Phase 79 · Peer-review readiness

Stage-1 Package

A Registered-Report-shaped dossier for a study that has not been reviewed, registered, built or run. Everything below is fixed before data exists, so that no result can be rescued after the fact.

internal red team: ACTIVE / SUBSTANTIALexternal technical review: NOT STARTEDformal peer review: NOT STARTEDpreregistration: NOT FILEDhardware: NOT BUILTphysical evidence: NONE

Section 1

Research question

In a programmable locally-coupled oscillator apparatus, can genuinely new information — a payload generated by a hardware TRNG only after arming — be decoded at endpoint B earlier than the conservative empirical lower bound of every inventoried physical causal path from endpoint A?

The expected answer is no. The study is designed so that 'no' is publishable, informative and fully specified in advance.

Section 2

Competing models M0–M5

Every model that could produce the data, with our own hypothesis ranked last on prior plausibility.

prior high

Ordinary local finite-speed physics (null)

T_decode is consistent with D_tau at every inserted delay setting. Δ ≈ 0 within the uncertainty budget.

prior moderate

Latent-connectivity / architecture effect

T_decode beats the intended logical route C0 but never beats D_tau. Margin does not scale with inserted delay.

prior moderate

Instrumentation artefact

A roughly constant apparent margin, independent of inserted delay, and disagreement between the two timing stacks.

prior moderate

Hidden conventional shortcut

Margin tracks the bypass delay and is caught by the channel-inventory audit and the B3 positive controls.

prior moderate

Conventional nonlinear mimic

Target-like latency and sparsity scaling produced by ordinary local nonlinearity, caching or scheduling; fails the mechanism audit.

prior extremely low

Causal-bound anomaly (Adjacency Theory hypothesis)

Δ > 0 by the preregistered guard band, preserved or growing across delay settings, surviving every control.

Section 3

Hypotheses

H0primary

T_decode(B) ≥ D_tau(A,B) − ε_cal at every delay setting.

Supports M0/M1. This is the expected outcome and is accepted, not merely 'not rejected'.

H1primary

Δ = D_tau,lower − T_decode,upper ≥ G at every delay setting, where G is the fixed guard band.

Only outcome that opens a causal-bound anomaly candidate. Requires all quality gates and all controls to pass.

H2secondary

T_decode < T_C0 while T_decode ≥ D_tau − ε_cal.

Latent-connectivity / architecture effect. Interesting engineering, no causal claim.

H3secondary

Observed margin is independent of inserted calibrated delay.

Fixed instrumentation offset (M2). Auto-classifies as artefact regardless of magnitude.

H4secondary

Observed margin scales with a specific port, node ID or wiring instance.

Topology dependence (M3/M4). Invalidates the run set.

Section 4

Apparatus diagram

Placeholder — schematic not frozen

ENDPOINT A→TRNG payload→timestamp A→16-node oscillator substrate→calibrated delay bank→timestamp B→ENDPOINT B decode

Independent timing stacks at A and B, offline blinded DAQ, RF and power isolation, environmental sensing, and a hidden-shortcut injection fixture sit around this chain. The frozen wiring diagram is a B0 deliverable and does not exist yet.

Section 5

Stage gates B0–B8

Each gate has entry criteria, a test, a pass condition and a stated failure response. Gates run in order and none may be skipped.

B0Clock loopback + uncertainty budgetnot run

Entry: Both timing stacks powered, calibration standards in place.
Test: Loopback timestamping of a known path with both stacks, across temperature.
Pass: Measured σ_total established and both stacks agree within their combined uncertainty.
On failure: Programme halts. No later phase runs on an unmeasured budget.

B1Post-arming entropy and decodenot run

Entry: B0 passed. TRNG installed at A.
Test: Payload generated strictly after arming; hashed at creation; decoded at B over a known route.
Pass: Entropy health logged and ≥99% payload fidelity over the baseline route.
On failure: Fix entropy chain or decoder before any timing claim.

B2Delay bank calibration and driftnot run

Entry: B1 passed.
Test: Bidirectional group-delay characterisation of every delay segment, repeated over temperature and time.
Pass: Per-segment delay known to within budget and drift bounded over a run window.
On failure: Re-characterise or replace segments; extend guard band only before freeze.

B3Hidden-shortcut detection challengenot run

Entry: B2 passed. Custodian installs secret bypasses.
Test: Blind set of runs, some containing a deliberately concealed conventional bypass.
Pass: 100% detection. Anything below 100% is a hard stop.
On failure: Channel inventory is incomplete by demonstration. No confirmatory run is permitted.

B4Ordinary C0 propagation baselinenot run

Entry: B3 passed.
Test: Measure conventional route latency under strongest permitted signalling.
Pass: Stable, reproducible T_C0 with characterised spread.
On failure: Diagnose apparatus instability before proceeding.

B5Nonlinear / moving-front analoguenot run

Entry: B4 passed.
Test: Run the oscillator substrate dynamics with the three-ruler mechanism audit active.
Pass: Dynamics reproducible and auditable. No causal claim is made at this gate.
On failure: Substrate redesign; verification clock does not advance.

B6Blinded mixed controlsnot run

Entry: B5 passed. Sealed schedule held outside the run team.
Test: Interleaved null, sham, cable-swap and bypass conditions under blinding.
Pass: Analysis pipeline classifies every control condition correctly while blind.
On failure: Pipeline or blinding repaired, then B6 repeats in full.

B7Frozen confirmatory causal-bound campaignnot run

Entry: B0–B6 all passed. Protocol, firmware, topology and analysis hashes lodged.
Test: The preregistered run counts, executed once, against the frozen decision rule.
Pass: The rule returns its verdict. Any verdict is a result.
On failure: Deviation logged and published. No re-run under a revised rule.

B8Independently assembled second apparatusnot run

Entry: B7 complete with any outcome.
Test: A second rig built from the specification by different hands reproduces the campaign.
Pass: Preregistered result reproduces, or does not, and both are published.
On failure: Single-apparatus results are downgraded, not defended.

Section 6

Causal-channel inventory

Every physical route by which influence could travel from A to B, and how each will be calibrated.

	Channel	Calibration method	Risk
E1	Intended optical / fibre delay routes	Bidirectional group delay per spool, OTDR continuity, thermal sweep.	low
E2	Analogue coupling wires between oscillator nodes	Pairwise pulse-response earliest onset.	medium
E3	Digital FPGA IO and control links	Per-pin loopback with firmware frozen.	high
E4	Clock distribution and any shared reference	Explicit inventory; independence demonstrated, never assumed.	high
E5	Power rail and ground coupling	Isolated / battery domains where possible; coupling characterised where not.	high
E6	RF, capacitive and inductive leakage	Shielded enclosure comparison, spectrum scans, radios disabled.	high
E7	DAQ, storage and analysis host paths	Offline and blinded; no live path between endpoints during a run.	medium
E8	Optical leakage and stray line-of-sight	Dark-box checks and photodiode scans.	medium

Section 7

Primary statistic, guard band and uncertainty budget

Δ = D_τ,lower(A,B) − T_decode,upper(B)

The single preregistered statistic. Nothing else decides the outcome.

D_tau,lower is the minimum over all inventoried physical paths of the summed per-edge earliest influence delays, each taken at its conservative lower confidence limit.

T_decode,upper is the upper confidence limit on the earliest time B decodes the post-arming payload above the preregistered bit-error threshold. First physical influence and first valid decode are recorded separately; only decode enters the statistic.

A positive Δ that does not exceed the guard band is reported as null. There is no 'suggestive' category.

Fixed guard band

G = max(10 · σ_total, 2 ns engineering floor)

G is fixed at protocol freeze and cannot be revised after any data exists, in either direction.

The delay-amplified design puts µs-scale delay on every permitted route, so a ns-scale guard band is a vanishing fraction of the bound rather than a knife-edge.

Uncertainty budget — design only

source trigger jitter	15 ps	Arming-to-emission latency spread at A.
timestamp A	25 ps	TDC quantisation plus differential nonlinearity.
timestamp B	25 ps	Independent stack, different calibration model.
cable / fibre group-delay model	20 ps	Residual after bidirectional characterisation.
FPGA IO calibration	30 ps	Dominant term; must be measured at B0.
environmental drift	20 ps	Thermal and supply drift between calibration and run.
σ_total (quadrature)	≈ 56 ps	Assumed, not measured.

Section 8

Sampling, exclusions and stop rules

Sample size and power

Delay settings: ≥4 inserted calibrated delays spanning at least one decade.
Runs per condition: ≥20 valid runs per condition per delay setting, fixed before collection.
Conditions: Target, null, sham, cable-swap, hidden-bypass positive control.
Stopping rule: Collection stops at the preregistered count. No optional stopping, no extension on an interesting trend.
Power basis: Guard band G against measured σ_total from B0. The full calculation requires a metrologist and is not yet complete.

Exclusion criteria

Calibration certificate expired at run time.
Configuration, firmware or topology hash mismatch against the frozen record.
The two timing stacks disagree beyond their combined uncertainty.
Entropy health check failed or payload generated before arming.
Environmental sensor out of the preregistered window.
Raw-data integrity checksum failure.

Every exclusion criterion is outcome-neutral: it can be evaluated while blind, and none references the value of Δ. Excluded runs are published alongside valid runs with their exclusion reason.

Stop rules

B3 hidden-shortcut detection below 100% — hard stop, no confirmatory campaign.
Any prohibited controller read or write during a run — the entire run set is invalidated, not the sample.
Timing-stack disagreement persisting across a session — stop and recalibrate from B0.
Blind key exposure to any member of the run or analysis team — campaign restarts with a new schedule.
Any change to frozen firmware, topology or analysis code — the freeze is broken and the campaign is exploratory from that point.

Section 9

Blinding, randomisation and controls

Blinding

Sealed condition schedule generated and held by a custodian outside the run team.
Analysts blind to condition labels until the frozen analysis has been executed and its output hashed.
Destination decoder blind to the source bits.
Fake timing-offset injection used to verify that analysts cannot infer condition from the data.

Randomisation

Condition order randomised within the sealed schedule; controls interleaved, never batched at the end.
Physical wiring randomised after firmware freeze.
Topology relabelling and endpoint remapping between run sets so no lookup can key off fixed node IDs.
Independently generated topologies loaded post-freeze; behaviour must transfer with no retuning.

Controls — null, sham and hidden-shortcut positive

K1	Hidden-shortcut positive control	Concealed conventional bypasses installed by the custodian; audit must detect 100%.
K2	Null condition	Apparatus armed, no payload emitted. Any decode is a pipeline failure.
K3	Sham condition	Payload emitted into a disconnected route. Decode must not occur.
K4	Cable-swap equivalence	Long route replaced by an equivalent short route; classification must change exactly as predicted.
K5	Shuffled payload	Decoder run against a shuffled payload; must fail the bit-error threshold.
K6	Source disconnection	Physical disconnection at A mid-set; all downstream decode must cease.
K7	Shielding comparison	Runs with and without RF enclosure and absorber; margin must not track shielding state.
K8	Sham firmware	Adaptive caching and centralised scheduling firmware loaded as positive controls; mechanism audit must reject them.
K9	Delay-scaling artefact test	Margin plotted against inserted delay; a flat margin classifies as artefact automatically.

Section 10

Data, code, hashes and deviations

Raw-data policy

Raw unsmoothed timestamps and waveforms published for every run, valid or excluded.
Analysis code published with the hash lodged before unblinding.
Firmware, topology, wiring map and calibration certificate IDs published with each run manifest.
Lab log published, including the entries that record mistakes.
No arrival metric may depend on smoothing, windowing or any parameter chosen after data exists.

Code and firmware hashes

Protocol document	At freeze, before B0	PENDING
Analysis code	Before any unblinding	PENDING
Endpoint firmware (A and B)	Before wiring randomisation	PENDING
Topology / wiring map	Per run set	PENDING
Raw dataset	At acquisition close	PENDING
Sealed blind schedule	At generation, held by custodian	PENDING

Section 11

Replication plan and claim ladder

R1 — same rig, new operators

Removes operator-specific procedure effects.

R2 — independently rebuilt timing stack

Removes shared calibration-model error.

R3 — second apparatus, same specification

Removes apparatus-specific systematics.

R4 — external laboratory, independent team

The only rung that permits language beyond 'candidate'.

C0	Synthetic / exploratory result	Statements about our own code. Nothing about nature. Everything published so far sits here.
C1	Apparatus characterisation	Statements about this rig's measured behaviour.
C2	Architecture / latent-connectivity effect	Statements about routing efficiency. Explicitly not causal.
C3	Causal-bound anomaly candidate (single apparatus)	A request for replication. No physical interpretation.
C4	Replicated causal-bound anomaly	A statement that something is unexplained. Still not a mechanism, still not FTL.
C5	New physics	Requires an independent theory, external confirmation and exclusion of every systematic. Not reachable by this programme alone.

Section 12

OSF preregistration checklist

Rigorous preregistration requires every decision to exist before the data does.

NOT YET REGISTERED10 of 12 items drafted

P1	Hypotheses stated explicitly and directionally	H0–H4 written with quantitative thresholds.	drafted
P2	All variables defined operationally	Δ, D_tau, T_decode, fidelity, active fraction and audit outcome are defined with measurement procedures.	drafted
P3	Every decision made before data exists	Guard band, run counts, exclusions and decision rule fixed at freeze.	drafted
P4	If/then contingencies for each possible outcome	The classification ladder maps every outcome branch in advance.	drafted
P5	Statistical tests and decision criteria named	Frozen decision rule with zero free parameters; slope and endpoint thresholds fixed.	drafted
P6	Exclusion criteria stated and outcome-neutral	Six criteria, all evaluable while blind.	drafted
P7	Model forms and covariates specified	Regression form across log2 N and delay settings specified; covariates limited to temperature and supply.	drafted
P8	Commitment to report all outcomes	Publication of valid and excluded runs regardless of result.	drafted
P9	Planned exploratory analyses labelled as such	Exploratory analyses enumerated and firewalled from the confirmatory rule.	drafted
P10	Sample size and power justification	Run counts drafted; the formal power calculation needs a metrologist and measured σ_total.	not ready
P11	Deviation reporting policy	Written and published.	drafted
P12	Registration lodged, timestamped and public	Not filed. Nothing above counts until this is done.	not ready

Section 13

Reviewer attack log

Written against ourselves, using role IDs because no real reviewer exists yet.

attacks logged

critical

open

closed

need hardware to close

X01R-MET-1 · MetrologycriticalOPENhardware required

The causal-path inventory is incomplete, so D_tau is too slow and any advance is an unlisted wire.

Owner: Metrology lead (unfilled)
Evidence required to close: Per-channel empirical calibration on built hardware plus 100% detection in the B3 bypass challenge.

X02R-EXP-1 · Experimental physicscriticalMITIGATED BY DESIGNhardware required

Entropy leaked before t0 — the payload was knowable to B before arming.

Owner: FPGA lead (unfilled)
Evidence required to close: Arming-order traces, TRNG health logs and payload hash timestamps from B1.

X03R-MET-2 · MetrologycriticalMITIGATED BY DESIGNhardware required

Common-mode clock error makes both stacks wrong in the same direction.

Owner: Timing lead (unfilled)
Evidence required to close: Demonstrated absence of a shared reference chain plus two calibration models that disagree by construction.

X04R-RF-1 · RF / EMIcriticalOPENhardware required

RF, ground or power-rail leakage carried the payload outside the inventory.

Owner: RF engineer (unfilled)
Evidence required to close: Shielded/unshielded comparison, spectrum scans, isolated supply runs, all showing no margin dependence.

X05R-NLD-1 · Nonlinear dynamicscriticalOPEN

An ordinary local nonlinear mechanism reproduces the behaviour; uniqueness is unproven.

Owner: Nonlinear dynamics lead (unfilled)
Evidence required to close: A mechanism-discrimination experiment, not an argument. None currently exists.

X06R-FPGA-1 · FPGA / embeddedcriticalMITIGATED BY DESIGNhardware required

A hidden path inside the FPGA fabric or its IO carries state between endpoints.

Owner: FPGA lead (unfilled)
Evidence required to close: Frozen bitstream hash, per-pin loopback calibration and independent netlist review.

X07R-STAT-1 · StatisticsmajorMITIGATED BY DESIGN

Analysis flexibility: window, threshold or smoothing chosen after seeing data.

Owner: Independent statistician (unfilled)
Evidence required to close: Hash of the analysis code lodged before unblinding; zero free parameters in the rule.

X08R-NLD-2 · Nonlinear dynamicsmajorMITIGATED BY DESIGNhardware required

The effect follows specific ports or node IDs rather than the dynamics.

Owner: Experimental lead (unfilled)
Evidence required to close: Post-freeze topology relabelling with behaviour transferring untuned.

X09R-EXP-2 · Experimental physicsmajorOPENhardware required

Scaling beyond N=64 is unverified, so the claimed trend may not exist.

Owner: Architecture lead
Evidence required to close: Substrate runs at larger N, in simulation first and then on hardware.

X10R-SKEP-1 · Adversarial skepticmajorOPEN

Exact-fabric concurrency remains unresolved on your own kill board; the architecture is not ready to be built.

Owner: Architecture lead
Evidence required to close: A concurrency resolution or an explicit statement that the bench test does not depend on it.

X11R-MET-3 · MetrologymajorMITIGATED BY DESIGNhardware required

Calibration drifts between characterisation and run, so D_tau is stale.

Owner: Timing lead (unfilled)
Evidence required to close: Drift bounds measured at B2 and calibration expiry enforced as a stop rule.

X12R-STAT-2 · StatisticsmoderateMITIGATED BY DESIGN

Multiple testing across delay settings and conditions inflates the false-positive rate.

Owner: Independent statistician (unfilled)
Evidence required to close: A single joint decision rule requiring the signature at every delay setting, not a family of tests.

X13R-SKEP-2 · Adversarial skepticmajorMITIGATED BY DESIGNhardware required

The blind was compromised — the run team could infer conditions.

Owner: Blind custodian (unfilled)
Evidence required to close: Custodian outside the team, fake-offset injection and a post-hoc guessing test the team must fail.

X14R-SKEP-3 · Adversarial skepticcriticalOPENhardware required

Replication is not independent — same people, same parts, same assumptions.

Owner: Programme lead
Evidence required to close: R4: an external laboratory reproducing from the specification with no shared personnel.

Section 14

Journal fit

PROCESS OPTION — FIT NOT ASSESSED

Nature Methods (Registered Reports)

Its Registered Report format is aimed at comprehensive comparisons of established, related methods and tools, and it states explicitly that the format is not suitable for method-development papers.

LIKELY NOT A FIT — listed to prevent us assuming otherwise.

Scientific Reports

Operates a two-stage Registered Report format: Stage 1 review of methods and analysis before data collection, Stage 2 review after the study.

POSSIBLE FIT TO INVESTIGATE — not a target guarantee.

Other multidisciplinary or metrology venues

Scope, format eligibility and editorial interest all unknown.

TO BE IDENTIFIED WITH COLLABORATORS.

Section 15

Partner review board

Target 5–7 external reviewers, at least 2 fully independent. Currently 0 filled.

Precision timing / metrology

unfilled

Attack the uncertainty budget, the calibration chain and the independence of the two timing stacks.

Experimental physics

unfilled

Decide whether the measurement can answer the question at all.

Nonlinear dynamics

unfilled

Propose ordinary local mechanisms that reproduce the behaviour.

FPGA / embedded

unfilled

Attack firmware freeze, arming order and every path where a bit could arrive early.

RF / EMI

unfilled

Assume an unlisted coupling path exists and find it.

Statistics / reproducibility

unfilled

Attack power, sampling, exclusions, blinding and remaining analytic freedom.

Adversarial skeptic

unfilled

No stake, no relationship, mandated to recommend rejection if rejection is right.

Reviewer outputs required

A signed and dated written critique.
Objections ranked by severity, using the same scale as the attack log.
An explicit list of protocol changes required before Stage 1 submission.
A final readiness vote: ready / ready with changes / not ready.

At least two reviewers must hold no funding, equity, advisory or employment relationship with the programme. Independence is declared in writing and published with the critique.

Workflow

1Technical brief issuedTwo pages plus the frozen protocol and its hash.
230-minute scientific critique callNo NDA unless the reviewer requires one. Criticism is the deliverable.
3Protocol review windowTwo weeks with full access to models, code and synthetic red-team results.
4Written critique returnedSigned, dated, severity-ranked, published verbatim with our response.
5Revision cycleWe rewrite against the critique, including the parts that hurt.
6Readiness voteRecorded per reviewer. A split vote is published as a split vote.

Section 16

Infrastructure relevance

Potential infrastructure relevance — not an endorsement

The Brookhaven National Laboratory Quantum Network Facility is open to the user community to benchmark performance and validate concepts, and the BNL–Stony Brook prototype uses White Rabbit-supported timing switches with purpose-built classical control across a roughly 70 km fibre span. That combination — long calibrated fibre, distributed sub-nanosecond timing and instrumentation expertise — is exactly the class of infrastructure a delay-amplified causal-bound test needs.

We have not approached this facility and no relationship exists.
Our experiment is not a quantum experiment; the relevance is timing, fibre and metrology expertise.
Listing a facility is not endorsement by that facility, and access would be decided by them on scientific merit.