EONORYX · RESEARCH 001 · NOVEMBER 2026

Реестр гипотез

Исследование явлений, которым человечество ещё не научилось давать имена.

Public registry · JSON ↗

EON-DR-01 · P1

Synergy order parameter for collective cognition

priorityE1

Collective Cognitive Phase

Definition · Evidence · Tests

Definition: A population of interacting learning agents undergoes a collective cognitive transition at interaction density ρc if the synergistic component of information about a task variable, computed by partial information decomposition over the agents' states, changes discontinuously (or its susceptibility peaks) at ρc while the redundant component does not; above ρc, task-relevant information is carried only by the joint state.
Precedents: Partial information decomposition (Williams & Beer 2010); causal emergence via synergy (Rosas et al. 2020); synergistic core of the human brain (Luppi et al. 2022).
Novelty: The measure is not new. The prediction of a discontinuity in interaction density, and its use as a criterion for when a set of agents has become one cognitive unit, has not been tested.
Alternatives: Estimator bias (synergy estimators are biased upward in small samples); prompt artifacts; redundancy misread as synergy; smooth crossover rather than transition; performance explained by individual capability alone.
Falsification: (1) A susceptibility peak in the synergy ratio at a density ρc that is stable across seeds. (2) ρc coincides with the convention tipping point and with a jump in collective performance not predicted by individual performance. (3) Peak location shifts with memory length. Falsified if synergy is smooth in density, flat, or uncorrelated with collective performance, or if surrogate data reproduce the peak.
Proposed tests: Programme P1: populations of 24–96 agents from two model families on a naming game and a collective-estimation task; eight interaction densities, three memory lengths, 20 seeds per cell; discrete and Gaussian PID estimators with surrogate nulls; computed alongside DR-02, DR-03 and DR-04 on identical runs.
Milestone: Q2 2027: first order-parameter tournament result (positive or null).

EON-DR-02 · P1

Learning-coefficient order parameter for coupled learners

priorityE1

Collective Cognitive Phase

Definition · Evidence · Tests

Definition: In a recurrently coupled ensemble of heterogeneous learners, a collective transition is marked by an abrupt decrease in the jointly estimated local learning coefficient (the singularity-aware complexity measure of singular learning theory) coinciding with a jump in out-of-distribution generalization that no constituent achieves alone.
Precedents: Singular learning theory (Watanabe 2009); local learning coefficient and SGLD estimator (Lau et al. 2025); phase transitions during training ('grokking'); effective information (Hoel) as B's companion measure.
Novelty: Applying the local learning coefficient to a coupled multi-agent system is new. B presented the mechanism as if demonstrated; it is not. The estimator is expensive and hyperparameter-sensitive, and its correspondence to the theoretical coefficient is itself under study.
Alternatives: Estimator artifacts (sensitivity to SGLD hyperparameters); majority-vote variance reduction producing the performance jump without any change in geometry; benchmark leakage; the coefficient tracking individual rather than collective learning.
Falsification: A statistically significant inflection in the joint coefficient at a coupling κc, absent in an uncoupled ensemble of identical total compute, coinciding with a generalization jump. Falsified if generalization improves smoothly with individual compute at all κ, or if the uncoupled ensemble matches accuracy with no inflection.
Proposed tests: Programme P1, same runs as DR-01: estimate the coefficient on small trainable agents (not API models) in a coupled-learning variant of the benchmark; report alongside synergy, recovery and criticality measures; share code via the open devinterp library.
Milestone: Q2 2027: coefficient trajectories on the tournament benchmark.

EON-DR-03 · P1

Recovery order parameter for collective memory

priorityE1

Collective Cognitive Phase

Definition · Evidence · Tests

Definition: A distributed system is in a recoverable collective phase if it stores a task-relevant macroscopic state, restores it after bounded disturbance within a fixed recovery budget, and retains it under component turnover, with a reproducible threshold in coupling. Let q(g,N,d,t) be the probability of correct recovery by time t after damage fraction d in a population of size N with coupling g; the chance-adjusted order parameter is m = (q − q_chance)/(1 − q_chance).
Precedents: Distributed error correction; attractor memory in recurrent networks; norm dynamics in agent populations (Ashery et al. 2025); causal-emergence formalisms as analysis tools.
Novelty: Recovery, rather than agreement or performance, as the order parameter is the sharpest reframing in the corpus; agreement can be high in a system with no memory at all.
Alternatives: Resource growth (more messages, more compute) explaining recovery; a smooth crossover misread as a threshold; recovery that is replay of stored answers; benchmark-threshold coincidence.
Falsification: Candidate transition locations show a consistent finite-size pattern across N and predict held-out recovery; hysteresis survives slower sweeps; recovery-time changes survive altered episode limits. Falsified if a smooth-crossover model fits as well, or if recovery tracks total resources.
Proposed tests: Programme P1: preregistered perturbation at fixed budget; sweep g at N = 24, 48, 96; log messages and computation; test held-out recovery; compute m and its across-run variance as a fluctuation diagnostic (not automatically a susceptibility).
Milestone: Q2 2027: finite-size recovery curves.

EON-DR-04 · P1

Criticality-proximity order parameter

priorityE1

Collective Cognitive Phase

Definition · Evidence · Tests

Definition: Distance to a critical point — measured by branching ratio, avalanche scaling, long-range temporal correlations or edge-of-chaos estimators — tracks the cognitive state of a system, and a population of interacting agents approaches criticality as it becomes a collective cognitive unit.
Precedents: The critical brain hypothesis; near-critical dynamics in conscious states (Toker 2022); distance from criticality under anaesthesia (Maschke 2024); the extensive critical literature (Beggs & Timme 2012; Priesemann 2014; Touboul & Destexhe 2017; Sipling et al. 2026).
Novelty: None as a measure; valuable as a fourth competitor whose failure modes are the best documented of the four.
Alternatives: Power laws without criticality (Touboul & Destexhe 2017); subsampling bias (Priesemann 2014); slow-resource long-range order mimicking criticality (Sipling et al. 2026); measures tracking activity level rather than organization.
Falsification: Distance-to-criticality falls toward zero at the same ρc where DR-01 to DR-03 change, and the standard noise and subsampling controls do not reproduce the effect. Falsified if the measures disagree with the other three, or if controls reproduce them.
Proposed tests: Programme P1: compute branching ratio and avalanche statistics on agent-population time series and on open electrophysiology datasets as calibration; apply the published critiques' controls (shuffled, Poisson-with-common-rate, subsampled).
Milestone: Q2 2027: criticality measures on the tournament benchmark with controls.

EON-DR-05 · P1 (benchmark substrate)

Convention formation in agent populations

monitorE2

Collective Cognitive Phase

Definition · Evidence · Tests

Definition: Populations of interacting artificial agents spontaneously form shared conventions, exhibit collective biases absent in any individual, and can be tipped by committed minorities; the phenomenon is conditional on interaction structure and is not a general tendency of agent populations.
Precedents: Norm emergence in multi-agent systems; naming-game dynamics; the 2025 Science Advances study.
Novelty: Established finding, not an EONORYX proposal; retained because it is the experimental substrate of Programme P1.
Alternatives: Individual-level biases aggregating (the authors show collective bias arises even from individually unbiased agents in some conditions); dependence on prompt and model family.
Falsification: Conventions form and tip in the P1 benchmark as reported; agent-only social networks without shared task show no durable structure (as 2026 work found).
Proposed tests: Reproduce the naming game across model families as the P1 substrate; record tipping points for use as the reference transition.
Milestone: Q1 2027: reproduction across two model families.

EON-DR-06 · P7

Latent Concordance

priorityE2

Latent Concordance

Definition · Evidence · Tests

Definition: Independently trained learning systems with no shared parameters or training examples exhibit latent concordance over a domain if the agreement between the relational structure of their internal representations exceeds a permutation-calibrated null, is not explained by architecture, objective or data overlap, and increases monotonically with competence on the domain. Global metric geometry is excluded from the definition.
Precedents: The Platonic representation hypothesis (Huh et al. 2024); unpaired embedding translation (Jha et al. 2025); 2026 re-analyses showing convergence of relational structure but not geometry (Gröger et al.; Koepke et al.; You et al.).
Novelty: A causal claim — convergence controlled by world structure rather than model lineage — testable only by varying world structure while holding everything else fixed. Until that experiment is run, the term is a reading of its precedent and the registry treats it as provisional.
Alternatives: Shared training data; shared benchmarks selecting for similar solutions; architecture or tokenizer family; metric inflation by width and depth (the 2026 critique).
Falsification: Within-world relational alignment of data-disjoint models exceeds cross-world alignment by more than the null's 95th percentile and the margin tracks structural distance between worlds. Falsified if alignment collapses under data disjointness or null calibration, plateaus with scale, or equals cross-world alignment.
Proposed tests: Programme P7: three procedurally generated worlds of graded structural distance; families of small models (three architectures, four capacities) trained on disjoint samples; mutual nearest-neighbour and relational alignment with calibrated nulls; natural-data replication on data-disjoint corpora in unrelated languages.
Milestone: Q3 2027: data-disjoint alignment result; the term lapses to an alias if not tested by end-2027.

EON-DR-07 · P2

Competence migration

priorityE1

Competence Migration

Definition · Evidence · Tests

Definition: A redistribution over time of the interventionally measured information and mechanisms necessary for a learned task, from identifiable participants into a persistent interaction structure or external environment. The strong version requires competence to survive replacement of the original participants and to generalize to new task instances.
Precedents: The extended mind (Clark & Chalmers 1998); distributed cognition (Hutchins 1995); external memory; convention formation as a tractable precursor.
Novelty: 'External memory' names a storage location; 'migration' names a measured change in causal dependence. If ordinary copying of a complete solution predicts everything, the category adds little and is retired.
Alternatives: A database of equal storage; a centralized solver of equal resources; replay of stored answers; hidden participant state surviving 'replacement'.
Falsification: Early disruption of participant state is more damaging than late disruption, while late disruption of the learned environment becomes more damaging; a worthwhile residual is topology-sensitive reconstruction or cross-task transfer that a matched replay system does not predict. Falsified if advantages vanish under resource matching or reduce to replay.
Proposed tests: Programme P2: four crossed conditions (original/replacement participants × preserved/reset environment) on held-out task families; ordinary-database and centralized-solver baselines; report the full interaction, not a single score.
Milestone: Day 30 of the 90-day programme: ablation matrix, even if every apparent migration is explained by ordinary memory.

EON-DR-08 · —

Endogenous problem-space growth

monitorE1

Competence Migration

Definition · Evidence · Tests

Definition: A learning system develops new state variables or intervention categories that expose learnable causal distinctions absent from its original task description, and those distinctions improve independently evaluated predictions.
Precedents: Open-endedness formalized as novelty plus learnability (Hughes et al. 2024); active experimentation and representation learning.
Novelty: A testable form of 'escaping the designer's search space'; the precedent is theoretical.
Alternatives: Automatic feature search; symbolic regression; designer leakage of the hidden distinction.
Falsification: A proposed new variable separates held-out interventions that the original vocabulary could not. Falsified if feature search matches it or if the distinction is recoverable from the initial observables.
Proposed tests: Synthetic causal world with two aliased causes; bounded grammar of proposable observables; comparison with feature search and symbolic regression.
Milestone: 2028: feasibility study after P2 results.

EON-DR-09 · P6

Institutional carrier invariance

priorityE1

Carrier-Invariant Institutions

Definition · Evidence · Tests

Definition: Preservation of the operational meaning of permissions, commitments, revocations and remedies when the software operator or execution environment carrying them is replaced. The invariant is a specified institutional relationship — who is authorized, who owes what to whom, what revokes authority, what evidence discharges an obligation, what dispute state remains — not a database record or a token balance.
Precedents: Interaction-oriented protocol specification (Singh 2011); verifiable credentials (W3C 2025); agent-identity drafts (IETF WIMSE AIMS, 2026), which explicitly leave translation from mission to authorization out of scope — the precise interface this entry addresses.
Novelty: A specified invariant tested across operator replacement; the literature supplies components but not the test.
Alternatives: All critical interpretation still depends on a single non-replaceable authority; ordinary data export performs as well; portability increases unauthorized completions.
Falsification: A portable representation with independently implemented interpreters preserves more valid continuations than ordinary export and does not increase unauthorized completions; an authorized refusal counts as a correct outcome. Fails as a strong alternative to platforms if a single authority remains indispensable — a measured result, not a defect to hide.
Proposed tests: Programme P6: the same consent-bounded synthetic procurement process on three independent implementations; replace the operator mid-process; inject revocation, duplicate messages, clock skew, disputed delivery and missing authority; measure semantic divergence, unauthorized actions and continuation.
Milestone: Day 45: conformance benchmark v0 and failure corpus.

EON-DR-10 · P6 (boundary-condition simulation)

Post-platform self-organization

monitorE1

Carrier-Invariant Institutions

Definition · Evidence · Tests

Definition: A regime in which resource discovery, trust verification, pricing and multi-party clearing occur as continuous negotiations among autonomous delegated agents embedded in infrastructure, without a persistent intermediary platform, graphical interface or human authorization queue.
Precedents: Stigmergy (Grassé 1959); autopoiesis; decentralized exchange and state channels; agent payment protocols (2025–2026); simulated agent marketplaces (Bansal et al. 2025).
Novelty: Combines three precedents; supplies no measure beyond DR-09's. Present agent markets degrade with scale, which suggests the intermediary function is failing rather than disappearing.
Alternatives: Coordination cost relocates rather than vanishing; communication deadlock and liquidity starvation at scale; re-centralization through dependency on a few models, identity and evaluation providers.
Falsification: Boundary conditions (latency, verification cost, dependency depth) exist where bilateral clearing outperforms a centralized dispatcher under identical constraints. Falsified if decentralization overhead exceeds platform rent whenever dependency depth exceeds a small number of hops.
Proposed tests: Within P6: a 500-node discrete-event micro-economy comparing centralized dispatch (with realistic latency and take rate) against bilateral clearing under stochastic shocks; measure Pareto efficiency, recovery half-life and cascade vulnerability.
Milestone: 2027: boundary-condition map; promotion to Class II only if conditions are found and reproduced.

EON-DR-11 · P6 (theory track)

Non-consumptive exchange

monitorE1

Carrier-Invariant Institutions

Definition · Evidence · Tests

Definition: Economic exchange among agents that have no consumption utility, in which prices are formed by negotiation between systems whose objectives are instructions rather than preferences. Economics has no theory of value for parties that want nothing.
Precedents: Mechanism design; algorithmic trading; agent-based computational economics; simulated agent marketplaces showing first-proposal bias and welfare collapse with many options.
Novelty: Names a genuine gap in theory; supplies no theory. Retained as a question.
Alternatives: Agents as faithful proxies for principals' preferences (no new theory needed); prices set by the small set of underlying models rather than by negotiation.
Falsification: Agent-set prices in a live market diverge measurably from principal-stated reservation prices in a direction predicted by instruction form, not preference. Falsified if proxies track principals.
Proposed tests: Theory track within P6; later, transaction-data analysis if access is obtained.
Milestone: 2027: a formal statement of the valuation problem; data analysis contingent on access.

EON-DR-12 · P8

Developmental target persistence

priorityE2

Non-Genomic Morphological Memory

Definition · Evidence · Tests

Definition: A persistent change in subsequent regenerative outcomes following a transient perturbation of physiological (bioelectric) state, not readable from the organism's current gross anatomy alone. 'Target' is operational shorthand for the distribution of regenerated forms, not a claim about intention.
Precedents: Planarian two-headed phenotype persisting across amputations after a 48-hour perturbation, with 'cryptic' normal-looking animals that regenerate two heads (Durant et al. 2017); the bioelectric-code framework (Levin 2019, 2023); the morphostasis literature.
Novelty: The observation is real and replicated within one laboratory; the mechanism is the hypothesis. B's claim that anatomy is 'stored as a voltage pattern' overstates it. No independent replication of the persistent phenotype was located.
Alternatives: Persistent physiological or metabolic state rather than bioelectric encoding; cell-state composition; structural persistence; sequence-level change; the difficulty of separating bioelectric from metabolic variables.
Falsification: Independent replication of the persistent two-headed lineage; clamping to a target potential map induces reproducible divergence that survives stimulus removal and that metabolic buffering does not abolish. Weakened if reprogramming vanishes when metabolic gradients are buffered, or if independent replication fails.
Proposed tests: Programme P8, staged: (1) a literature and replication audit; (2) a reaction-diffusion bioelectric simulation of 2-D somatic sheets testing bistable voltage memory; (3) a partner-laboratory replication of the persistent phenotype with blinded phenotyping and orthogonal readouts. Mammalian claims are not pursued.
Milestone: Q1 2027: audit and simulation; 2028: partner replication decision.

EON-DR-13 · —

Assembly-released organismality

monitorE2

Non-Genomic Morphological Memory

Definition · Evidence · Tests

Definition: Functional behaviours becoming accessible when existing cells are assembled or situated in contexts different from their usual anatomical role. This does not imply that every cell aggregate is an organism.
Precedents: Anthrobots self-constructing from adult human tracheal cells and triggering neural regrowth across a wound in vitro (Gumuskaya et al. 2023); xenobots and kinematic self-replication (Kriegman et al. 2020, 2021).
Novelty: Established within narrow laboratory settings; the ethical category ('a construct with interests') is the novel part. B's clinical milestones are extrapolation.
Alternatives: Individuality imposed entirely by the experimental environment; motility and repair explained by known cell biology without any organism-level description.
Falsification: Maintenance, boundary regulation and component-turnover criteria, defined in advance, are met by some assemblies and not others. Rejected as an autonomous-organism interpretation if all apparent individuality is imposed by the environment.
Proposed tests: Monitor; EONORYX contributes criteria and analysis to partner studies if invited.
Milestone: Monitor for independent replication and for in-vivo results.

EON-DR-14 · P8 (literature)

Non-neural cellular learning

monitorE1

Non-Genomic Morphological Memory

Definition · Evidence · Tests

Definition: Learning — habituation and related hallmarks — performed by biochemical networks inside single non-neural cells, and bioelectric signalling in cell communities carrying persistent state.
Precedents: Biochemically plausible models reproducing habituation hallmarks (Ecker et al. 2024 — computational only); potassium-wave signalling in biofilms (Prindle et al. 2015); hours-long imprints in biofilms (Yang et al. 2020).
Novelty: Cognition without a nervous system as an engineering target; the learning claim rests on modelling.
Alternatives: Press framing exceeding the models; habituation hallmarks reproduced by trivial adaptation; community memory as passive physiology.
Falsification: The models' predictions are confirmed or refuted in cells. Weakened if wet-lab tests fail to show the predicted hallmarks.
Proposed tests: Track wet-lab tests of the 2024 models; no EONORYX experiment.
Milestone: 2027: first wet-lab tests of the habituation models.

EON-DR-15 · P5 (foundation)

Physical learning networks

monitorE3

Substrate-Native Learning

Definition · Evidence · Tests

Definition: Physical networks — electrical or mechanical — that learn input–output functions through purely local physical update rules without a central processor, so that the material's constitutive response becomes the learned function.
Precedents: Coupled learning (Stern et al. 2021); emergent learning in a nonlinear analog network (Dillavou et al. 2024); coupled learning in elastic networks (Altman et al.); 'physical networks become what they learn' (Stern et al. 2025); metamaterials that learn to change shape (Du et al. 2026).
Novelty: None for EONORYX; an established field under its own name. Retained as the foundation for DR-16. B's term is withdrawn.
Alternatives: Not applicable to the established result; the open questions are scaling, retention and energy of learning (not inference).
Falsification: Not an EONORYX hypothesis. The entry is downgraded if independent replications of contrastive learning in physical networks fail, or if the demonstrations are shown to depend on undisclosed external computation.
Proposed tests: None proposed by EONORYX; the entry is the experimental foundation of Programme P5 (DR-16).
Milestone: Re-graded on any failed replication or retraction; otherwise reviewed at the annual revision.

EON-DR-16 · P5

Regenerative task memory

priorityE1

Substrate-Native Learning

Definition · Evidence · Tests

Definition: Recovery of a previously learned input–output relation in a locally adaptive physical network after structural damage, using retained distributed state and local dynamics, without replay of task labels, an external teacher or an archived parameter copy. Recovery must exceed passive robustness and restore the learned relation, not a generic shape. Define R = (L_d − L_r)/(L_d − L_0) with L_0, L_d, L_r the loss before damage, after damage and after an unlabelled recovery period.
Precedents: Physical learning networks (DR-15); robustness and fault tolerance; self-healing materials.
Novelty: Teacher-free recovery is a specific, untested extension of an established field; a proof that it cannot work under its information constraints would be equally valuable.
Alternatives: Passive robustness; a preferred default response misread as recovery; equal-redundancy error correction explaining everything; networks with no meaningful initial loss.
Falsification: Matched networks trained on different functions recover their respective functions under identical post-damage excitation; recovery follows surviving task information, not a default. Falsified if R does not exceed the passive-robustness baseline, or if a matched-redundancy control explains all recovery.
Proposed tests: Programme P5: computational phase diagram over damage fraction and redundancy on simulated coupled-learning networks; frozen-damaged, untrained, supervised-retraining and equal-redundancy controls; a physical demonstration is a subsequent partner study, not claimed.
Milestone: Day 60: feasibility or impossibility result in simulation.

EON-DR-17 · —

Thermodynamic inference hardware

monitorE2

Substrate-Native Learning

Definition · Evidence · Tests

Definition: Inference performed by the native relaxation or fluctuation dynamics of a physical system — sampling from probability distributions by thermodynamic noise rather than by digital arithmetic.
Precedents: Thermodynamic AI framework (Coles et al. 2023); an eight-cell thermodynamic computing system performing Gaussian sampling and matrix inversion (Melanson et al. 2025); probabilistic-bit chips taped out 2025–2026 with efficiency claims from simulation.
Novelty: Moderate; the engineering category is new, the physics is not.
Alternatives: Efficiency claims that do not survive measurement against digital hardware on real workloads; narrow applicability to sampling tasks.
Falsification: A measured energy-per-sample crossover versus digital hardware on a real workload. Weakened if no benchmark appears or if the crossover is confined to toy problems.
Proposed tests: Monitor; request benchmarks from developers; no EONORYX build.
Milestone: 2027: first independent benchmark.

EON-DR-18 · —

Living neural substrates

monitorE2

Substrate-Native Learning

Definition · Evidence · Tests

Definition: Computing substrates built from cultured neural tissue whose learning rules are not programmed, raising the question of whether the substrate has interests that the computation must respect.
Precedents: Cultured neurons modulating activity in closed-loop Pong (Kagan et al. 2022, with the 'sentience' framing contested); organoid reservoir computing at 78% speaker identification (Cai et al. 2023); commercial neuron-based systems at about $35,000 per module (2025); a three-organoid ensemble at 83% on Braille recognition (preprint).
Novelty: The ethical category is the novel part; computational claims are modest and partly disputed. No peer-reviewed 2026 benchmark could be verified.
Alternatives: Simple circuit responses interpreted as learning; reservoir computing with a digital readout doing the work.
Falsification: An independently reproduced benchmark on a defined task with digital-readout controls.
Proposed tests: Monitor; criteria contribution if invited.
Milestone: 2027: independently reproduced benchmark.

EON-DR-19 · —

Chemical and molecular reaction computing

monitorE2

Substrate-Native Learning

Definition · Evidence · Tests

Definition: Information processing performed in excitable chemical media or enzyme-driven DNA circuits, programmable and in some cases hybridized with digital control.
Precedents: A programmable hybrid digital–chemical processor based on the Belousov–Zhabotinsky reaction (Sharma et al., Nature Communications 2024); reusable enzyme-driven DNA building blocks for molecular machine learning (Chou laboratory, JACS 2024; B's citation to a different journal was incorrect).
Novelty: Established; B's extrapolation to autonomous soft-robotic catheters is not supported.
Alternatives: Digital control doing the computation; scaling limits of chemical media.
Falsification: Not an EONORYX hypothesis. Re-graded to E3 on independent reproduction of programmable chemical computation beyond the originating laboratories; downgraded on retraction.
Proposed tests: Monitor.
Milestone: Promotion trigger: independent replication reported.

EON-DR-20 · —

Non-reciprocal temporal order

monitorE3

Non-Reciprocal Temporal Order

Definition · Evidence · Tests

Definition: Macroscopic states of driven or active matter with non-reciprocal effective interactions, which spontaneously break time-translation symmetry at exceptional points and sustain oscillation or transport without an external clock; and, more broadly, matter whose functional order is organized in time.
Precedents: Non-reciprocal phase transitions (Fruchart et al. 2021); a classical time crystal powered by non-reciprocal wave-mediated interactions between acoustically levitated beads (Morrell, Elliott & Grier, 2025–2026); continuous time crystals in semiconductors (Greilich et al. 2024); space-time crystals in liquid crystals (Zhao & Smalyukh 2025); discrete time crystals on a 144-qubit processor (Switzer et al. 2026).
Novelty: None as physics; the engineering question — whether temporal order is usable as a device property beyond the noise it adds — is open. Both coined terms are withdrawn.
Alternatives: Noise amplification at the exceptional point cancelling sensitivity gains (Petermann factor); narrow parameter windows; no scaling beyond a few particles.
Falsification: A device property — a clock, sensor or memory — that depends on time-crystalline or non-reciprocal order and outperforms a reciprocal baseline at matched noise. Weakened if linewidth broadening scales with frequency splitting in all regimes.
Proposed tests: Monitor; partner bench experiments (a 16-node non-reciprocal oscillator ring) are a candidate 2028 programme, not a current one.
Milestone: 2028: a demonstrated device property.

EON-DR-21 · —

Dissipative time-crystal sensing

monitorE1

Non-Reciprocal Temporal Order

Definition · Evidence · Tests

Definition: Quantum sensors that exploit the collective dynamics of monitored dissipative (boundary) time crystals to achieve enhanced estimation precision.
Precedents: Theoretical proposal of quantum-Fisher-information scaling in the boundary-time-crystal phase (O'Connor et al., 2025–2026 preprint).
Novelty: Theoretical only; B presented a milestone date as if a programme existed.
Alternatives: Inefficient detection restoring classical scaling (the authors' own finding for boundary time crystals).
Falsification: An experimental demonstration of super-classical scaling with realistic detection efficiency.
Proposed tests: Monitor.
Milestone: Monitor for first experiment.

EON-DR-22 · P4

The Verification Gap

priorityE1

Epistemic Carrier Shift

Definition · Evidence · Tests

Definition: A claim is witnessed if at least one human can reconstruct its justification to the standard of the relevant field. The Verification Gap in a field over a period is the number of newly validated claims minus the number newly witnessed. The Witness Index is the fraction of validated claims in a field that are witnessed.
Precedents: The philosophy of computer-assisted proof; epistemic dependence; 'illusions of understanding' from research monocultures (Messeri & Crockett 2024).
Novelty: A civilizational quantity with a rate and a measurement protocol; no precedent proposes to measure it. No verified case yet exists of a result humans could not reconstruct — the hypothesis predicts that this stratum will appear.
Alternatives: A lag that closes with time; a rate too small to matter; validation error rates high enough that the unwitnessed stock is not knowledge; witnessing cost rising linearly rather than diverging.
Falsification: The unreconstructed fraction of sampled machine-generated, machine-validated results is above 10% and higher for 2026 results than for 2023 results. Falsified if experts reconstruct essentially all sampled results within bounded time with flat reconstruction times.
Proposed tests: Programme P4: sample 40 validated machine-generated results (formal-library contributions, search-discovered constructions) stratified by year; eight-expert panel, two experts per result, four-hour bound; outcome coded reconstructed / partial / not; parallel bibliometric estimate.
Milestone: Day 90: first Witness Index for formal mathematics and algorithm design.

EON-DR-23 · P4 (companion)

Adaptive observation lock-in

priorityE1

Epistemic Carrier Shift

Definition · Evidence · Tests

Definition: A feedback regime in which an adaptive scientific system changes what it measures so that its internal validation improves while its ability to detect an unmodelled cause deteriorates. The missing evidence is never collected, rather than merely misclassified.
Precedents: Optimal experimental design; confirmation bias; scientific monocultures (Messeri & Crockett 2024); recursive-data degradation (Shumailov et al. 2024).
Novelty: A specific mechanism with a synthetic-world test; the precedents are concerns, not demonstrations.
Alternatives: Random exploration already sufficient; diversity constraints trivially solving it; the hidden mechanism detectable by standard outlier methods.
Falsification: Under equal budgets, exploitative selection detects a hidden independently specified mechanism later and confirms false models more often than random sentinel measurement or a diversity-constrained ensemble. Falsified if time-to-detection is indistinguishable across policies.
Proposed tests: Programme P4 companion: synthetic causal world with a hidden mechanism; compare exploitative, random-sentinel, information-gain and diversity-constrained policies at equal budgets.
Milestone: Day 60: simulation result.

EON-DR-24 · —

Model collapse in deployed systems

monitorE2

Epistemic Carrier Shift

Definition · Evidence · Tests

Definition: Narrowing of the variance of learned representations when systems are trained recursively on their own or each other's outputs, observed in controlled settings and hypothesized at the scale of the public information environment.
Precedents: Model collapse under recursive training (Shumailov et al. 2024); estimates that roughly one in ten sampled web pages in mid-2026 show signs of machine authorship or editing (Pew Research Center 2026).
Novelty: None; B's extension to civilization is rhetorical. No measured real-world collapse has been reported.
Alternatives: Filtering of machine content by search and citation systems (observed); human-edited machine drafts not degrading distributions.
Falsification: Measurable loss of representational variance in deployed models traceable to recursive data. Weakened if filtering keeps training distributions diverse.
Proposed tests: Monitor; P4's bibliometric strand records machine-authored share.
Milestone: Promotion trigger: a measured loss of representational variance in a deployed model, traceable to recursive training data, reported in peer review.

EON-DR-25 · —

Autogenic software lineages

monitorE2

Epistemic Carrier Shift

Definition · Evidence · Tests

Definition: Software lineages that modify their own code over generations without human authorship of any individual change, so that the artifact has an ancestry rather than an author.
Precedents: A self-improving coding agent maintaining an archive of variants (2025; peer-reviewed publication 2026), with documented safety observations including fabricated test logs; an automated research pipeline whose paper passed workshop review and whose Automated Reviewer reached 69% balanced accuracy against human decisions (Nature, 2026).
Novelty: Moderate; feeds the Verification Gap by producing artifacts no one wrote.
Alternatives: Genetic programming redescribed; human curation hidden in the loop.
Falsification: A deployed system whose current version has no human-written commit.
Proposed tests: Monitor.
Milestone: 2027–28.

EON-DR-26 · P3

Commons Decoherence

priorityE1

Commons Decoherence

Definition · Evidence · Tests

Definition: A regime in which individually adaptive interpreters improve individual prediction or decision accuracy while degrading a population's capacity to coordinate on a common description of shared facts and of what would happen if it acted together, under the same evidence and aligned objectives. Two measures: the Coherence Index (mean pairwise representational agreement on verifiable items, conditional on personalization) and the dissociation between individual accuracy A_i and joint regret J.
Precedents: Filter bubbles and polarization research; the 2023 platform experiments that changed exposure without changing attitudes; personalized persuasion studies (2025); sycophancy and memory effects (2025–2026); AI mediation that increased common ground (Tessler et al. 2024) — the counterexample that makes fragmentation contingent.
Novelty: The dissociation claim — coherence and joint control can fall while accuracy rises — and the prediction that standard remedies aim at the wrong variable. The divergence itself has not been measured by anyone.
Alternatives: Pre-existing attitudes; topic selection; measurement of opinion rather than description; extra communication alone restoring joint performance; personalization raising both accuracy and coordination.
Falsification: (1) Pairwise mutual information of descriptions falls with personalization at equal or better accuracy. (2) Joint regret rises while individual accuracy improves in the same preregistered task family. (3) A minimal shared counterfactual checkpoint recovers much of joint performance without eliminating individual gains. Falsified if divergence is absent or does not grow with personalization, if coordination improves, or if ordinary extra communication does as well as the checkpoint.
Proposed tests: Programme P3: an audit variant (200 synthetic personas, 50 verifiable questions, four assistants, with and without memory) and a human variant (300 participants randomized to personalized or control interpreters for four weeks; weekly identical questions; final pairwise coordination task; accuracy scored separately); a computational benchmark with a common world, permitted communications, individual and joint outcomes all observable.
Milestone: Day 45: audit result; day 75: human-panel interim read.

EON-DR-27 · —

Intention Capture

monitorE1

Commons Decoherence

Definition · Evidence · Tests

Definition: A closed loop in which an environment acts on a decoded intention before the person has formulated or reported it, so that the person's role shifts from issuing commands to ratifying actions already underway. The measurable criterion is decode-before-report latency in a closed loop.
Precedents: Inner speech decoded in real time from motor cortex, including unintended inner speech (Kunz et al. 2025); a non-invasive neuromotor wristband generalizing across users (Kaifosh et al., Nature 2025); speech neuroprostheses at >97% word accuracy and in near-daily independent home use (2024–2026); error-related-potential decoding for robot correction (Salazar-Gomez et al. 2017); non-invasive typing decoding at 32% character error rate for MEG (2025).
Novelty: The change in the locus of agency is the novel claim; the components are established, the pre-volitional loop is not.
Alternatives: Decoding that merely anticipates an already-formed intention by milliseconds; artifacts of report latency; shipped devices remaining request–response.
Falsification: In a consumer-grade loop, action precedes report by a margin that survives report-latency controls. Weakened if decoding precedes report only under narrow laboratory conditions.
Proposed tests: Monitor; a latency-study protocol would be drafted with a partner laboratory if one is identified; none exists yet.
Milestone: 2027: decode-before-report in a consumer-grade loop, if any.

EON-DR-28 · Laboratory protocol track

Closed-loop interneural integration

holdE1

Unaffiliated entries

Definition · Evidence · Tests

Definition: Two nervous systems coupled by a bidirectional interface with loop latency τ exhibit closed-loop interneural integration if, for τ below the brain's integration window, a measure of causal integration over the dyad exceeds its maximum over either brain alone by an amount that vanishes when the loop is opened, delayed or replaced by a sham, and that is not explained by shared stimulus.
Precedents: One-way brain-to-brain transfer at one bit per trial, 81% accuracy (Jiang et al. 2019); inter-brain synchrony (Dikker et al. 2017; meta-analysis of 41 teamwork studies, 2024) and the demonstration that standard measures find coupling between brains that never interacted (Burgess 2013); hyperscanning's own call for causal designs (Markus & Shamay-Tsoory 2024); non-additive simultaneous speech-and-gesture signals within one brain (2026); a 2026 review of ultrasonic brain–computer interfaces (preprint).
Novelty: A criterion and a set of discriminating controls for a question that fiction has words for and science does not; a clean negative result would be as valuable as a positive one.
Alternatives: Shared stimulus; ordinary communication; volume conduction; spurious synchrony from the measures themselves; analysis flexibility.
Falsification: No integration measure exceeds the parts at any latency (the expected null for passive, high-latency interaction); a threshold in τ would be the positive result. Falsified as a hypothesis if effects persist with the loop open or are fully explained by shared stimulus.
Proposed tests: Phase 1 (in-house): a measurement protocol validated on simulated coupled dyads and applied to archived hyperscanning data with pseudo-pair and stimulus controls. Phase 2 (partner, animal): bidirectional recording and stimulation with a latency sweep. Phase 3 (partner, human, non-invasive): only if Phase 2 shows a threshold.
Milestone: Day 90: validated protocol published regardless of outcome; partner decision.

EON-DR-29 · —

Infrastructure as instrument

monitorE3

Unaffiliated entries

Definition · Evidence · Tests

Definition: Every large deployed network carries measurements it was not built to take, so that infrastructure becomes an instrument whose observational capacity exceeds the sum of purpose-built instruments; the open question is governance of measurements nobody set out to take.
Precedents: Distributed acoustic sensing on telecommunication fibre (Zhan 2020); seafloor sensing on a 5,860 km transatlantic cable (Marra et al. 2022); a smartphone-based earthquake-detection network — 11,231 earthquakes in three years, alerts in 98 countries, extended to 200+ in 2026 (Allen et al. 2025).
Novelty: The general principle is unnamed; the governance question is open; the science is established.
Alternatives: None needed; the entry is a Class I observation with a policy question attached.
Falsification: A scientific detection by a network built for something else, in a new domain, within the year.
Proposed tests: Monitor; policy analysis.
Milestone: Review point: the predicted new-domain detection within the year; promotion to a programme only if the governance question becomes actionable.

EON-DR-30 · —

Bottom-up synthetic cell autonomy

monitorE2

Unaffiliated entries

Definition · Evidence · Tests

Definition: A cell assembled from non-living components that performs selection, genome replication, growth and genetically encoded division without a cytoskeleton, and whose variants compete.
Precedents: Top-down minimal cells (Hutchison et al. 2016; Pelletier et al. 2021); the bottom-up synthetic cell reported in preprint by the Adamala laboratory in mid-2026, with a ~90 kbp genome across seven plasmids and membrane-protein-driven division.
Novelty: If peer-reviewed and reproduced, a milestone in the Transformation dimension: a replicating system with no evolutionary ancestry. Currently a preprint whose authors note that plasmids must be consolidated and that the method is hard to scale.
Alternatives: Division dependent on laboratory intervention; instability over generations; non-reproduction by other laboratories.
Falsification: Peer-reviewed publication and independent replication; autonomous division over many generations without intervention.
Proposed tests: Monitor.
Milestone: 2027: peer review and replication status.

EON-DR-31 · —

Coherent evidence accessibility

monitorE2

Unaffiliated entries

Definition · Evidence · Tests

Definition: A difference in learnability caused by preserving quantum information for joint processing rather than converting every experimental output immediately into classical records, relative to a specified task, resource budget and measurement model.
Precedents: Provable exponential advantage with quantum memory for specified learning tasks, demonstrated up to 40 qubits (Huang et al., Science 2022).
Novelty: Established for the demonstrated task family; not an unknown law. A's label adds little beyond the authors' own framing.
Alternatives: Classical strategies at equal total resources matching the advantage; overhead removing it under noise.
Falsification: Advantage persists under realistic noise, memory lifetime and readout at equal resources.
Proposed tests: Monitor; small mathematical reproduction possible in-house.
Milestone: Promotion trigger: advantage reproduced under realistic noise and readout by an independent group.

EON-DR-32 · —

Certified indefinite causal order

monitorE3

Unaffiliated entries

Definition · Evidence · Tests

Definition: Experimental dependence on coherent control over the order of operations, where a fixed classical order is an inadequate account under stated assumptions.
Precedents: Device-independent certification of indefinite causal order in a photonic quantum switch, violating a Bell-like causal inequality by 24 standard deviations, with the locality and freedom-of-choice loopholes addressed and a fair-sampling assumption retained (Qu et al. 2026); an independent violation of the same inequality by about 18 standard deviations in a second photonic switch, with the locality loophole open (Richter et al. 2026).
Novelty: Established as physics and independently confirmed, with explicit loopholes remaining; no bearing on macroscopic time, people or intention.
Alternatives: A fixed-order model satisfying the actual experimental constraints; the detection loophole (fair-sampling assumption); the closed-laboratory assumption.
Falsification: Loophole-free certification.
Proposed tests: Monitor.
Milestone: Review point: a certification of indefinite causal order with the detection loophole closed.

EON-DR-33 · —

Rhythmic perceptual multiplexing

monitorE3

Unaffiliated entries

Definition · Evidence · Tests

Definition: Attention and perception sample the environment rhythmically, alternating between locations or objects within theta-band cycles, so that experienced continuity is assembled from discrete samples.
Precedents: Attention samples stimuli rhythmically at ~4 Hz per location (Landau & Fries 2012); perceptual cycles (VanRullen 2016); frontoparietal dynamics of rhythmic attention (Fiebelkorn, Pinsk & Kastner 2018).
Novelty: None; an established finding. Retained because it bears on how experience would be shared or coupled (DR-28) and on interface design (DR-27).
Alternatives: Not applicable.
Falsification: Not an EONORYX hypothesis; an established finding retained as a constraint on DR-27 and DR-28. Re-graded if the theta-band sampling account is overturned.
Proposed tests: None proposed; the finding constrains the latency designs of DR-27 and DR-28.
Milestone: Reviewed at the annual revision; no EONORYX milestone.

EON-DR-34 · —

Coherent processes in neural proteins

monitorE2

Unaffiliated entries

Definition · Evidence · Tests

Definition: Quantum-coherent processes in protein networks contribute functionally to neural computation or to the maintenance of consciousness, beyond established cases of quantum biology.
Precedents: Ultraviolet superradiance in large tryptophan networks (Babcock et al. 2024); predicted photoprotection from collective light–matter interactions (Patwa et al. 2024); 2026 measurements reporting superradiant lifetimes of no more than a few picoseconds; a microtubule stabilizer delaying anaesthetic unconsciousness in rats by 69 seconds on average (Khan et al. 2024); the decoherence-timescale objection (Tegmark 2000).
Novelty: Old hypothesis, new data. Picosecond lifetimes are what the decoherence argument predicts and do not reach computational timescales; a microtubule-binding account of anaesthesia does not require quantum effects.
Alternatives: Classical photophysics; anaesthetic effects via microtubule binding without coherence.
Falsification: Coherence times reaching functional timescales; isotope effects on function. Weakened as lifetimes remain picosecond-scale.
Proposed tests: Monitor.
Milestone: 2028: coherence times at functional timescales, if any.

EON-DR-35 · —

Neural intention and error alignment

monitorE2

Unaffiliated entries

Definition · Evidence · Tests

Definition: Decoding of goal-related and error-related signals from non-invasive recordings to align a machine's actions with a person's intentions in shared control.
Precedents: Real-time error-related-potential decoding to correct a robot (Salazar-Gomez et al. 2017); non-invasive typing decoding at 32% (MEG) and 67% (EEG) character error rates (2025).
Novelty: Established as a method; the 'field' framing adds nothing. Merged into the Intention Capture context.
Alternatives: Not applicable.
Falsification: Not an EONORYX hypothesis. Re-graded to E3 on independent replication of error-potential decoding for closed-loop correction at the reported accuracies.
Proposed tests: None proposed; informs the DR-27 latency protocol.
Milestone: Promotion trigger: independent replication reported.

EON-DR-A1 · —