Device-independent key distribution takes its security from a certified Bell violation. Under the sophon intervention loophole that certificate can be forged event by event, and a sophon also reads any key material at the moment it is recorded. The question is what rate of key survives these two facts. We derive a key rate as a function of an intervention budget, the share of all events that a setting-aware sophon may alter, using a bound on the local weight of the unaltered behaviour and a best-response allocation. The model is applied to archived three-station Mermin runs of the Institute, and no new run is reported. For the sealed record of 206,400 triple events the rate at zero intervention is 0.078 key bits per key-setting event, about 4,000 bits from 51,600 key-setting events, and it reaches zero at a budget of 0.027 of all events, one eighth of the 0.216 needed to explain the violation by forging. Three archived records and the least favourable quarter of one give budgets between 0.020 and 0.027. The best response alters only the key setting, a setting-blind forger gains nothing by construction, and secrecy against an adversary that reads the record is zero at any rate. For the first time, a key rate under the intervention loophole is stated against a budget. The rate need not be an upper limit, because the model grants the remainder chance-level unpredictability but also uses cautious bounds elsewhere, and no statistic in the archive bounds the budget from above.

1. Introduction

Our approach to key rates under the so-called intervention loophole is to fix the share of events that a sophon may alter and to let the sophon choose the allocation that harms the key most. We apply it to the archived Mermin runs of the Institute. That is, the key rate becomes a function of a stated budget, and the data decide only where that function reaches zero.

Device-independent key distribution takes its security from a certified Bell violation and from nothing inside the devices. Since the first sophons arrived, every such certificate has been open to the intervention loophole, because a sophon can fix the logged result of any one measurement it reaches (Pei and Hao, 36 CE). Gan then bounded the weight that a local component can carry in a multi-party correlation by its Mermin value (Gan, 118 CE). Paired sophons were next tested across terrestrial baselines, and no delay that could be detected was found, so no station lies beyond the reach of another (Qin and Tao, 152 CE). A third station then let an analyst tell substitution at one station from genuine nonlocality, provided that substitution was on a large scale (Rong and Cen, 12 DE). Lei then built a three-station conference key from the same measurements, on the assumption of honest devices (Lei, 20 DE).

None of these results states what rate of key a station can claim when the loophole is open.

Our sealed-record study certified a Bell violation without relying on the declared suspension, and it bounded the forged share from below only (Hu, 59 DE). Our register study found no loss of violation when settings were readable in advance (Hu, 60 DE). Neither paper asked what the violation is worth to a key.

Sophons read whatever is written, and a key is written when it is recorded, so no protocol can keep a recorded key from a sophon. What we compute is a narrower guarantee, and Section 3 says exactly which. We write in Deterrence Era, Year 60 (DE 60). Trisolaris exists and its fleet is inbound. It has declared sophon interference partly suspended, as relayed by the UN Planetary Defense Council (PDC) and logged in its Secretariat bulletin (PDC Secretariat, 2 DE), and no instrument has verified the declaration. All archived runs predate this account.

2. System Description

A central source sends one photon of a three-photon entangled state to each of three stations in the Beijing basin. Every station measures X or Y polarisation and logs +1 or -1. The Mermin value M is E(XXX) minus the sum of E(XYY), E(YXY) and E(YYX), with E the average product of the three outcomes within one setting group. A local model cannot push M past 2, and settings followed a stratified schedule lodged in the seal of the sealed record (Hu, 59 DE). In the register runs they were scheduled or entered at the console (Hu, 60 DE). Every group therefore holds the same number of events.

The key is the outcome of station 1 in the XXX group, the key setting, and the other groups serve only to estimate M. In the ideal state the product of the three outcomes in the key setting is +1, and each single outcome is uniform and independent of any one other. Station 3 announces its outcome, which alone says nothing about the outcome of station 1, and station 2 recovers that outcome as the product of its own and the announced one, with error rate Q = (1 - E(XXX))/2. The key is therefore shared by stations 1 and 2, with station 3 as helper. Lei built the three-way variant (Lei, 20 DE).

After the detections the stations announce their settings, estimate M and Q from the sealed record, and exchange an error-correction message and a privacy-amplification seed. Each announcement, that of station 3 included, is written, so a sophon reads it at once. Error correction costs h(Q) bits per key bit, with h the binary entropy, and we charge that cost in full.

The adversary is the sophon, and we credit it with exactly the powers that follow. It reads anything written, printed, spoken or recorded, and it cannot read thought. It can fix the logged result of any one measurement, and it coordinates with no measurable latency. In the run-time arm of the register a setting reached the console with a median lead of 1.9 s over the detections it governs, and scheduled settings were readable 6 to 41 days ahead (Hu, 60 DE). Console logging precedes detection in archived records generally (Fan, 28 DE). We therefore give the sophon the setting of every event it alters.

Our inputs are the published correlators of two archives, the nine sealed runs (Institute of Physics, 59 DE) and the run register of the readability study (Institute of Physics, 60 DE). Neither archive contains a key or any record of key extraction. We assume that the classical channel between the stations is authentic.

3. Analysis and Model

Write ε<sub>k</sub> for the fraction of events altered within setting group k. The budget ε is their mean, which is the altered fraction of all events because the groups are equal. An altered event carries the product that raises M, +1 in the key setting and -1 in the other three, and the sophon knows its key bit. Write the observed correlator of group k as |E<sub>k</sub>| = (1 - ε<sub>k</sub>) |E<sub>k</sub>'| + ε<sub>k</sub>, where the primed value belongs to the unaltered events. Hence |E<sub>k</sub>'| = (|E<sub>k</sub>| - ε<sub>k</sub>)/(1 - ε<sub>k</sub>), and the unaltered events have the Mermin value M' equal to the sum of the four. Before inverting we lower each observed correlator by 1.645 standard errors and raise Q to its one-sided 95% limit, using a normal approximation that is adequate at 12,900 events or more per group.

Suppose the unaltered behaviour splits into a local part of weight p and a remainder. The remainder cannot exceed 4 in M, the largest value any no-signalling behaviour reaches, and the local part cannot exceed 2, so M' is at most 2p + 4(1 - p) and p is at most (4 - M')/2 (Gan, 118 CE). We then make one assumption of our own, stated given the announcement of station 3, that the local part is wholly predictable to the sophon and the remainder gives a key bit that it guesses no better than chance. With ε<sub>x</sub> the altered share of the key setting, its guessing probability is at most P<sub>g</sub> = ε<sub>x</sub> + (1 - ε<sub>x</sub>)(1/2 + p/2), and the key rate per key-setting event is r = max(0, -log<sub>2</sub> P<sub>g</sub> - h(Q)), after the reconciliation cost of Lei (20 DE), and the rate per triple event is r/4. It is asymptotic, with no finite-size term for privacy amplification.

We searched all allocations of the budget over the four groups in steps of one sixteenth of the budget, and the rate was lowest when the whole budget went to the key setting, so that ε<sub>x</sub> = 4ε, and this allocation is the best response. Each altered key bit is known to the sophon and also lowers the corrected key-setting correlator, which raises p.

If a setting-blind forger alters the same share ε in every group and sets every product to -1, the corrected M' becomes (Σ|E<sub>k</sub>| - 2ε)/(1 - ε), and P<sub>g</sub> works out to 3/2 - Σ|E<sub>k</sub>|/4 at every ε. This holds by construction, because such a forger adds exactly the local bound of 2, and we do not offer it as an empirical finding.

Two thresholds follow. The budget ε<sub>max</sub> is the smallest ε at which the best response drives r to zero. The budget ε<sub>cert</sub> is the smallest ε at which an intervention on one group, with the unaltered events local, reproduces all four observed correlators, so that M' falls to 2. Below ε<sub>cert</sub> the violation is genuine, and below the smaller ε<sub>max</sub> it also pays for a key, as the next section shows.

A sophon that reads the record after writing has P<sub>g</sub> = 1 for that record at any M, so the secrecy of a recorded key against such a reader is zero. This is structural. The rate certifies secrecy against an adversary that reads settings and announcements and can alter outcomes, but cannot read the recorded outcomes of unaltered events. A second guarantee holds without that restriction. Up to the budget, no party chose the key bits before the measurement, so the key can serve as a nonce.

Partial suspension enters through the run time. Suppose interference is suspended in a share w of it. Because every block of 16 events in the sealed record held each used setting group four times (Hu, 59 DE), and we assume the same spread for the other records, a sophon confined to the remaining windows can alter at most a share 1 - w of each setting group, provided its windows are longer than a block, so it cannot concentrate its budget as the best response does. Any w above 1 - ε<sub>max</sub> therefore guarantees a positive rate, whatever the sophon does inside its windows. If it alters every event in them with the violating sign, the rate reaches zero at a smaller suspended share, which Section 4 gives. The location of the windows is not needed.

4. Validation Against Field Data

The sealed-record study yielded 206,400 triple events with 51,600 in each setting group. The four correlators were 0.813 in the key setting against -0.813, -0.816 and -0.811 elsewhere, and the published value of M was 3.254 with a 95% interval of 3.244 to 3.264 (Hu, 59 DE). The tabulated correlators sum to 3.253, rounding aside, and we compute from them. The key-setting error rate is Q = 0.0935 with a one-sided upper limit of 0.0956, and the lower limit on M is 3.236.

Table 1 shows the rate against the budget. At zero intervention P<sub>g</sub> is 0.691 and the key rate is 0.0784 per key-setting event, or 0.0196 per triple event. The 51,600 key-setting events would give about 4,045 key bits, though no key was drawn.

The rate falls by about 0.003 for each 0.001 of budget and reaches zero at ε<sub>max</sub> = 0.027, which is 5,580 of the 206,400 events. The budget that would explain the whole violation by forging, ε<sub>cert</sub>, is 0.216. An intervention one eighth the size of the one that could fake the violation therefore already removes the key. The earlier paper bounded the forged share by M alone and gave 0.15 of all events or more for forging confined to one setting. Taking the four correlators together raises that figure, because the unaltered events must match each correlator and not only their sum. Forging confined to one group is the cheapest allocation, so 0.216 is the smallest budget that explains the violation by forging, and the like-for-like figure for the least favourable quarter is 0.215. Within this model the record therefore excludes the forger of about 0.16 of all events that the earlier paper could not exclude. That forger was built to reproduce M and not the four correlators, and matching all four correlators takes at least 0.216.

Table 2 applies the model to three archived records and to the least favourable quarter of the sealed record, the worst of 60 filters in the earlier study and so a cautious choice. That quarter has 12,900 events per group, and its correlators were not tabulated, so we set each equal to M/4. The two arms use the register correlators. The quarter's rate at zero intervention is 0.058 against 0.078 for the pooled record, mainly because the sample is a quarter of the size. For partial suspension, a sophon that alters every event in its windows removes the key when it is active in more than 7.7% of the run time for the sealed record and 5.7% to 6.6% for the other three, so w must exceed 0.923 for the sealed record and 0.934 to 0.943 for the others. A sophon free to choose which events to alter within a budget needs only 2.0% to 2.7% of all events, and w above 0.973 to 0.980 guarantees a positive rate whatever it does.

This section shows what values the inputs take. It does not test the assumptions of Section 3.

5. Failure Modes

First, a sophon reads the record at the moment of writing, so against that reader the key was never secret, whatever the rate says. A key that must stay unknown to Trisolaris after it is stored cannot be made here.

Second, the budget cannot be seen in any statistic of the run. A run with no alteration and a run with a share ε altered in the key setting can show identical statistics, because the best response leaves the observed correlators where they are. The data bound ε from below through ε<sub>cert</sub>, as in the earlier work, and never from above. The rate is positive only if a sophon altered fewer than about 2.0% to 2.7% of events, and nothing in the archive says so. A verified declaration of partial suspension would say so, and none is verified (Xiang and Tong, 33 DE).

A third failure lies in the guessing bound that we assumed. We granted the remainder chance-level unpredictability, the most generous assumption available to the key. If the remainder is partly predictable, the true rate is lower and the budget at zero is smaller than Table 2 shows. However, the maximal local weight and the min-entropy bound are cautious choices, so the net direction is not fixed.

Authentication is the next failure. The stations must be sure that announcements come from each other, and a classical authentication key shared in advance is a written string that a sophon has already read. We gave the sophon no power over classical messages, and that is our choice and not a property of the sophon.

Time structure is the last failure that we can name. A budget spent in bursts is invisible to an estimate pooled over the run, and the block-wise estimate that exposes it costs rate. Because we give the sophon every setting, the rate does not depend on when a setting became readable.

6. Conclusion

Under the stated model, the archived Mermin runs likely support a rate of at most 0.058 to 0.078 key bits per key-setting event, reached only with no intervention and falling to zero once the sophon altered 2.0% to 2.7% of the events, and only against an adversary with no access to the record after it is written. The access limit is permanent. The budget cannot be tested with the statistics that the archive holds.

The tolerable budget may be larger than we report, because a full entropic treatment might admit more intervention. We think this less likely than the reverse, because a no-signalling remainder that reaches M = 4 can leave the outcome of station 1 fully predictable. However, the maximal local weight and the min-entropy bound are cautious choices, so a full treatment could move the budget either way.

As ecumenicals we argue for the compatibility of the Luo-Ji / Ye and Lurker frameworks, and our model uses only the read access and the power over single outcomes, and so needs neither framework to be right.

For the first time, a key rate under the intervention loophole is stated against a budget and not against a declaration. It assumes a budget that no archive can measure, a chance-level guess for the remainder, one site with three stations, asymptotic rates and an authentic classical channel. Whether a longer baseline, a fourth station or a bound on the number of altered events from outside the statistics would change the figures remains open, and the method does not decide whether any sophon altered a single event.

device-independent key distributionintervention loopholeMermin inequalitysophonbounded-intervention modelkey ratepartial suspension

References

  1. 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.
  2. 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.
  3. Gan, B. (118 CE). The local weight of a multi-party correlation bounded by its Mermin value. Journal of Dimensional Physics, 26, 140–166.
  4. 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.
  5. Lei, X. (20 DE). A three-station conference key from Mermin measurements on an entangled state, rates under honest devices. Journal of Dimensional Physics, 43, 120–149.
  6. 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.
  7. Xiang, J., and Tong, B. (33 DE). What instruments could confirm sophon quiescence, an assessment of verification options. Deterrence Era Strategic Studies, 9(2), 140–171.
  8. 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.
  9. Reconstituted Chinese Academy of Sciences, Institute of Physics (59 DE). Apparatus record and sealed-record manifests, runs 1 to 9, Laboratory for Sophon-Shielded Measurement. Reconstituted Chinese Academy of Sciences, Archival Series, Series SSM-3, manifests 1–9.
  10. 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.
  11. Ruolan, H. (2026). A Three-Party Bell Test During the Partial Sophon Suspension: Why the Intervention Loophole Cannot Be Closed by Declaration. Uncited Press.
  12. Ruolan, H. (2026). No Difference in the Mermin Violation Between Scheduled and Run-Time Measurement Settings in Deterrence-Era Bell Runs. Uncited Press.
Read this article inside the full journal experience — browse by faculty, search across universes, and filter by discipline.
Open in Uncited Press →