This manuscript specifies a mid-level, standards-ready controller for networked quantum sensing under a Four-Solves-Three (4S3) contract. Four fixed roles (D, A, B, C) decide three gates—parity hygiene, centered-network co-lock, and corridor severity—without threshold travel and with tokens-only replay. The physical setting is a compact continuous-variable network that enables weighted-sum estimation on the determinant lane (D). I/Q centering (L1/L2) anchors the estimator; z_tension operates as a stop-token budget. Device-independent falsifiers (flip-only phase, disable-L, withhold-node, leave-one-out) are executed at network scale. Heisenberg claims are bound-respecting via a preregistered Fisher ratio band and an identifiability angle token. Two guard-safe routes are contrasted under the same contract: Route A (Magnon Electric Readout, direct D-lane weighted sum) and Route B (Angular Ribbon Mid-Stream Tap, advisory-only centering/stop). A minimal replication kit (frozen graph/weights/mask, detector model, and JSON schemas) allows third parties to reproduce IoU and rho, rerun falsifiers, and verify acceptance from tokens alone.
This manuscript specifies a mid-level, standards-ready controller for networked quantum sensing under a Four-Solves-Three (4S3) contract. Four fixed roles (D, A, B, C) decide three gates—parity hygiene, centered-network co-lock, and corridor severity—without threshold travel and with tokens-only replay. The physical setting is a compact continuous-variable network that enables weighted-sum estimation on the determinant lane (D). I/Q centering (L1/L2) anchors the estimator; z_tension operates as a stop-token budget. Device-independent falsifiers (flip-only phase, disable-L, withhold-node, leave-one-out) are executed at network scale. Heisenberg claims are bound-respecting via a preregistered Fisher ratio band and an identifiability angle token. Two guard-safe routes are contrasted under the same contract: Route A (Magnon Electric Readout, direct D-lane weighted sum) and Route B (Angular Ribbon Mid-Stream Tap, advisory-only centering/stop). A minimal replication kit (frozen graph/weights/mask, detector model, and JSON schemas) allows third parties to reproduce IoU and rho, rerun falsifiers, and verify acceptance from tokens alone.
This is the first in three significant manuscripts, and it establishes a mid-level Four-Solves-Three (4S3) controller for wave platforms, with magnons and L1/L2 geometry as the primary arena and a [[5,1,3]] stabilizer twin for device-independent containment. Decisions are made by four fixed roles closing three independent gates under guard-safe containment: parity hygiene, centered-X balance-line co-lock, and corridor-mapping severity. Acceptance lines are locked (IoU >= 0.75; rho <= -0.60) and never moved. Engine-invariant falsifiers (flip-only Z, disable-L, withhold-carrier reproducibility, and a 3-of-4 stabilizer audit) bound artifacts without threshold travel. Reproducibility is delivered through tokens-only replay with minimal hashed schemas (e.g., window_hash, corridor_hash, grid_spec_hash, kpi_script_hash) and a worked micro-example. A rectifier KPI and a preregistered Tz_star budget provide compact auditing of parity and flatness; a one-tick Future_ur lease gate ensures survival consistency before any directive update. A concise concentric-ladder checkpoint (4->5) is identified for stress testing while guards remain fixed. The result elevates wave-scale demonstrations to a replication-ready standard: same grammar across magnons, L1/L2 geometry, Floquet/SRO mapping, and the stabilizer twin, with leave-one-out redundancy and tokens-only replay enabling independent verification.
This manuscript establishes the L lane as a fully disciplined coordinate layer within the G–Sigma–P–L stack, transforming the ledger from an informal bookkeeping device into a contract-bound, falsifiable decision system. It defines contracts, capsules, observation tokens, and ledgers in a platform-independent manner and enforces that every decision be reproducible from token data alone under fixed guards, with no threshold travel and no hidden state. The work formalizes a three-level falsification hierarchy—replayability, no-threshold-travel, and one-way federation with Four-Solves-Three closure—and extends it with explicitly non-gating metadata and audit mechanisms, including Strong versus Strong-to-Weak tagging and Sum-of-Zeros shadow-ledger analysis. By insisting on immutability, replay, and separation between live certification and long-horizon auditing, the manuscript provides a durable operational blueprint that applies equally to photonic, magnetic, valleytronic, phononic, and Hilbert-style experimental programs, enabling cross-platform comparison and long-term integrity without exposing raw traces or proprietary control logic.
This manuscript establishes a fixed, platform-agnostic measurement corridor for cross-bench certification and applies it uniformly to both on-chip magnon devices and Gaussian homodyne photonics. External papers supply only their hardware and readout templates; all scoring, falsifiers, z-Tension budgeting, and replay rules originate in the present G–Σ–P–L grammar. The work formalizes a native three-parameter witness—F_guarded, Tz_center, and rho_back—under acceptance lines that never move, and introduces a registrar-verified, tokens-only replay path anchored by a one-tick time-triad lease. Reverse UBL contributes the P-lane magnitude (R′), while Centered-X, programmable UBL ⟂ DHP, and local TWA+ calibration provide the structural mapping that locks both platforms to a common contract. This bridge manuscript is technically important: it fixes the corridor for all upcoming triad-based evaluations and ensures that future device claims, photonic or magnonic, are falsifiable, replayable, and computed under the same immovable G–Σ–P–L rules.
Modern quantum error-correction pipelines are increasingly assembled from heterogeneouscomponents: neural predecoders trained on stabilizer syndromes, heuristic local correctors,and symbolic exact routines. Each component is correct in its regime and wrong outsideit, and a deployment decision must be made per workload, per round, under real-timeconstraints. We present a reproducible trust-layer and routing demonstrator for NVIDIA-style Ising-family surface-code predecoders, packaged as a standalone public artifact. Thesystem combines four load-bearing ingredients: (i) a hash-pinned benchmark surface of512 dual-basis distance-13 samples with SHA-256 contracts on every committed artifact;(ii) a provenance-forensics layer that executably localizes replay mismatches to the specificpipeline stage responsible for them; (iii) a structured-regime routing policy derived froma signed benchmark delta rather than a hand-tuned threshold, with a falsifiable mutationtest that proves the lane selection follows the evidence and not the implementer; and (iv) aCUDA-Q preflight that reports deployment readiness honestly, distinguishing “importable”from “realtime-ready.” We report exact historical parity: 512/512 matching samples acrossboth X and Z stabilizer bases, with the heuristic lane and the NVIDIA-lane outputs agreeingsample-by-sample with the committed reference. On the 512-sample benchmark the NVIDIAlane lifts mean resolved fraction by ∆ = +0.638 on line-propagation regimes and ∆ = +0.595on cycle-event regimes, while incurring ∆ = −0.080 on local-noise regimes — supportinga three-lane routing rule that the infrastructure encodes and tests. The complete artifact,including Lean-compatible reproducibility contracts and a patch-ready CUDA-Q executionpath, is published at https://github.com/Abraxas1010/ising-trust-routing.
We report three principal findings. First, two independent discrete-event quantum-network simulators—NetSquid 1.1.8 and SeQUeNCe 1.2.0—agree with an analytical Werner-swap oracle and with each other to machine precision in 160/160 state-level cases (max |ΔF| = 8.88×10−16) and in 50/50 protocol-level cases under matched source, channel, detector, heralding, retry, cutoff, BSM, readout, occupancy, latency, and goodput semantics. Second, when the shared harness is released and each simulator runs its own asynchronous event engine with independent random streams, mean output fidelity can be equivalent while the fidelity distributions differ strongly (Kolmogorov–Smirnov p≈10−27); a systematic timestamp-replay diagnosis localises this discrepancy to control/timing sampling rather than quantum-kernel disagreement. Third, a lightweight simulator (QMRB v0.2) with an incorrect Werner swap cross-term (/9 instead of /3) passes all counter-level parity checks but fails all 80 repeater fidelity comparisons with a mean error of 0.0235, demonstrating that modelform errors can be invisible at the counter level and silent until a fidelity threshold is crossed. To obtain these results we introduce a layered, pre-declared validation methodology comprising four layers: state-level parity, protocol-level parity, native-event parity, and model-form diagnosis. Each layer has an acceptance criterion frozen before execution. The methodology distinguishes implementation parity from model-form parity and from protocollevel parity, and reports explicit boundaries on what each layer does and does not establish.
After an experiment runs, exactly two decisions could still be made by a person: does this outcome exclude the theory's declared region, or did an auxiliary — source, tap, detector, calibration — fail, voiding the run instead; and who is licensed to say so. Any protocol that leaves either decision open after data exist has left a seat for a post-hoc refutation authority — the Global Falsifier, the imagined adjudicator who owns the modus tollens. Paper 44 of this sequence closed its photonic tapped-switch probe F_A at pre-registration-grade design ("the IV gate itself remains future laboratory work"); Paper 48 froze the shared adjudication layer as a handover template, typing everything design-specific as named pre-data slots behind a failable acceptance gate. What stands between that design and a laboratory's first shot is a single object: a completed, gate-passing handover instance. This paper — the forty-ninth in the sequence, centre panel of its Execution Program (Papers 48-50) — builds that object, the reference instance H_A, at the executable level the template promised. Delivered: the delta specification from the cited apparatus class (the Goswami-type photonic quantum switch) to 44's tapped configuration, component by component against the frozen budget rows; the schema completion with field-level primitives for every field, extensions add-only and partition-admissible; the frozen analysis pipeline in algorithmic normal form — IV gate, measured-calibration transfer, estimators, and the leaf router implementing 48's tap rows verbatim, under a language-independent rounding convention so floating-point artifacts cannot flip a frozen boundary direction — with the reference implementation shipped as a hash-anchored code appendix (deposited companion file) behind a declared production-scoring boundary; the concrete exclusion rule list with its veto-impact report and a frozen post-exclusion N rule disposing shortfall, surplus, and stopping at once; the S1-S4 blinding runbook; and the mock-data challenge, executed in three parts — a deterministic core exhaustive over every router branch and both tie directions, an adversarial set generated through a recorded, deposited exchange by a party in the auditor role that did not implement the scoring code, and a public-beacon-seeded bulk for totality and determinism. All three parts pass; the routing tables are generated by executing the shipped implementation and are typed non-evidence in the sentence that reports them. The acceptance gate is then run condition by condition on the deposited pair. One finding is published along the way: 48's coverage condition, read literally, would force every per-design instance to claim all seven leaves; the pre-committed completion evaluates coverage over the design's assigned leaf subset plus the shared non-execution leaf — an annotation to 48, never a retro-edit. The verdict is InstanceStatus_A = reference-complete, with an empty instance gap register: every gate condition passes, and what remains is ownership — acceptance transfers when a laboratory signs, auditor-reproducibly, over the instance plus its re-gated extensions; reference-complete never shortens to accepted, and an immutable core guarantees that no admissible extension can change what a null excludes, who may say so, or how boundary cases route. Displayed at the close is the channel's frozen exclusion content — the carve a future null already names, [1.7e-1, 2.1] at the conservative corner — beside the pre-priced Duhem escape map. Nothing is carved, scored, or measured; no band or feasibility verdict moves; the data paper 49-D stays reserved behind its unlock conditions. What retires is the interpretive seat: refutation content as a public, frozen, executable object — falsifiability without a Falsifier.