Live state-models over classical ledgers: a record-and-belief architecture whose belief is a density matrix

Working title: Live state-models over classical ledgers: a record-and-belief architecture whose belief is a density matrix

Author: [withheld in the repository copy]. Contact: [withheld]. Status: DRAFT — not submitted. Luke submits by hand, and only if he chooses to. Revision: 2026-11-02 second sweep (related work, venues). Issue: none filed; slate entry P1 (companion to the published truth-tracking paper; the qi-paper-1 card on the site).

Tiers: PINNED (CI gate on every change), MEASURED (run once on the owner's real data), DESIGNED (built, evaluation owed), DENIED (measured and rejected, kept with numbers), DARK (no measurement; no number typed).

Venues

Venue	Deadline (time zone)	URL	Fee / barrier	Checked
arXiv cs.AI, cross-list quant-ph	Rolling. Cutoff 14:00 US Eastern daily; announced 20:00 ET, Sunday–Thursday	https://info.arxiv.org/help/endorsement.html ; https://blog.arxiv.org/2026/01/21/attention-authors-updated-endorsement-policy	Free. One cs endorsement; quant-ph is its own domain and may ask for a second. "Independent" accepted as organization. Moderators may reclassify. Policy since 2026-01-21: automatic endorsement needs an academic email and prior arXiv authorship in the domain; every other new author takes the personal-endorsement route, which "arXiv staff cannot waive". No numeric endorser threshold is published	READ (policy post READ 2026-11-02)
Entropy (MDPI)	Rolling; SI "Explaining Economic and Social Science Phenomena Through Physical Models, Third Edition" closes 2027-02-28 (no time zone)	https://doaj.org/api/search/journals/issn:1099-4300 ; https://www.mdpi.com/journal/entropy/special_issues/U96HVBQ5X5	Fully OA, CC BY; APC up to 2600 CHF (DOAJ); waiver policy unread; single-anonymous; scope and special issues UNVERIFIED (mdpi.com still 403 on 2026-11-02). The SI's keywords include quantum-like models and decision making, known from a search snippet only (UNVERIFIED after search 2026-11-02); secondary to arXiv plus Entropy regular or FoP	DOAJ registry only; SI snippet only
Foundations of Physics (Springer)	Rolling	https://link.springer.com/journal/10701	Hybrid; OA APC £2490 / $3690 / €2890; subscription route free; no hybrid waivers; single-blind; desk-rejections final; affiliation required. The "belief" application is outside the stated scope; GKSL and filtering content is inside. Re-read 2026-11-02: median first decision 19 days; the journal carries a continuous QBism-and-agent debate 2010–2026 (Fuchs and Schack 2010; Pienaar 2020, 2021; Barzegar 2020; Gao 2021; Elia 2026), so the "which update rule should an agent use" framing has precedent there	READ (re-read 2026-11-02)
TMLR (alternative)	Rolling; holiday pause about Dec 2 → Jan 5 (pattern; no 2026–27 pause posted as of 2026-11-02)	https://jmlr.org/tmlr/	No fees; double-blind on OpenReview; public-email profile moderation up to 2 weeks; first criterion is evidence for claims	READ (re-read 2026-11-02)

No page cap at FoP or TMLR; Entropy limits UNVERIFIED. Public-safe figures: docs/figures/empty_diff_figure.png, berry_loop_trace.png; never the origin-data figures.

Abstract (draft)

Any system that acts in a world it cannot fully see keeps two things: a record of what happened, an append-only classical event log, and a belief about what is probably true now. We present an architecture in which the record is the only truth and the belief is a cache rebuilt by replay. The belief is a density matrix ρ over a small graph of two-level systems ("qubit clusters", one per role of a node), evolved by open-system (GKSL) dynamics and updated by Kraus maps, all simulated classically. Three coupling maps carry the loop: κ (ingest into ρ), ε (read-only projection of ρ) and χ (egress that decides visibly and dispatches nothing until promoted). We state a blank-slate theorem as an executable gate, the "log is truth" property as a daemon-against-replay parity test, and its measured limit: at the snapshot cursor the twins share the field hash but differ in 2235 walked process slots. Honest uncertainty is built in: an observation with efficiency η = 0 is designed as a no-op (DARK: proof or test not yet public), every value carries a nameplate (latent, live or dark), and a frame export marks each node current, stale or dark by its own age. We run one engine under a market feed, a game's play tapes and a home export, and report the estimator rung that lost. We concede that a classical model with more parameters would fit the same data. The claim is a representation with built-in honesty constraints and a measured ladder, not quantum advantage.

Contributions

1. The architecture as code paths. Record (append-only log) and belief (ρ over qubit clusters, dense or second-order cumulant closure), the coupling maps κ / ε / χ named by their modules, and a lint that fails the build if a domain word enters the engine (PINNED).
2. The blank-slate theorem as an executable gate. A maximum-entropy engine boots a world it has never seen, replays a deterministic synthetic day through the production ingest path, and every declared binding drives the field (PINNED).
3. Daemon parity and its measured limit. A killed worker recovers the exact field hash from snapshot plus log tail. The honest limit, the "continuation gap", is 2235 walked process slots differing between snapshot twin and from-log twin (MEASURED, declared lower bound).
4. Honest uncertainty by construction. η = 0 is a no-op on the field; nameplates record evidence status, never geometry; the frame/1 export leaves every unsignalled node dark and promotes nothing (PINNED).
5. Worked instances in foreign domains with the DENIED tier shown. A market feed over two disjoint years, a game's play tapes and a 13-day home export, with the Belavkin rung that never beat the blank engine (MEASURED, DENIED).
6. Exact correspondences and a pre-empted objection. The α-blend is a CPTP replacement channel (propagation, not conditioning); η is detection efficiency, so η = 0 recovers the unconditional master equation; the "more classical parameters" objection is conceded, not fought.

Outline (journal length, about 9,000 words)

1. Introduction (700 words). A thermostat, a trading desk and a game act on a picture of the world they cannot verify. Each keeps a record and a belief; most architectures blur them. We separate them by law: the record is truth, the belief is a cache. The belief is a density matrix because one object then carries mixture, uncertainty and a mixedness measure.
2. The record and the belief (900 words). The record is an event log; the belief is ρ over a graph of qubit clusters, one qubit per (node, role), run dense (exact 2ⁿ) or as a second-order cumulant closure (O(n²)). A DomainSpec of nodes, roles, bridges, bindings, outputs and drivers makes the world data; the engine carries no domain vocabulary and no engine module may import an adapter.
3. Three coupling maps and the executable theorems (1,200 words). κ ingests observations into ρ; ε is a read-only projection of ρ; χ is a shadow-first egress that decides visibly and dispatches nothing until promoted. The design pattern is stated here; its code paths, the first-sentence replay and its counts are withheld with the private engine. (DARK: needs public evidence (a public repro or measurement) before it can be claimed.)
4. The continuation gap (800 words). The daemon-parity test ends a world driven over HTTP at the exact canon hash of a library-direct replay. At the snapshot cursor the twins share the hash but differ in 2235 walked process slots (41 changed, 1481 only from-log, 636 only snapshot, 77 alias moves; 6061 vs 5216); a learning-hot tail then forks the hash. Incremental parity is xfail-pinned; chains that must referee clean re-evolve from the log. We relate this to Chandy–Lamport channel state, CRIU's inventory and failure transparency.
5. Honest uncertainty by construction (900 words). Every observation carries η ∈ [0, 1]; η = 0 is pinned as a no-op. Nameplates are latent (never observed), live, or dark (last η = 0) and never copy ρ. The frame/1 export derives current / stale / dark from each value's age against a producer-stated threshold; a blank boot exports every node dark with null and leaves the hash untouched. Blank construction mixes analog-dissipative roles to maximum entropy rather than a pure pole.
6. Exact correspondences: blend, η, instruments (1,000 words). ρ ↦ (1−α)ρ + α·ρ_target mixes the identity channel with a replacement channel; both are CPTP, so the blend is a quantum operation. In Leifer and Spekkens' vocabulary it is belief propagation through a partially trusted channel, not conditioning; in the diagonal case it is a linear opinion pool, a Jeffrey-style update. In the stochastic master equation (Jacobs and Steck 2006, eq. 91) the innovation term carries √η while the back-action does not, so at η = 0 the conditional state obeys the unconditional master equation. The engine updates by Kraus maps of partial efficiency, never projections, which Khrennikov et al. (2014) show is needed for replicability and order effects to coexist.
7. Worked instances in foreign domains (1,200 words). Market, 2025-07 → 2026-06: persistence AUC 0.4902, α-blend 0.4932, Belavkin 0.4595, blank engine 0.5170; n = 375 cells, 5 tickers, 76 held-out sessions, "inside the noise band at this n". Market, 2024-07 → 2025-06: persistence 0.4492, Belavkin 0.4754, engine 0.5563 on 75 sessions. Home export, 13 days, private: 45 bindings drove the field; fiber 0.054 → 0.129; belief AUC 0.94 / 0.87 / 0.79 in three well-sensed rooms against 0.60–0.67 for persistence of own inputs. Game tape, 138 ticks: flat, reported as an honest null.
8. Pre-empting the parameter objection (600 words). A 2ⁿ-dimensional ρ holds any classical distribution over n bits in its diagonal, so a classical model with more parameters would fit the same data; we concede this. Boyer-Kassem et al. (2016) found non-degenerate quantum models failed Grand Reciprocity in 65 of 72 datasets; psychology's reply is a parameter-free prediction (Wang et al. 2014). Ours is a measured ladder against persistence and a classical blend with the loss published, plus honesty constraints enforced by construction. We claim no contextuality; a sensor graph has direct influences everywhere (Dzhafarov et al. 2016).
9. Related work and the negative space (1,000 words). As below.
10. Threats to validity, DARK, what this paper is not (500 words). As below.

Related work and the negative space

The record side. Helland (2015), CIDR: "The truth is the log. The database is a cache of a subset of the log." Fowler (2005), martinfowler.com: state "purely derivable from the event log"; replay reaching external systems needs a Gateway. Young (2010), CQRS Documents: rolling snapshots are "a heuristic". Kleppmann and Kreps (2015), IEEE Data Engineering Bulletin: a materialized view as a cached subset of the log. Chandy and Lamport (1985), ACM TOCS: a consistent snapshot must record channel state. Veresov, Spenger, Carbone and Haller (2024), ECOOP: failure transparency, a formal snapshot-plus-replay equivalence. CRIU wiki, "What cannot be checkpointed" (practitioner, informal).

The record side, 2026: event-sourced agents and world models over append-only logs. Nakajima (2026), arXiv:2605.21997, "The Log is the Agent": an append-only event log as source of truth, the working graph a deterministic projection of it, a "determinism contract" for replay, forking at any event; no snapshots, no uncertainty object. Nakajima (2026), arXiv:2606.10241, "Regimes": every promotion or discard is itself an event. Rovai (2026), arXiv:2605.15967: agent state as "an append-only log of typed RDF triples", counterfactuals by forking the log, no learned components. McCann (2026), arXiv:2605.24036, "Intent-Driven Computing": programs emit intents, a governed runtime records every decision in a tamper-evident ledger before any effect, "no alternative path to effects", "event sourcing by construction"; the closest published analogue of a χ map that logs before it acts. Chang and Chang (2026), arXiv:2607.12480, TRACE: "no durable state change without a record" (illustrative only). Su, Xu, Zuo and Bertino (2026), arXiv:2607.27773, ChronoMem: agent memory "as an append-only event log" with "a whole-memory snapshot materialized from a prefix of the log"; the log-plus-snapshot design named as a storage-for-restoration trade. Brito dos Santos Filho (2026), arXiv:2602.23193, ESAA: agents emit only structured intentions, a deterministic orchestrator appends and projects, esaa verify replays with hashing. Goes (2026), arXiv:2605.17401: append-only logs "underpin immutability, replay, and history-derived state". Zhou et al. (2026), arXiv:2607.19949, SenWorld: a deterministic, event-sourced digital-twin simulation labelled by pointer to an existing record. Beber (2025), arXiv:2504.11635: event sourcing for systems-biology models, one log and "multiple specialized read models". Yang, Liu, Wu and Wang (2026), StateLedger, NeurIPS 2026 poster (OpenReview aNVVPOAxzx): memory as "durable state evolution" with CommitPages recording drift against Checkpoints; append-only and replay are not stated on the page. SkillFoundry (GitHub, archived 2026-05-09; software, not a paper): append-only immutable observation events with a Beta-Bernoulli belief node replaced atomically on each update, the exact record/belief split with a classical belief. None of these abstract pages shows peer review yet except StateLedger. Hu et al. (2025/26), arXiv:2512.13564, "Memory in the Age of AI Agents", is the survey hub and carries no event-log framing.

Belief states and belief change. Åström (1965), J. Math. Anal. Appl., and Kaelbling, Littman and Cassandra (1998), Artificial Intelligence: the belief is a sufficient statistic of the history, hence a fold over the log. Katsuno and Mendelzon (1991), KR / CUP 1992: revision (new information, static world) against update (the world changed). Diaconis and Zabell (1982), JASA: Jeffrey conditionalization; an observation at the prior leaves belief unchanged. Baltag and Smets (2006), MSCS: quantum measurements as epistemic actions in dynamic logic.

Density matrices as belief. Bouten, van Handel and James (2007), SIAM J. Control Optim.: the Belavkin filter as conditional expectation given the record. Jacobs and Steck (2006), Contemporary Physics: the SME with detection efficiency η. Wiseman and Milburn (1993), PRA, and (2009), CUP §4.8.1: inefficient detection. Leifer and Spekkens (2013), Phys. Rev. A 88, 052130, DOI 10.1103/PhysRevA.88.052130 (journal reference confirmed from the arXiv:1107.5849 record on 2026-11-02; later versions drop "Towards" from the title): conditional states; the projection postulate is "a misleading analogy" for Bayesian conditioning. Warmuth (2005), NIPS: a density-matrix Bayes rule that reduces to the classical rule when matrices are diagonal. Warmuth and Kuzmin (2006), UAI 2006, pp. 503–511, arXiv:1408.3100: joint and conditional density matrices, a total-probability analogue and Bayes-style rules returning a posterior density matrix from a prior and data, classical probability recovered when all matrices are diagonal; the classical-side anchor for "belief = ρ updated by evidence" with no quantum hardware and no dynamics. Monras, Beige and Wiesner (2011), AMCS, and Srinivasan, Gordon and Boots (2018), AISTATS: hidden quantum Markov models, ρ as belief over classical symbols updated by Kraus maps. Ning (2026), arXiv:2608.06554: an HQMM claimed to beat an EM-trained HMM on non-quantum data (synthetic; Splice only at higher latent dimension), a positive result with no record layer. Lanza, Solinas and Mastrogiovanni (2020), arXiv:2006.16404, published in Experimental Robotics (Springer Proceedings in Advanced Robotics, 2021, pp. 502–509, DOI 10.1007/978-3-030-71151-1_44): a quantum-like robot perception model with belief-query operators; resolved from the full text on 2026-11-02 as a pure state, simulation only, with no open-system dynamics, log or classical baseline. Radha and Goktas (2026), arXiv:2605.25313, UWM-JEPA: a density-matrix latent on a joint system–environment space holding beliefs over hidden futures, with a learned unitary predictor that "cannot shrink the represented uncertainty" (the opposite design choice to a dissipative belief); hidden-velocity task 0.77 against LSTM-JEPA 0.53; code public. quantum-mind (Yeshwanth Guru; PyPI 3.1.0 uploaded 2026-10-12; software, not a paper; self-description from the PyPI record and README, not peer-reviewed): "keeps the robot's belief about a person as a density matrix over hypotheses", core listed as Lüders rule, density matrices, Lindblad; synthetic data only, "No study with real people and a robot has been run"; self-reports a loss to a Gaussian Bayes filter (0.347 against 0.280 at 60 training sequences) and claims no quantum advantage.

Quantum cognition, QBism and their critics. Busemeyer and Bruza (2012), CUP, and Pothos and Busemeyer (2013), BBS: quantum probability as a language for belief. Caves, Fuchs and Schack (2002), PRA: ρ as an agent's degree of belief. Boyer-Kassem, Duchêne and Guerci (2016), Mathematical Social Sciences: the flexibility objection made exact. Wang, Solloway, Shiffrin and Busemeyer (2014), PNAS: the QQ equality on 70 surveys. Khrennikov, Basieva, Dzhafarov and Busemeyer (2014), PLoS ONE: projective models cannot have both replicability and order effects. Dzhafarov, Kujala, Cervantes, Zhang and Jones (2016), Phil. Trans. R. Soc. A: most behavioural "contextuality" is signalling.

Digital twins. Kapteyn, Pretorius and Willcox (2021), Nature Computational Science: the twin as a probabilistic graphical model. Suhail et al. (2022), ACM Computing Surveys: ledger plus twin, the ledger used for integrity. "Quantum digital twin" is a term collision to disambiguate in print: Otgonbaatar and Jennings (2024), arXiv:2410.23311, means a twin of a quantum processing unit; Taheri Monfared, Bombarda, Gargantini and Haghparast (2026), arXiv:2602.15477, means a quantum-accelerated twin of a classical system (Nimbe et al. 2026, Int. J. Adv. Manuf. Technol., DOI 10.1007/s00170-026-19078-2, likewise; Crossref record only, abstract not retrieved).

Umwelt in AI, 2023–2026. The word is now live in AI and must be cited: Sang (2025/26), arXiv:2511.05540, a Bayesian "Umwelt" world model for driving that cites von Uexküll; Bentley, Lim and Ishikawa (2025), ABMHuB'25 at ALife 2025, arXiv:2509.03380, umwelt-like "aspects" for LLM agents; Jehu-Appiah (2026), arXiv:2603.27626, "Umwelt engineering" of LLM vocabularies; Bosch, Sommers, Doerig and Kietzmann (2026), arXiv:2604.17960, an "Umwelt Representation Hypothesis" for ANN–brain alignment; Ay and Löhr (2015), Theory in Biosciences 134(3): 105–116, DOI 10.1007/s12064-015-0217-3, the measure-theoretic Umwelt as an agent's intrinsic σ-algebra on a sensorimotor loop. None uses a density matrix, κ / ε / χ maps, dark states or shadow egress.

The negative space. The nearest neighbours are the HQMM line (Monras 2011; Srinivasan 2018), which has ρ as belief over classical symbols but no record, no replay and no snapshot layer; Kapteyn et al. (2021), who have the record → model → decision twin but keep the model classical; Suhail et al. (2022), where the ledger protects integrity and never boots the twin; Lanza et al. (2020), who have sensors in and belief queries out but no open-system dynamics, coupled role graph or classical baseline (and, from the full text, a pure state in simulation only); Veresov et al. (2024), who formalise snapshot-plus-replay for dataflow, not for a belief state; and, since 2026, the event-sourced-agent cluster (Nakajima; Rovai; McCann; ChronoMem; StateLedger; ESAA; SkillFoundry), which has the record half entire, an append-only log as sole truth with state as its deterministic projection, and a classical or absent belief. On the belief side the nearest are Warmuth and Kuzmin (2006), a posterior density matrix by a Bayes rule with no log and no dynamics; quantum-mind, a density matrix with a Lindblad core inside a robotics-framed library, synthetic only; and UWM-JEPA, a density-matrix latent under a unitary predictor. The novelty sentence is therefore narrower than the first draft's: the record half of this architecture is now a 2026 wave; what remains unfound is a density-matrix belief as the derived view over such a log, and a measured continuation gap between a snapshot-booted belief and its from-log twin. None states "belief is a fold over the log" across control (Åström; Kaelbling), revision (Katsuno–Mendelzon), epistemic logic (Baltag–Smets), the density-matrix Bayes rule (Warmuth–Kuzmin) and quantum filtering (Bouten et al.) in one architecture note. Machine-checked claims ledgers with refuted rows now exist in repositories (Chen et al. 2026, arXiv:2609.26160, a public scorecard of 18 supported / 5 failed / 2 untested / 1 mixed pre-registered claims; the claimkeeper and claim-ledger tools; Fetna 2026, SSRN, index only, UNVERIFIED), but none ships the quantum rung with its loss recorded under a referee-enforced DENIED tier. The research notes mark these "not found", not "does not exist". The owed arXiv search ran on 2026-11-02 and returned zero relevant hits: abstracts matching "density matrix" with "event sourcing", none; "density matrix" with "append-only", "event log" or "event-sourced", one irrelevant; the quantum or density-operator variants with event-sourcing terms, two irrelevant physics items; OpenAlex full text for "density matrix" "event sourcing", 16 works, none relevant. Semantic Scholar was rate-limited and not searched.

Second sweep (2026-11-02). Web search was restored for a second pass over the record and belief notes (19 and 20 queries before the shared budget closed; the rest ran as direct fetches against arXiv, OpenAlex and Crossref). Verdicts per negative-space item: the log-as-truth-plus-ρ-belief pairing is UNCHANGED, with the nearest neighbours now named above; "belief as a fold over the log" is NARROWED by Warmuth and Kuzmin (2006), which the synthesis must include, and by Goes (2026) and Rovai (2026) stating "history-derived state" as a kernel or world-model property; the continuation gap is NARROWED on the practitioner side and UNCHANGED on the academic side: a running daily replay-equivalence job with a diff oracle and a documented residual-nondeterminism list exists in one public repository (vaultquest-archive issue #751 and merged PR #762, September 2026), one project withdrew a years-old "state is derived from replay" claim after finding no commit had ever built it (holomush issue #4784 and its ADR, 2026-07-11), and two practitioner posts (nilus.be; Anhaia, dev.to) prescribe sampled nightly reconciliation of snapshot-derived state against full replay, while no academic empirical study of snapshot/replay divergence was found (Shyngys et al. 2026 measures recovery latency only; Kumar and Saminathan 2026, arXiv:2609.28779, output agreement and completion only; Trofimov et al. 2019, arXiv:1907.06250, formal; Breitner and Smith 2017, PACMPL ICFP, DOI 10.1145/3110247, an experience report). The shadow-first χ map is NARROWED on the egress side by McCann (2026), TRACE (2026) and ESAA (2026); learning gated by the actuator's own causal footprint has no neighbour. The use of the Umwelt term is NARROWED: "Umwelt plus world model" is no longer empty (Sang; Ay and Löhr) and the paper must cite both; the quantum-Umwelt mapping is unclaimed. Dark-as-first-class, decay-as-uncertainty and vocabulary linting are UNCHANGED. Still UNVERIFIED after search 2026-11-02 and not cited above as verified: Kreps (2013) verbatim sentences (index-located, page 404; the Kleppmann–Kreps 2015 bulletin is the stand-in); "Memory as Distributed State in Multi-Agent LLM Systems" (no arXiv record); Fetna (2026) SSRN (403); the Carreras (2026) "claims ledger" detail; the ACM Artifact Review and Badging page (403). Owner numbers in this file are unchanged by the sweep.

Evidence each claim rests on

Claim	Pointer	Tier
A world is data (DomainSpec); no domain word in the engine; no engine import of adapters	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
Belief is ρ over qubit clusters, dense (2ⁿ) or second-order cumulant closure (O(n²))	substrate/cumulant_cluster.py, substrate/cluster.py	PINNED (code)
Cumulant closure: 100.00% decision parity vs full ρ on ZZ; 0.02 max divergence at J = 0.6	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
Blank-slate theorem	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
Log is truth, snapshot is cache (daemon parity)	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
Continuation gap: 2235 slots (41 / 1481 / 636 / 77; 6061 vs 5216); 5 of 49 state owners not spliceable	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
κ / ε / χ as code paths; egress shadow-first	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
First sentence: 605 batches, 2396 events; 9 of 10 beliefs distinct over 1.95 of 2.00; fiber −9.39 bits / 87 qubits; world +11.97 bits / 21 qubits; 18 of 19 axes; 0 actions	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
Blank boot mixes analog-dissipative roles to maximum entropy	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
η ∈ [0,1]; η = 0 is a no-op on the field	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
Nameplates latent / live / dark never copy ρ	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
frame/1 export: dark with null on blank boot; nothing promoted; hash untouched	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
Market 2025-07 → 2026-06: persistence 0.4902, L0 0.4932, L4 0.4595, engine 0.5170; n = 375 cells. 2024-07 → 2025-06: persistence 0.4492, L4 0.4754, engine 0.5563 (5/5 tickers)	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
Belavkin rung default OFF ([private, withheld])	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
Home export, 13 days: 45 bindings; fiber 0.054 → 0.129; AUC 0.94 / 0.87 / 0.79 vs 0.60–0.67	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK
Game tape: 138 ticks flat, reported as a null	DARK: needs public evidence (a public repro or measurement) before it can be claimed	DARK

DARK (do not claim a number)

• Live production decisions: no χ output has been promoted to a consequential live action.
• Any human study: none.
• Coherence advantage on real data (U1 > U0): owed.
• Forge authoring and warden proposal quality: owed.
• Anything measured on the earlier home deployment the engine descends from, including the self-crediting effect size: the rescission ceiling applies.

Threats to validity

One author, one engine at version 0.1.0, one owner's data. The market result is "inside the noise band at this n" (375 cells); the dossier records no interval, so one must be computed before submission. The home export is private; a reader cannot re-run its AUCs. The continuation gap is a declared lower bound and will move. PINNED means a gate runs on every change, not that a theorem is proved; the "blank-slate theorem" is a named executable check. The paper's own sentences are claims (η 0.25) under the truth-tracking ladder; only the pointers are referees.

What this paper is not

• Not a quantum-advantage claim; no quantum hardware is used or implied.
• Not quantum cognition: the claim is engineering, not a theory of human judgement.
• Not a claim that the Belavkin rung helps; it lost (P2 reports the loss).
• Not an evaluation of the forge, the warden or the proposed constitution (docs/LAW.md).
• Not a report on any earlier home deployment.
• Not a user study; not a claim of contextuality or non-classicality.

Luke's steps, if he wants to submit

1. Register at arxiv.org with any verified email, organization "Independent"; link an ORCID (free).
2. Start a submission with primary cs.AI; arXiv emails a six-character endorsement code. On a nearest-neighbour abstract (Lanza, Solinas and Mastrogiovanni, arXiv:2006.16404; or a 2026 event-sourced-agent paper such as Nakajima, arXiv:2605.21997, or Rovai, arXiv:2605.15967) click "Which authors of this paper are endorsers?" and send one endorser the code plus the PDF. Endorsers need domain papers dated 3 months to 5 years ago. Do not mass-mail. Since 2026-01-21 this personal route is the only one for an author without an academic email and a prior arXiv paper in the domain, and "arXiv staff cannot waive endorsement requirements" (READ 2026-11-02).
3. If the quant-ph cross-list asks for endorsement, repeat with a quant-ph neighbour (Khrennikov, arXiv:2304.08599; Asano and Khrennikov, arXiv:2604.18643).
4. Pick a license: arXiv perpetual non-exclusive if the journal is undecided; CC BY 4.0 if Entropy is the target; irrevocable per version.
5. Submit before 14:00 ET; nothing announces Friday or Saturday.
6. Choose one journal. Entropy: read the APC, waiver and special-issue pages in a browser (bot-walled). Foundations of Physics: subscription route; frame on the GKSL and filtering foundations; affiliation "Independent researcher" (handling UNVERIFIED). TMLR: create the OpenReview profile now with the portfolio site as homepage showing name and email, wait out moderation, use the TMLR class, anonymize, ship code anonymized.
7. Before any submission re-run [private, withheld], [private, withheld] and [private, withheld]; record the dates and the new gap counts; compute an interval for the market AUCs. (DARK: needs public evidence (a public repro or measurement) before it can be claimed.)
8. Log the submission in the authority record when there is a receipt.