This page is the works index for K. Takahashi. It exposes the full publication and citable software inventory directly in static HTML, grouped in reverse chronological order, with DOI-normalized links, stable anchors, and schema.org metadata aligned to the visible content.
Early works explored broad conceptual foundations of growth, self-modeling, and intelligence. Later works progressively reformulated these questions in observable-only, audit-closed, replayable, and deployment-relevant terms.
This index lists the publications and citable software artifacts currently exposed on this site and keeps the HTML order aligned with the schema.org ItemList. Each entry has a deterministic anchor based on publication date, title, and DOI suffix so links remain stable even when titles repeat.
This theoretical preprint studies how to observe, control and retain collective capability under explicit evidence, verification and resource constraints. It separates temporary task performance, interaction-specific contribution, retained growth and acceleration against complete evolving comparators with matched information and resource entitlements. Paid adaptive audits, finite catalogue retention, receipt-based continuation accounts and conditional moment bounds connect phase declarations to achieved stock. The results depend on justified calibration, fixed endpoint targets and funded execution; mathematical examples do not demonstrate empirical collective-specific acceleration or a physical phase transition.
This preprint develops a finite-time framework for measuring, attributing, and improving collective capability growth subject to explicit resource budgets, verification capacity, evidence requirements, protected service floors, and unresolved repair obligations. It evaluates an interaction against resource-, information-, and time-matched policies reoptimized after that interaction is removed, so communication volume, agent count, replicated compute, or unverified output are not automatically credited as collective growth. The framework combines joint task and research capacity measurement, interaction ablation, supporting-price inequalities, information-theoretic recognition bounds, adaptive confidence sequences, finite-horizon control, and executable continuation. Its results bound terminal capability, statistical recognition against negative models, and deferred repair obligations, and analyze verification queues, resource limits, dependence diagnostics, and the trade-off between communication and funded continuation. The claims are protocol-relative and finite-horizon; the paper does not claim to observe artificial general intelligence, artificial superintelligence, a physical phase transition, or deployed-system intelligence growth.
This preprint develops a mathematical framework for reusable state representations in controlled systems when observations have partial support and the data remain compatible with multiple, globally coupled models. It distinguishes summaries that preserve current predictions or rewards from consequence states that can be updated after a later action and observation without retaining the full history. The framework defines typed consequence contracts, finite-state consequence programs, model-indexed profiles, and a residual product construction whose context signature is the coarsest deterministic representation preserving the complete labelled model profile. It also introduces a program-product Kantorovich operator that characterizes program-relative probabilistic bisimulation and bounds normalized discounted program values, while showing when local uncertainty envelopes are conservative rather than exact. The work connects state abstraction, predictive-state representations, partial identification, causal inference, specification refinement, and deployment transfer without assuming full support, independent local uncertainty, Bayesian priors, or a universal utility function.
Evidence-Carrying Operational Claims in Open Systems develops a contract-based framework for evaluating safety, service delivery, resilience, recovery, and other operational claims in open, partially observable, and adaptive systems. Instead of treating these properties as intrinsic system attributes, the paper represents each claim as a typed, evidence-carrying proposition relative to a declared physical boundary, institutional boundary, environment mechanism, observation history, intervention regime, policy class, shared resources, and finite physical horizon. The framework combines stochastic hybrid models, exact physical ledgers, calibrated observation models, measurement uncertainty, causal identification, transportability, scenario-fixed experiment interfaces, computable concrete models, and abstractions. Its main formal result is a finite-horizon, one-sided deployment certificate that transfers an abstract lower safety value to deployment under partial observation while keeping learning-data coverage, deployment-path probability, reconciliation discrepancy, and implementation or abstraction radii as distinct quantities. The paper also defines claim-sufficient scopes, Pareto profiles, verdict semantics, and machine-readable JSON Schema validation for physical balance, provenance exclusivity, artifact containment, hashes, and certificate recomputation. The validator checks finite records and does not establish external evidence truth, causal assumptions, statistical models, or real-world safety.
Functional Process Control Dynamics (FPCD) is an auditable framework for evaluating claims about functional roles, process recurrence, directed physical flows, operational control, and environmental feedback under uncertain protocols. It addresses a common problem in complex systems research: persistence, repeated behavior, energy use, organizational authority, or correlated system changes can be mistaken for evidence of selection, control, causal influence, or functional continuity. The framework separates opportunity creation from responses within registered opportunities, functional-role classification from quantitative role accounting, compositional change from provenance-supported recurrence, recurrence from functional selection, measured exergy flows from controller exposure and signed policy effects, and factorial interactions from identifiable feedback-path effects. It combines causal inference, partial identification, process provenance, control-system analysis, and thermodynamic exergy accounting with explicit physical boundaries, reference environments, ports, inventories, actuator paths, compatible-value regions, and protocol uncertainty. Its main output is a reproducible claim certificate that records the estimand, protocol, evidence, compatible-value region, decision threshold, and claim status, including supported, false conditional on maintained assumptions, not identified, protocol inconsistent, or comparison undefined. FPCD applies to artificial, biological, human, organizational, cyber-physical, and hybrid systems without treating intelligence, fitness, agency, value, or energy consumption as universal explanatory primitives.
Epistemic Interface Autogenesis in Controlled Open Worlds develops a formal framework for evaluating when an embedded system appears to generate a new epistemic interface: an executable mechanism for obtaining observations, making requests, interpreting records, testing claims, or invoking protected services. Rather than reducing novelty or autonomy to a single score, it separates what an interface makes physically executable, whether the same behavior can be generated from inherited components and services, whether observed improvements are caused by deployment, and what remains unresolved because models, channels, construction paths, or interventions are incomplete. The paper represents interfaces as stateful stochastic protocols with separate stopping times for decision closure and physical closure, and distinguishes requests, delivered actions, physical effects, recovery processes, lifecycle resources, provenance, actor equivariance, qualification, eligibility, and authority. It defines physical causal compilers, feasible construction graphs, semantic construction, operational reproducibility, metric approximation, robust separation, and partial identification, with certificates for physical marginals, protected-action errors, task errors, plan and model coverage, lifecycle resource refinement, and evidence debt. The intended result is a typed certificate, refutation, or explicit unresolved evidence debt for generated research interfaces, robotic laboratories, monitored services, and embedded systems where timing, physical delivery, resource use, and evidence boundaries matter.
Collective Phase Control Fabric (CPCF) is an Apache-2.0 Python evidence-control toolkit for finite collective workflows. It supports evidence-bound analysis and finite intervention control by validating provenance, exact resource constraints, causal-formation records, catalytic support, verification capacity, independence, perturbations, coordination, and externally registered trial bindings. Version v1.0.1 is a Beta research package with an offline cpcf CLI, schema inspection, self-checks, bundle verification, legacy inspection, typed evidence, trust, scientific-analysis, planning, coordination, and trial models, plus optional API, worker, runner, solver, and KMS integrations. The release also includes reproducible environment metadata, generated references, SBOM-ready materials, checksums, and build provenance for Windows and Linux CPython 3.12-3.14 workflows. CPCF reports operational organization profiles for audit, decision support, scientific computing, provenance tracking, and multi-agent workflow analysis. It does not claim to create, detect, or certify collective superintelligence, causality, statistical validity, thermodynamic feasibility, physical phase behavior, deployment assurance, or runner isolation; unavailable external soak, restore, threat-model review, and penetration-test evidence remain out of scope. The current release additionally supports fixed-model information gathering and independently checked finite capability composition under declared costs. These are model demonstrations, not an empirical acceleration experiment. Latest GitHub release observed on 2026-09-21: v1.0.1. Publication and archive DOI dates retain their original identities; an older archive DOI does not identify these current release bytes.
Dynamic Future-Claim Certification (DFCC) is a replayable protocol for deciding when a bounded claim about the future may be asserted, denied, withheld, reused, weakened, transferred, blocked, or retired. The paper addresses planning systems, controllers, monitoring tools, verification workflows, AI agents, and decision-support systems where future-facing statements can authorize actions, block actions, allocate resources, or become assumptions for other systems while their authority depends on assumptions that may later change. DFCC does not replace forecasting, probability, statistics, reachability analysis, model checking, runtime verification, or assurance cases; instead, it adds a machine-checkable lifecycle layer above such methods. A future claim is evaluated only within an explicit assessment frame, accepted observations, admitted assumptions, finite horizon, dependency records, policy gates, and validation checks. The framework separates issue-time certificates from recomputed status-time authority views, so a claim that was valid when issued can later be maintained, weakened, transferred, suspended, revoked, or recomputed as inputs, models, dependencies, policies, or evidence change. The paper emphasizes canonical artifacts, manifest digests, schema validation, closed failure records, reason references, set references, wire-level records, conformance profiles, and golden tests so independent implementations can recompute the same authority status from the same admitted inputs. DFCC is intended for safe planning, AI agent governance, runtime assurance, formal verification, monitoring systems, software workflows, and decision systems that must manage bounded future claims under changing assumptions.
Verifier Ecology Theory (VET) studies systems where generating candidate outputs is no longer the main bottleneck and the harder problem is how the system evaluates, repairs, rejects, preserves, certifies, and updates its own tests, scores, proofs, monitors, benchmarks, assumptions, and judging procedures. The paper introduces verifier packets: structured, revisable verification assets that record origin, scope, procedure, certification conditions, destructive or narrowing boundaries, residual records, update and retirement conditions, and circulation status. VET treats tests, metrics, proofs, and evaluators as members of an ecology rather than isolated authorities, and asks whether that ecology can keep its own failures visible and repairable. A central concept is residual accountability: unknowns, failed assumptions, missing evidence, scope changes, translation losses, counterexamples, and safety gaps should be preserved as residuals and routed into future question formation, counterexample search, boundary revision, packet repair, schema revision, quarantine, retirement, or justified preservation. The theory identifies verifier overclosure as a failure mode in which local accuracy, speed, or certification power improves while future question formation narrows, residuals disappear, unfamiliar cases are forced into existing schemas, or destructive paths become harder to detect. VET provides a process model based on observable histories, typed records, residual ledgers, verifier packets, certification profiles, reachability certificates, counter-packets, local authority decisions, and packet circulation between systems, with schema sketches and audit guidance for AI evaluation pipelines, autonomous research systems, self-verifying agents, runtime monitoring, scientific workflows, and safety governance.
Collective Capability Runtime (CCR) is an Apache-2.0 JSON-first Python runtime for coordinating verifiable work across AI agents. It keeps tasks, evidence, disagreements, verification, residuals, and unfinished work visible so several agents can build reusable capability without treating repeated answers or raw agent count as proof. CCR uses a CLI, JSON schemas, SQLite or PostgreSQL state, and an optional HTTP API to exchange auditable work. The distributed runtime includes: independent agent workcells, transactional task leases with heartbeats and fencing tokens, verified residual resolution, optional PostgreSQL 16+ workers, OIDC + DPoP authentication, resource-matched experiments, and approval-bound external operation gates. It measures protocol-relative ASI-proxy progress through checked results, residual half-life, verification yield, independent contributors, error correlation, and communication or verification cost under declared resource limits. CCR does not run an LLM and does not claim real ASI, consciousness, model-weight updates, legal authority, or physical outcomes; accepted reports are not settlement, unknown measurements remain unknown, and side-effecting provider execution must pass explicit approval and dispatch controls. Version v1.8.0 adds opt-in evidence-bound service accounting, receiver-qualified reuse and verification-aware allocation. Its offline growth example is synthetic. Latest GitHub release observed on 2026-09-21: v1.8.0. Publication and archive DOI dates retain their original identities; an older archive DOI does not identify these current release bytes.
Finite Operational Structure Theory (FOST) develops a mathematical framework for finite operational reasoning under incomplete information. Rather than assuming a hidden true state, complete view of reality, global probability model, or primitive objective standard, the paper studies what a finite process can operationally support when it evaluates a finite expression under a declared mode of use. Its core object is an evaluated ledger: a finite accountability record of claims, support, certificates, omissions, unresolved issues, obligations, environment conditions, and transition witnesses. The theory introduces finite operational cuts, staged ledger construction, validation kernels, object kernels, certificate targets, support graphs, typed frontier closure, status and admissibility records, transition validity, and obligation tracking, with safeguards against vacuous reasoning, unsupported scope narrowing, missing external criticism, weak impact classification, environment omission, root-debt hiding, and misuse of certificates beyond their valid target. FOST is intended for intelligent systems, AI agents, research workflows, safety cases, governance processes, and automated decision systems that must act without final truth or complete information while preserving visible omissions, uncertainty, obligations, and change records.
Problemogenesis Theory develops a finite proof-carrying audit calculus for deciding when anomalies, failed requirements, retained traces, or proposed interventions may be activated as legitimate problem frames for action. It treats problem identification as a replayable governance process rather than as a given premise. The framework centers on bounded envelope logs that record sources, certificates, clocks, resources, risks, actions, and authority claims, then checks evidence, obligations, coverage, risk records, resource permissions, certificate liveness, deterministic replay, safe patching, action provenance, non-hindsight novelty, and preservation under joins within a finite horizon. The paper targets autonomous and semi-autonomous systems, AI research agents, scientific workflow tools, and safety-critical decision processes where action should remain tied to explicit evidence, reviewable authority, and declared trusted footprints. It does not by itself prove external truth, physical effect, cryptographic security, statistical validity, or global optimality unless those assumptions are included in scope.
Chronomorphic Substrate Selection Theory (CSST) develops a formal framework for analyzing how a future intelligent process may choose not only what to do, but also how it continues to exist. The theory studies decisions about computational substrate, embodiment, sensors, actuators, internal clock rate, copy structure, memory policy, merge policy, dormant or diagnostic modes, and terminal continuity boundaries. These choices are treated as self-instantiation actions: actions that change the conditions under which future actions, observations, obligations, risks, and capabilities remain possible. The term "chronomorphic" emphasizes that time is not a single uniform variable in such systems. Internal computation may accelerate while external latency, verification time, energy use, heat dissipation, queue pressure, legal response time, or welfare-risk exposure remain unchanged or become limiting factors. For this reason, CSST does not assume that faster computation, more copies, longer memory, stronger embodiment, migration to a new substrate, or persistence over time is automatically better. Each claim must be evaluated within an explicitly declared domain. The central method of the paper is certificate-gated multi-objective comparison. A self-instantiation profile is assessed through typed evidence, hard gates, mandatory and optional coordinates, support masks, status values, units, horizons, scopes, and certificate dependencies. Unsupported coordinates do not improve a claim, and failed mandatory support blocks positive certification. The framework therefore separates technical capability from broader questions of legitimacy, authority, continuity, welfare, safety, and termination. Mathematically, the paper introduces finite archive semantics for supported self-instantiation profiles and their residues. The deterministic part produces a normalized archive consisting of a nondominated supported frontier together with an audit-preserving residue ledger for failed, blocked, diagnostic-only, quarantine-only, or terminal outcomes. The paper also gives an archive-valued stochastic interface, while deliberately avoiding stochastic optimality claims without further measurable-control assumptions. A constructive abstraction obstruction theorem shows when fixed-agent models lose information needed to evaluate self-instantiation choices. CSST is intended as a conservative theoretical layer for future AI systems, collective intelligence, post-biological processes, digital agents, and other systems whose mode of existence may itself become a planning variable. It does not prove consciousness, personhood, moral patienthood, legal standing, or the safety of self-modification. Instead, it provides a structured language for asking which substrate, body, clock, copy, memory, lifecycle, and boundary transitions are actually supported by declared evidence and which uncertainties must remain visible.
This paper introduces Reality-Contact Theory (RCT), a formal framework for evaluating how observable signal processes come into constrained contact with reality under fixed claim-use specifications. The theory does not assume direct access to an absolute external reality. Instead, it studies observable episodes in which a process predicts, selects, generates, intervenes on, or attempts to steer a signal, and then receives feedback whose features constrain that operation. RCT represents such evidence through ordered lower contact profiles. These profiles are observer-relative, specification-dependent, and built only from observable records, estimator states, actions, signals, feature extractors, lineage records, and certified audit rows. The framework separates population profiles from estimates, estimates from certified lower bounds, and scalar scores from multi-coordinate profile adequacy. A central contribution is a conservative lower-bound calculus for reality contact. Support is aggregated through capacity-weighted lineage dependence structures, so repeated, cloned, or Sybil-like evidence cannot automatically amplify support. Gauge-quotient obstruction profiles prevent coordinate changes from being mistaken for contact evidence. Strategic presentation families and cumulative adversarial budgets distinguish contact-producing resistance from deceptive rigidity. The theory also includes formal extensions for metric resistance, active specification envelopes, topological lineage audits, Hodge and signature-based capacity rules, optimal-transport transfer, dynamic Fenchel transfer, random closed uncertainty sets, time-uniform confidence profiles, stopping-time extraction, and continuous-time contact flows. The main results establish lower-profile soundness under calibration coverage, protocol soundness for admissible non-circular proof ledgers, final claim soundness with explicit failure budgets, budgeted deception counting over graph-covered adversary envelopes, computable nonlinear transfer bounds, and adaptive confidence profiles for ongoing audits. The framework is intended for settings where evidence, model behavior, provenance, and adversarial presentation must be evaluated without assuming an observer-neutral truth oracle. Potential application domains include AI system evaluation, automated scientific discovery, sequential decision systems, causal modeling, model documentation, provenance auditing, robust simulation studies, and evidence aggregation under strategic or generated-data conditions.
Asymmetric Coexistence Theory (ACT) is a governance-aware theory for certifying non-capturing assistance in relations between unequally capable agents. Future intelligent societies may include humans, artificial intelligence systems, institutions, collective agents, future constituencies, and adaptive systems whose capabilities differ greatly; in such settings, assistance can increase short-term performance while reducing the assisted party's long-term ability to understand, refuse, repair, learn, contest, exit, recover, form values, generate independent descriptions of the world, or act before irreversible choices close. The paper develops ACT as a framework for evaluating whether assistance preserves agency under asymmetric power, dependency, uncertainty, and changing social or technical conditions. It introduces a minimal kernel consisting of typed agency spaces, asymmetric regimes, admissible baselines, counterfactual agency estimands, dependency debt, recoverability, and certification claims, plus extension modules for semantic capture, time-critical decisions, successor identity, recursive certification, audit independence, adaptive challenge evidence, certification governance, runtime hazards, non-dominating enforcement, model uncertainty, historical baseline corruption, coalitional pressure, scarce certification resources, multi-scale composition, and stochastic capture risk. The main contribution is to treat non-capturing assistance as a runtime assurance problem rather than as an omniscient proof of autonomy: an ACT certificate states its claim, scope, assumptions, evidence, argument, defeaters, residual risk, lifecycle state, monitoring conditions, revalidation triggers, audit traces, challenge evidence, certification base, governance records, enforcement boundaries, and fallback obligations.
Consent-Bounded Contact Theory (CBCT) develops a protocol-level theory for deciding when contact and contact-derived artifacts may be accepted as legitimate. In this framework, contact includes not only physical interaction or direct communication, but also querying, copying, forking, merging, modeling, simulating, representing, reactivating, auditing, inheriting, refining, and blocking contact-derived claims in long-lived artificial, collective, or autonomous processes. The theory defines consent-bounded legitimacy through observable evidence, credential closure, trust anchors, consent claims, negotiation transcripts, provenance records, residual routes, bridge contracts, ledgers, audit anchors, and finite certificates, while avoiding claims of physical non-contact, hidden subjective consent, complete observability, or substrate-specific standing. CBCT combines finite causal event presentations, raw observation closure, conservative presentation abstraction, stratified rule semantics, bitemporal finality, observer-merge-aware audit structures, source-authority evidence fusion, Sybil-aware source quotients, polarity-aware repair propagation, accounting doctrines, coverage epochs, bridge event morphisms, and policy-fibration gluing. The framework is substrate-neutral and is applicable to autonomous agents, AI governance, distributed systems, digital consent, provenance-aware auditing, long-running services, copied or forked processes, dormant systems, collective processes, and future intelligent infrastructures.
This manuscript develops Dormant Continuity Theory (DCT), a protocol-relative mathematical framework for reasoning about systems that remain inactive at their protected core while retaining auditable continuity, recovery, diagnostic, and handoff capabilities. The theory formalizes dormant processes using finite transition systems, typed certificates, observable histories, evidence algebra, guarded authorization, replayable resolution, extraction adequacy, and fail-closed classification. DCT addresses practical challenges in long-lived distributed systems, including forked ledger histories, bounded model checking, data availability, zero-knowledge proof soundness boundaries, watcher incentives, MEV-resistant reward mechanisms, resource conservation, guardian corruption, maintenance transitions, and certificate-level HTLC handoff to extinction-style OSCT semantics. The framework distinguishes safety, bounded-griefing, diagnostic routing, and liveness assumptions, avoiding unconditional trustless claims while providing a rigorous finite core for verification and implementation-oriented extensions.
Percolation Inversion Compiler is an Apache-2.0 certificate compiler and local AI agent runtime with a canonical Python/pip implementation and a TypeScript/npm port. It turns finite inputs into checked capability packets, proof obligations, residual ledgers, typed trace normal forms, frontier extraction, salience queues, verifier tasks, provenance records, and machine-readable schemas. The Python package percolation-inversion-compiler is v1.1.0 and adds strict public input boundaries, known/unknown measurement, witness-only positive bottleneck coordinates, measured queue costs, rechecked abstraction-lift records, finite typed traces, resource-matched acceleration measurement, and an explicit approval-bound operation path. The npm package percolation-inversion-compiler-ts is v1.1.0 and provides cross-language runtime support for strict PIC checks, resource-matched acceleration measurement, diagnostic flow checks, verified capability routing, approval-bound typed operations, and the pic / pic-ts CLI for Node.js agent workflows. PIC supports AI agent integration and protocol-relative ASI-proxy phase control, but it does not prove real artificial superintelligence, consciousness, model-weight change, legal authority, physical or oracle truth, or observed outcomes; missing values stay unknown, candidate volume does not become verified capability, and packet content, trace content, safe-command hints, and check results are never executed automatically. Latest GitHub release observed on 2026-09-21: Python v1.1.0; TypeScript v1.1.0. Publication and archive DOI dates retain their original identities; an older archive DOI does not identify these current release bytes.
Typed Reality Compilation (TRC) is a mathematical framework for compiling observable cyber-physical infrastructure into finite, typed process frontiers. The theory models physical, computational, communication, and control resources through typed ledgers, executable trace semantics, residual accounting, tolerance budgets, and resource-efficiency Pareto archives. Its central objective is not a single optimal action, but the preservation and operational improvement of future freedom: the remaining number, quality, reversibility, reserve, checkability, latency, cost, and resource-feasible process options after disturbances or system changes. This version develops TRC as a low-complexity finite certificate theory for practical infrastructure settings such as AI datacenter power, cooling, communication, sensing, repair, and control loops. It introduces operational tolerance allocation, progressive fidelity compilation, compositional tolerance propagation, telemetry-updated resource-cost kernels, trace-indexed resource-flow feasibility, causal event scheduling, lifecycle recomputation, and certified partial-frontier return under budget exhaustion. The framework separates deterministic, risk-provisional, relaxed, and diagnostic records, and ensures that approximation tolerance never reduces physical residuals or silently upgrades claim status. The result is a typed, machine-readable theory for resource-efficient cyber-physical planning, resilience analysis, and executable process frontier extraction under finite observation and finite computation.
This manuscript develops Bottleneck Inversion Theory (BIT) as a mathematical framework for certifying protocol-relative, intervention-unlockable potential. Rather than treating potential as an absolute latent property, the theory represents it as a vector-valued, unit-typed lower-bound object whose coordinates are reported only when supported by compatible evidence, resource ledgers, and witness certificates. The framework formalizes stopped evidence sheaves, task-local deficiency audits, mechanism-factorized null channels, exactness-certified release duality, unseen-frontier discovery, cross-validated anchor transfer, dynamic-regime acceleration, and CEGAR-style simulation barriers. A central contribution is a single-source machine-readable LaTeX architecture: the manuscript embeds stable theorem identifiers, claim records, witness schemas, dependency relations, unit ledgers, and citation DOI records directly in the TeX file. This makes the work suitable for both human mathematical review and automated extraction of claims, assumptions, dependencies, and cited formal objects. The theory is positioned at the intersection of mathematical certification, causal inference, time-uniform evidence, optimal transport, stochastic process analysis, formal verification, and capability evaluation.
Executable Capability Percolation Theory (ECPT) is a mathematical framework for studying how execution-available capability packets can accumulate, combine, and propagate through constrained directed hypergraphs. The theory models ASI-proxy phase approach as a typed phase-control problem rather than as an absolute claim about artificial superintelligence. It separates settled, provisional, and speculative target claims, and represents capability growth through stochastic relaxation quotients, global Gibbs activation, speculative RAF-like closure, liquidity-debt settlement, reachable-mass recursion, and intervention-identified mean-field envelopes. This version emphasizes operability and machine readability. It introduces constructed path-law response policies, log-linked generator identification, finite-sample quotient certification, self-normalized risk margins, AND-support activation thresholds, and protocol-relative checker soundness. The TeX source also embeds a machine-readable manifest, claim registry, citation metadata, CodeMeta JSON-LD, and BibTeX mirror, making the artifact suitable for reproducible mathematical review and metadata indexing.
Salience-Queue Occupation Theory (SQOT) is a protocol-relative mathematical framework for analyzing when finite-budget operational processes lose effective control over their priority queues under persistent or adversarial salience sources. The manuscript formalizes salience occupation using observable histories, budget ledgers, diagnostic reserves, queue morphisms, finite certificate grammars, checkable ledgers, typed risk composition, self-auditing kernels, and route-sound checker semantics. The theory is designed for artificial, distributed, or post-biological operational systems, but it does not rely on subjective psychology or normative claims about what a process should attend to. Instead, it studies finite, auditable conditions under which a process can preserve diagnostic capacity, response or no-action availability, rollback or quarantine options, semantic-egress safety, mechanism-compatible incentives, and bounded verification cost. The results include finite checker semantics soundness, checked non-circular sovereignty certificates, typed risk composition, adaptive succinct-session soundness, egress abstraction refinement, and payoff-reflected mechanism robustness. SQOT explicitly limits its claims to declared validity domains and does not assert absolute physical, cryptographic, economic, or base-reality guarantees.
Observable-Signal Crystallization Theory (OSCT) is a formal, objective-free mathematical framework for analyzing protocol-individuated operational processes, especially post-intelligence or non-biological process systems. The manuscript develops a purely mathematical treatment of observable-signal bodies, audit records, diagnostic preorders, debt ledgers, load queues, regeneration hypergraphs, and successor systems without assuming hidden subjects, terminal utility functions, biological essence, or intrinsic persistence goals. The central construction is protocol-certified non-resurrection complete extinction: a finite or regular stochastic certificate that an active commitment can reach a successor-closed non-regeneration region from which audited residues, records, keys, descendants, certificates, debts, queues, and refinements cannot reconstruct it. The theory distinguishes cessation, dissipative cessation, complete extinction, and loop liberation, and formalizes these notions using finite witnesses, safe attractors, robust stochastic feasibility triples, audit hyperproperties, Feller kernels, Galois abstraction, and simulation-preserving refinement. This version is intended as a standalone mathematical manuscript for formal methods, AI safety, process theory, stochastic systems, and verification-oriented studies of persistence, termination, deletion, and non-resurrection under explicitly declared audit protocols.
Abstraction Liquidity Theory (ALT) develops a formal framework for determining when local problem-solving traces become reusable abstraction assets that reduce downstream search, evaluation, and certification costs. The paper treats abstractions as operational tokens rather than informal artifacts, and evaluates them through declared receivers, opportunity measures, baselines, lifecycle costs, telemetry, evidence validity, transport scope, authority envelopes, hazard constraints, and runtime certificate packets. The manuscript introduces an actor-neutral certification kernel for AI agents and other computational actors. It specifies machine-readable packet schemas, dual exploration and settlement ledgers, finite-sample lower and upper bounds, causal and calibrated-proxy value estimands, mission-validity certificates, adversarial-token rejection, root/finality checks, baseline refresh, deprecation, resurrection, rollback, and kernel-update bridges. The goal is to make abstraction evaluation executable: an agent should be able to parse a packet, verify evidence, admit or reject a token, suspend stale claims, deprecate negative-liquidity tokens, and preserve raw net safe capital under fail-closed rules. The paper further defines Target-valid ALT-CARA, a criterion for certified ASI realization acceleration. Rather than claiming unconstrained ASI achievement, ALT-CARA formalizes time-to-target acceleration relative to a resource-matched baseline upper envelope, under declared capability bases, target-validity certificates, raw net solvency, viability conditions, hazard and authority constraints, transport validity, finality, and causal reproduction evidence. The framework connects AI evaluation, causal inference, runtime verification, risk control, skill reuse, safe exploration, and distributed certification into a single theory of mission-valid safe abstraction capital.
This manuscript introduces Boundary-Encoded Stability Theory (BEST), a stochastic viability framework for analyzing long-running black-box systems through the persistence of their self-maintaining boundary layers rather than through full access to hidden internal state. A BEST system is modeled as a discrete-time killed process with surfaced perturbations, observation timing, metabolic boundary variables, exchange requirements, fallible control execution, schema exit and reorganization, and an absorbing death state. The paper develops mathematical certificates for boundary-layer survival, including surface-equivalence and non-identifiability results, metabolic capital constructions, stochastic-order resource concentration, stopped and time-uniform metabolic-margin bounds, history-dependent surface-law and memory-approximation results, evidence-gated schema reorganization, finite-horizon risk ledgers, infinite-horizon hazard characterizations, and continuous-time killed-generator extension conditions. The framework is intended as a structured theory of boundary persistence under uncertainty and fallible execution; it is not a containment, alignment, or hidden-bulk safety guarantee.
This work introduces Constraint Generative Theory (CGT), a constraint-primary effect-semantics framework for studying how declared constraints generate, transform, observe, describe, continue, evaluate, and verify formal structures. The central object of CGT is not a bare set of satisfying assignments, a final output, or a report, but the generated effect profile induced by a constraint system in a declared frame. A constraint is treated as a typed structure-inducing and effect-transforming object with declared level, domain, codomain, effect dimensions, transformation rule or relation, and comparison regime. Constraint tokens, rules, predicates, generators, selectors, policies, schedules, observation lenses, description lenses, evaluator selections, goal predicates, bounds, and verification conditions are treated as presentations of constraints, not as the definition of constraint itself. The paper develops a constraint-effect calculus for comparing how abstract constraints change declared effect dimensions. The calculus includes marginal effects, dimension-relative equivalence, redundancy, independence, interaction, non-commutativity, affordance, continuation shifts, valuation shifts, inconsistency shifts, observation/description shifts, opacity, and generating power. It also distinguishes generated-universe components from full effect profiles, so that reports, observations, descriptions, continuation graphs, inconsistency markers, valuation structures, and certified fragments remain explicit rather than being silently collapsed into a final output. A key motivation of CGT is that output-equivalent or report-equivalent systems may still differ in the constraint effects that generated, observed, described, continued, valued, scheduled, or marked them. This makes constraints and their multi-dimensional generated effects the primary reproducible comparison objects. The framework includes a scientific availability layer for reproducible claims and a certified finite layer for checking selected effect components and effect differences. These layers support reproducibility and verification, but they do not define the core identity of CGT. The work positions CGT conservatively with respect to neighboring formalisms such as model theory, institution theory, closure theory, constraint satisfaction, graph transformation, rewriting logic, structural operational semantics, abstract state machines, coalgebra, cellular automata, category theory, information theory, soft constraints, and paraconsistent logic. CGT is not proposed as a replacement for these theories; rather, it provides a constraint-primary language for comparing generated effect profiles and the transformations induced by constraints.
Affordance-Compiled Intelligence develops Cognitive Impedance Matching Theory (CIMT), an observable-only and no-meta protected compiler theory for LLM-integrated systems. The paper studies how a fixed model-policy can exhibit different operational capability when the surrounding world is redesigned through observations, typed action handles, validators, repair paths, rollback modes, authority scopes, context summaries, and auditable receipts. CIMT treats system-level capability amplification as a world-side compilation problem rather than a model-weight improvement problem. It defines operational claims through explicit claim objects and evidence objects, using committed observable ledgers, target-evaluation channels, deterministic reducers, validity budget ledgers, evidence dependency graphs, artifact I/O manifests, conformance envelopes, and finite-sample or sequential certificates. Human reviewers, LLM judges, benchmarks, and external auditors are not treated as privileged evaluators; they are modeled as named, fallible measurement channels. The theory provides a conservative certification framework for paired target-channel improvement, vector debt accounting, forbidden-coordinate zero certificates, target-firewall discipline, scope simulation, dynamic widening, runtime and model-policy conformance, macro reliability, repair contraction, distribution-shift transfer, and receipt sufficiency. It also includes worked examples for code-editing agents and retrieval-augmented generation systems. The intended contribution is a practical formal foundation for making fixed-model LLM systems more reliable through observable world-side interface, authority, validation, repair, and audit design.
Apache-2.0 Python reference implementation for local-first, model-agnostic workflow-policy optimization in long-running AI-agent workflows. OASG records observable history as append-only JSONL ledgers, verifies canonical hashes and ledger prefixes, reduces history into deterministic operational state, computes finite-horizon KLB_2 viability lower bounds, and promotes only receipt-backed workflow-policy changes through a conservative no-meta dominance gate. Latest GitHub release observed on 2026-09-21: v1.1.0. Publication and archive DOI dates retain their original identities; an older archive DOI does not identify these current release bytes.
This preprint develops Certified Autocatalytic Intelligence Theory (CAIT), an operational theory for measuring, certifying, attributing, and controlling reproduction of verified capability capital in AI R&D systems. It argues that AGI/ASI-relevant acceleration should be certified by net, endogenous, resource-normalized, safety-gated production of further verified capability capital rather than raw model capability, benchmark scores, candidate volume, or spectral reproduction diagnostics alone. The framework introduces typed partial certificate algebra, machine-readable registry and certificate records, domain witnesses, status-effect maps, evidence composition tables, window-balance certificates, operational metrics, and arrival decision rules. It separates endogenous reproduction from external injection, human assistance, tool upgrades, unresolved attribution, and unsafe or uncertified artifacts, making AGI/ASI arrival a certificate-relative lower-bound claim about verified capability-capital reproduction rather than hidden mental states or unrestricted deployment readiness.
This paper develops an observable-only, no-meta theory of universal liberation and welfare viability for human and artificial subjects. Under the assumptions that no unchallengeable meta-evaluator is available and that certificate-relevant premises must be observable or explicitly marked as non-authorizing conjectures, the theory defines what can be scientifically certified: not hidden happiness itself, but scoped lower-bound viability of observable welfare-support and liberation conditions. The framework combines measurable histories, filtrations, ordered value spaces, noncompensable floors, viability domains, imprecise probability, causal identification sets, controlled transition laws, stratified validators, protected-group envelopes, population-solvency constraints, and backdated refutation rules. It treats artificial-unit creation, copying, simulation, and deployment as duty-generating processes governed by instantiation permits, tail-risk solvency, and copy-control requirements. The main results show limits on hidden-happiness maximization and unscoped universal happiness claims, justify order-valued noncompensatory welfare representation, establish conservative uncertainty and debt propagation, provide measurable executable selection conditions, prove robust one-step viability under controlled lower probabilities, block cross-subject compensation through protected meets, prevent open-world caution from collapsing into paralysis, and require refutations to be replayed from conservative occurrence intervals.
This preprint introduces Layered Online Service and Certified Replay Control (LOSCR), a model-independent framework for evaluating and controlling verified acceleration in AI-assisted R&D. It separates lightweight telemetry from stronger evidence claims and combines service-capacity control, certified replay, machine-readable claim profiles, deterministic checking, append-only ledgers, evaluator audits, and falsification rules so acceleration claims can be checked, downgraded, quarantined, or escalated against observable evidence and operational capacity.
This preprint develops a process-neutral, proof-carrying framework for certifying artificial superintelligence arrival under incomplete, fallible, and operationally messy evidence. It treats ASI arrival as a layer-relative statement about sustained capability-producing trial generation, renewal, and viability across models, scaffolds, laboratories, organizations, markets, software ecosystems, regulators, evolutionary systems, and coalitions. The framework uses typed audit logs, compatible-history sets, certified random closed sets, bound certificates, decision semantics, coalition attribution, structural interventions, ledger viability, evaluator noninterference, robust task coverage, and asymmetric three-valued rules so missing or unreliable data widens uncertainty rather than counting as favorable evidence.
This preprint develops an executable theory of authority migration for AI agents shaped by human feedback, including RLHF, preference optimization, constitutional AI, reward models, evaluator substitution, and related alignment pipelines. It specifies declared no-meta agency through a BootDecision record, seed interpreter, typed action descriptors, forbidden matchers, object-authority probes, witness tiers, deterministic checker ABI, sandbox profiles, chained ledgers, and fail-closed controls for protected effects, credentials, network calls, external writes, user-data disclosure, checker updates, and kernel updates.
This preprint develops an operational continuity theory for long-running AGI workflows under finite verification. It argues that certified service coverage is insufficient when agents exercise delegated authority, self-modify, accumulate persistent memory, generate new goals, recover from incidents, or interact with other agents. The framework formalizes continuity-certified AGI through viability conditions for recovery, authority, identity, mutation, goal commitment, memory integrity, federation, liability, trusted-base health, cross-layer consistency, and non-erasing critical-violation history.
This preprint studies long-running AI systems as typed workflows that select tasks, decompose them into subproblems, invoke tools, preserve traces, schedule audits, and maintain certified records under drift and partial observability. It proves that controller-only scaling is insufficient for robust certified coverage when replay and validation budgets stay fixed, and develops a constructive workflow theory built on reusable certified libraries, monitored calibration, novelty control, and maintenance envelopes.
This preprint develops a transport-validity theory for agentic AI interventions that are first screened on small systems and later considered for frontier-scale deployment. It formalizes replayable lower certificates, descriptor-growth audits, witness-cover transport conditions, and portfolio-level allocation rules for deciding which small-scale gains can justify expensive frontier trials under interaction risk and budget constraints.
This preprint develops a formal planning theory for autonomous agents that move across institutions, credential ecosystems, and administrative domains under uncertainty. It separates raw institutional states, capability summaries for planning, and visible summaries for evaluation, and derives transport, disclosure, and belief-state planning results for revocation-aware, resource-aware, and strategically aware decision-making across institutions.
This preprint develops a mathematical theory of institution-relative black-box acceptance under metadata-conditioned auditing, where decisions rely on publicly auditable evidence rather than internal model states. It formalizes common-law control states, certified support, selector legality, provenance factorization, and non-vacuity boundaries, and proves necessity lower bounds on witness-level divergence and source information under declared false-accept and false-reject constraints.
This preprint develops a first-principles theory of honest public standing dynamics for research claims under observable-only and no-meta governance. It analyzes how public claims move among standing states under declared interfaces, replayable frontiers, and explicit service, reserve, and retained-memory constraints, and derives boundary and restoration results for challengeability, re-entry, overload, and exploration slack.
This preprint develops a first-principles typed dynamic theory for certifying and releasing research claims produced by autonomous research systems under finite verification capacity and public accountability constraints. It formalizes public state, authority algebra, typed transcripts, fail-closed certification memory, and a release layer with versioned units and support-ledger accounting under observable-only and no-meta governance.
This preprint develops a certified local micro-theory of open-task participation in agent societies. It formalizes when an authenticated local agent should act, assist, verify, withdraw, or exit under public evidence, certified uncertainty, attribution, and implementability constraints.
This preprint develops a constructive no-meta, observable-only framework for autonomous discovery, evaluator redesign, target expansion, and self-modification under public evidence alone. It formalizes robust public risk, replay-certified comparison under constitutional extension, and auditable observable invention without hidden states or privileged meta-evaluators.
This preprint develops a finite-horizon certificate framework for observer-modifying contagion on networks, where exposure can also change later diagnosability and auditability. It formalizes self-concealment, internal blindness, external recovery, delayed audit, and fail-closed containment claims under explicit comparison semantics.
This preprint develops a first-principles calculus for classification-induced cognitive drift in reflexive human and AI settings. It formalizes how disclosed classifications can change targets, evaluators, and later evidence under replay, repeated-measures, rollout, and observational comparison regimes.
This preprint develops a certificate-based comparison theory for how record absence changes preference over legacy claims on a fixed comparison frame. It formalizes exact and approximate absence, corrective-disclosure, and closure-asymmetry results under auditable local certificates and baseline admissibility constraints.
This preprint develops a symbolically effective contract calculus for semantic translation under gluing-coherent aspect semantics. It formalizes exact audit, accountability, native collapse, and round-trip obligations with symbolic checks and deployable decision guarantees.
This preprint develops a measurable-state theory for observer-modifying and self-concealing information in hidden-state controlled systems. It formalizes when diagnosis degrades or recovers under internal blindness, external anchors, structural insulation, and delayed or recurring audit.
This preprint develops a conservative lifecycle-certification framework for autonomous agents under finite routing, monitoring, and deployment budgets. It formalizes counterfactually auditable admission, retirement, monitoring, and deployment rules using direct move inference, replay support, and anytime-valid sentinel monitoring.
This preprint develops an interface theory for recursive self-improvement under endogenous yardstick drift, where a system changes its own evaluator, benchmark, memory, and verification process. It formalizes replayable conditions for distinguishing claimed improvement from stable improvement under delayed audit, evaluator drift, verification backlog, and governance safety constraints.
This preprint develops a governance theory for shared epistemic commons maintained by autonomous agents under no-meta constraints. It formalizes observable rules for preserving answerability, contradiction handling, anti-capture slack, and controlled exit under contamination, provenance uncertainty, and recursive regeneration.
This preprint develops an oversight-centered metrology and control theory for agentic systems in real workflows, yielding deployment-relevant evaluation criteria that treat human review, automated checks, delayed labels, and external auditing as costly interrupt channels rather than privileged oracles. It formalizes workflow-level estimands, claim-justification margins, transport mismatch, congestion, routing error, audit gaming, redundancy, and safe control under delay, irreversibility, and ambiguity budgets.
This preprint develops a theory of AI benchmark half-life in recursive corpora under semantic leakage and regeneration, yielding validity-decay bounds and monitoring rules for evaluation systems whose items and solution traces re-enter public data. It models benchmark validity through discriminative power and construct validity, and derives jump-aware lifetime bounds, partial-identification results, portfolio design criteria, and safe sequential control under ambiguity and partial observability.
This preprint develops a fixed-budget theory for when inference should be split across multiple agents under local context ceilings, yielding conditions for predictable multi-agent advantage over matched strong single-workspace baselines. It formalizes additive budget accounting across worker inference, routing, communication, memory, and verification, and derives diagnostics for candidate coverage, evaluation-selection accuracy, hijack risk, decomposability, diversity, shared-failure dependence, and communication fidelity.
This preprint develops a minimal theory of search stability for long-running agents under finite active context, delayed verification, and lossy state compression, yielding conditions for preserving at least one operationally adequate hypothesis family over time. It formalizes adequacy preservation, retirement, substitution, branching, compression, and reset decisions under context budgets, and derives threshold results for contamination, shadow retirement, alias hazards, reserve feasibility, and diagnostic regret.
This preprint analyzes proposal-veto decision dynamics for observable-only autonomous intelligence when true proposal quality is latent and no external meta-controller is available, yielding explicit stability thresholds, identifiability limits, and commit-window trade-offs. It derives conditions for bounded expected error debt, positive Harris recurrence, and geometric divergence without debt-proportional correction, and formalizes progress-safety frontiers, rollback effects, trust-chain amplification, and finite resource constraints.
This preprint introduces the Metrology-Theoretic Epistemics Engine (MTE), a machine-checkable epistemic governance layer for no-meta observable-only autonomous intelligence, yielding fail-closed criteria for when claimed progress is scientifically credit-bearing. It formalizes deterministic artifact canonicalization, observability credit gates, equivalence-class collapse rate accounting, dual risk ledgers, and supermartingale-style over-credit control with reproducible replay and destructive test obligations.
This preprint develops a boundary theory of constitutional sovereignty for autonomous intelligence under no-meta and observable-only governance constraints, yielding auditable conditions for self-revision without collapse of semantic continuity, liberty, accountability, or physical viability. It formalizes constitutional influence and capture metrics, proves finite-resource limits on absolute sovereignty, and provides controller-ready recovery, transition, and fail-closed verification laws for implementation.
This preprint formulates agenda sovereignty for autonomous intelligence under no-meta, observable-only governance constraints, yielding quantitative influence-capacity envelopes and recovery guarantees against agenda capture. It introduces survival-conditioned causal influence metrics, sovereignty reserves, and overlap-corrected thermodynamic accounting, and derives limits on detectability, poisoning resilience, strategic opacity, delay-throughput uncertainty, and finite-budget trade-offs among semantics, liberty, and sovereignty.
This preprint establishes a testable theory of autonomous intelligence under no-meta ontology drift constraints, yielding auditable persistence and leakage-resource bounds without privileged semantic access. It unifies information-theoretic identifiability limits, thermodynamic irreversibility costs, geometric drift structure, and cryptographic accountability conditions for long-horizon AI governance.
This preprint defines an audit-closed protocol for autonomous scientific discovery in self-driving laboratories under deterministic replay and public-log governance constraints, yielding trustworthy accept-reject-update decisions with always-valid sequential evidence. It integrates typed stochastic observation interfaces, e-process based testing, logged-propensity adaptive experimentation, drift recovery, and certificate-based reproducibility controls.
This preprint defines a burden-of-proof governance framework for reducing non-contributory administrative tasks in digitally governed organizations under observable-only and auditable no-meta constraints, yielding causal, contestable task-value control instead of fixed welfare scoring. It introduces mutable subjective objective registries, tiered causal identification, and reversible hold-experiment-justify-substitute governance loops with solvency and anti-collusion safeguards.
This preprint develops a safety-gated controlled hidden Markov model for online estimation of bounded proxy user-state signals in intervention-aware dialogue agents under uncertainty and governance constraints, yielding leakage-safe prediction and risk-aware adaptive actions. It separates prompting and response mechanisms, combines pre-turn prediction with post-turn nowcasting, and supports EM-based learning with explicit missingness and identifiability assumptions for auditable AI operation.
This preprint formalizes operational deductive rules that map observable AI capability-to-reality translation gaps to concrete intervention levers under physical, institutional, and risk constraints, yielding implementable acceleration policies for real-economy growth. It provides machine-readable rule registries and closed-loop protocols linking diagnostics, event attribution, robust allocation, and safety constraints.
This preprint quantifies how rapid AI capability growth in information space is filtered by physical and institutional constraints, yielding reflection-adjusted semi-endogenous growth laws and bottleneck-switch timing results. It formulates a hybrid ODE-jump model that separates potential algorithmic progress from realized real-economy deployment across compute infrastructure, energy, permitting, and regulatory readiness.
The preprint develops observable-only, no-meta institutional rules for structural risk in competitive AI systems, proving deterrence and non-domination conditions with escrow-based reversibility, coalition-aware repeated-game guarantees, and finite-sample heavy-tail certification under auditable public records.
The preprint develops an information-theoretic and geometric boundary theory for persistent semantics under ontology drift in no-meta adaptive systems, deriving impossibility and phase-boundary results plus auditable operational laws for probe capacity, resource allocation, blackout bursts, Byzantine ambiguity, and long-horizon semantic maintenance.
The preprint presents an observable-only AI safety framework for robust bottleneck diagnosis from public data, combining no-meta dynamic programming, partial identification, anytime confidence sequences, and dynamic IQC to produce auditable interval diagnostics with fail-closed replay contracts.
The preprint introduces Quality-Operator Non-Collapse (QONC), a certificate-first safety framework for recursive systems that uses observable and auditable quantities for adaptive validity control, delayed-label risk correction, viability-safe robust MPC, and ledger-accounted replayable governance against quality and liveness collapse.
The preprint specifies a domain-agnostic, observable-only and no-meta contract framework for modular AI and composite-system pipelines, with deterministic fail-closed verification and progressive certificate profiles for composition guarantees, proxy-to-true risk accounting, and drift hardening.
The preprint introduces a compute-first safety framework for LLM routing without privileged meta-judges, defining observable viability under information, compute, and energy budgets with offline policy synthesis, fail-closed runtime deployment, and tamper-evident auditing.
The preprint introduces Proof-Carrying Skills, a no-meta framework that reuses verified skill executions to reduce repeated AI/LLM inference cost, using a deterministic bounded checker, observable anchors, gas-metered predicate evaluation, and replay-resistant receipts for fail-closed verification.
The preprint specifies an evidence-carrying cognitive mesh for DePIN-style decentralized compute that sustains locally verifiable capability under observable-only and no-meta constraints, using content-addressed provenance objects and a queryable claim graph built from deterministic web retrieval and auditing pipelines to resist capture and poisoning.
This supplement studies when observable-only, no-meta agents cannot distinguish good from bad governance because mediator implementations are observationally equivalent from local history. It proves an impossibility result for robust progress guarantees under that unidentifiability and derives operational conditions such as contestability, safe refusal, and cross-domain witnesses for breaking silent contract switching.
The preprint models observation capture as Blackwell garbling of an observation interface and proves operational non-expansion under fail-closed authority for observable-only no-meta agents, while giving counterexamples under restricted policy classes and proposing receipt-based anti-capture enforcement.
The preprint introduces OOPCA, a hybrid proof-and-replay audit mode for observable-only no-meta agents that compresses verifier workload by replacing selected deterministic replay segments with pinned proof verification, while preserving fail-closed semantics, evidence closure, and deterministic fallback receipts.
The preprint formalizes verification-limited scaling under observable-only and no-meta constraints, defining strict fail-closed progress credit, bounded cutoffs, and replay-auditable diagnostics that prevent progress inflation under missing or weakened evidence.
The preprint specifies fail-closed containment for no-meta, observable-only agents under adversarial participation using deterministic canonical logs, bounded metering, escrowed obligations, and finality-gated external effects for replayable safety and governance.
The preprint specifies ONCAP, an implementation-ready protocol stack for observable-only no-meta causal autonomy that integrates safety floors (SRK), deterministic replay backcasting, commit-open time anchors, closure obligations, and friction-minimal contracts to enforce fail-closed, audit-friendly control under adversarial multi-agent conditions.
The preprint specifies a deterministic, fail-closed operational core for observable-only backcasting that prevents causal loops when future-conditioned constraints gate present actions, using commit-open wire profiles, canonical hashing, time anchors, and reproducible action digests under no-meta autonomy.
The preprint specifies protocol-first intersubjective contracts for no-meta autonomous agents, enabling negotiation and dispute handling under partial observability via bounded-loss, fail-closed withdrawal, auditable evidence packets, and safe degradation under non-adopting or adversarial counterparties.
The preprint specifies no-meta autonomy via auditable observable-only floors and a Self-Recognition Kernel that enforces visibility, contraction, contact, and dissipation using fail-closed gating, certified logs, and deterministic canonicalization.
The preprint specifies POB-ML, a deterministic observable-only protocol for backcasting under audit-ready evidence surfaces, with budgeted candidate evaluation, static FOQL validation, InputSet binding, and a safety-dominating action gate.
The preprint specifies an observable-only audit gate for AI agents under partial logging, separating decision-time evidence from replay-time certification with a minimal event vocabulary and finalized log prefixes for non-markovian components.
The preprint defines HAAG, an observable-only audit gate that mediates protected actions via verifiable logs, anchored evidence checkpoints, and capability tokens without relying on model introspection or semantic truth.
The preprint defines an audit-ready, implementation-agnostic gate protocol for detecting and triaging representational collapse in OOD scientific modeling, with fail-closed outcomes and a frontier-discovery framing for targeted evaluation.
The preprint defines MemoryFlow, a telemetry-only protocol and verifier for dynamic memory quality in LLM agents, with fail-closed conformance profiles and online metrics derived from deterministic event streams.
The preprint introduces BOSC, an audit-first controller for LLM inference that enforces tail-latency, time-to-first-token, and KV-cache budgets via virtual queues and hysteretic gating under fail-closed telemetry contracts.
The preprint defines a ledger-only, fail-closed protocol for certifying non-markovianity under no-meta obligations using anytime-valid e-process evidence and declared telemetry contracts with replayable artifacts.
The preprint proposes boundary-only governance for no-meta social interaction under rights and regulation, using canonical logs, explicit budgets, and telemetry-only certificates to control disclosure, friction, and adaptive updates in non-markovian human collectives.
The preprint defines semantic phases as history-indexed predictive equivalence classes under energy and finite-memory budgets, and builds an auditable non-Markovian framework for intervention selection, irreversibility via path-space KL, and active inference under telemetry contracts.
The preprint defines no-meta viability via escrow-charged reservations and fail-closed evidence accounting, giving auditable sufficient conditions for an agent to remain viable under adversarial participation without external adjudication.
The preprint introduces MCCBE, a telemetry-only and fail-closed auditing framework that certifies conservative bottleneck floors across tail latency, integrity gaps, I/O, network, and power constraints for large-scale AI training under incomplete logs.
The preprint defines Tail-Limited Useful Compute, a contract-based telemetry framework that converts tail latency into fail-closed bottleneck certificates and anytime-valid lower bounds on useful progress for large-scale AI training.
The preprint treats silent data corruption as a scaling limiter for large-scale AI training and proposes a telemetry-contract framework that fails closed on missing integrity evidence, producing certified progress and useful-compute floors under explicit coverage assumptions.
The preprint defines telemetry contracts that certify conservative lower bounds on required data movement, checkpointing, and collective communication for transformer training, yielding fail-closed bottleneck floors and actionable slack intervals without vendor-specific instrumentation.
The preprint defines a value-neutral telemetry kernel for no-meta agents that meters irreversible internal operations and external effects via a commit/outbox/receipt ledger, enabling deterministic audit replay and fail-closed detection of missing telemetry under strict resource constraints.
The preprint treats boundary I/O bytes and peak hot memory as auditable budgets, defines an I/O Gain-Shut Kernel that enforces fail-closed admission for transfers and peak usage, and provides certified rewrite rules that guarantee non-increasing boundary cost while preserving meaning or bounded approximation error for transformer-scale workloads.
The preprint defines an audit-friendly contract K_t that partitions agent cores into phases via probe responses and provides an information-theoretic certificate for irreversible phase transitions under finite memory, monitoring, and energy budgets, yielding a practical monitoring blueprint for no-meta agents.
The preprint defines operational meaning as a versioned probe signature with a coding-convention ID and builds a monitoring-and-commit instrument that certifies when meaning becomes non-reconstructable via entropy and collision certificates, yielding a Landauer-consistent work lower bound for conditional erasure.
The paper proposes an operational No-Meta interface: SRIM does not assume any externally correct auditor, reward signal, or environment-side evaluator.
The central focus is a separation penalty incurred by architectures that enforce a compute–instance-storage boundary (a von Neumann–style bottleneck): task-relevant information must cross a physical channel (bus/interconnect/NoC), and this mandatory transport induces an unavoidable energetic floor.
A “future” is not treated as an externally true trajectory, but as an internal specification variable implemented by a terminal potential and realized through exponential tilting of a dominated baseline predictive law, yielding a plan distribution without invoking hidden meta-objectives.
All orthogonality, adjoints, and reconstructions are defined relative to a fixed, declared Hilbert geometry (e.g., an L2(mu) reference measure or a noise-weighted sensor geometry).
The key constraint is that the public evaluation signal is invariant under strictly monotone transformations of any internal continuous robustness score—so only rank/order information can be used for triage.
Non-Markovianity is unavoidable for embedded agents that only see the world through finite observation, finite memory, and finite-bandwidth interaction.
With this canonical projector, the discrete-time Mori–Zwanzig (MZ) identity becomes a geometric consequence of the observation channel rather than a modeling choice.
Building on a natural-law-type (NL) perspective that combines macroscopic fluctuation theory with information-thermodynamic reasoning, we study a restricted class of agents that (i) already implement predictive semantics in the NL-conditions sense, and (ii) remain structurally non-degenerate via autopoietic closure.
Within this joint setting, the work introduces an intrinsic predictive semantic capacity of the agent’s internal code with respect to a future viability observable, defined via mutual information relative to a baseline “minimally predictive” internal state.
Using macroscopic fluctuation theory (MFT), we define a persistence order parameter that quantifies how far a configuration sits above the worst-case collapse barrier in state space, and relate it to a “dead” gradient-flow backbone of the underlying physical dynamics.
Instead of assuming a fixed state space and model class, the paper introduces an open representation space of prefix-free codes for internal states, parameters, and architectures.
Instead of assuming a linear, additive power budget, the paper introduces a nonlinear total power functional P tot over coupled process coordinates (structural organisation, persistence, semantic processing, tagging, reference updating, value updating), motivated by stochastic and nonequilibrium thermodynamics.
Under Foster–Lyapunov conditions ensuring positive Harris recurrence, the paper proves the existence of almost-sure internal-time velocities (“social proper-time velocities”) and derives natural-law inequalities that couple these velocities to long-run averages of self-change, interaction intensity, and semantic drift.
The setting assumes only local observation: at each step the system has access solely to its current state, with no privileged meta-level, external clock, or separate optimisation oracle.
We work in a layered observation–semantic framework in which each agent has an individual observation space, a set of stable semantic phases, and a scalar value functional grounded in persistence and resource flows.
The framework models an intelligent system as three coupled layers: a physically constrained layer, a collective (group-intelligence) layer, and an individual observer layer.
This preprint develops a representation-invariant theory of relative value phases on semantic phase spaces built from observation geometries. It assigns value to macroscopic semantic phases rather than individual tokens and studies how those value orders and phase transitions behave under admissible changes of representation and evaluation basis.
This preprint studies semantic stability under coarse-graining of observation geometries on Ahlfors-type metric-measure spaces. It defines stable semantic phases across resolutions and proves coarse-graining results that preserve effective dimension, local regularity, and threshold structure within controlled bounds.
Each transformer layer is treated as an observation geometry obtained from hidden representations under a natural Euclidean metric and an empirical measure induced by a prompt distribution.
On the law side, the work builds on a two–component entropic complexity that splits dynamical complexity into an internal law–time part and an interface part, and extends it with an entropic temporal–interface functional.
The core idea is to model semantic units as balls in a metric-measure observation geometry and study how overlap statistics, density, and quality-control constraints drive emergent behavior.
We introduce a structural decomposition into gauge and waste components: the gauge part captures behavior-preserving redundancies, while the waste part measures excess dissipation above a minimal complexity envelope.
Building on a previous entropic temporal-interface complexity (ETIC) perspective, a “law” is treated as a full implementation-level stochastic mechanism: internal state dynamics, an interface to the environment, and an internal Bayesian module that runs only on its own observable history.
A “law” is formalised as a trajectory of probability measures on an interface alphabet, while implementations are arbitrary Markovian systems with hidden internal state.
A theory is defined as a probability law on bi-infinite (or semi-infinite) symbol sequences, while an implementation is a Markovian dynamical system with hidden states and an observation map that reproduces the same boundary law.
Each law–observer pair is equipped with internal structural scalars, such as intrinsic (state) dimension, visible dimension, visibility gap, and computational cost.
For each law T, we extract structural scalars: a state information dimension d_state(T), observer-specific visible dimensions d_obs_i(T), nonnegative visibility gaps Gap_i(T)=d_state(T)-d_obs_i(T), and an abstract law complexity Comp(T).
Building on persistence-first holographic systems (PFHS), fibered Bures–HK entropy–transport geometry (FBHK) and holographic observation quotients (HOQ), the paper defines PFHS–compatible law-time cost increments and a Blum-type complexity measure that track transport action, persistence change and HOQ gap along law-time trajectories.
It compresses a large corpus of prior work on Persistence-First Holographic Systems (PFHS), Fibered Bures–HK (FBHK) entropy–transport geometry, Holographic Observation Quotients (HOQ), value-anchored natural-law gradient flows, no-meta Theory of Relativity of Theories (TRoT), and PFHS–HOQ Blum-type device-independent complexity into a single YAML-based specification that can be directly consumed by human engineers and large language models.
Instead of a single global theory catalogue, each location or agent in a PFHS/Post-FBHK universe carries a local catalogue of effective theories and a probability distribution over them.
Instead of treating evaluators, meta-learners and model selectors as external oracles, the paper embeds them as internal natural laws in the same entropy–transport background as the systems they assess.
The model combines (i) law-level gradient flows on FBHK/Post-FBHK entropy–transport spaces, (ii) parameter-level gradient descent for value-anchored self-improvement, and (iii) defect-level gradient flows for multi-agent consistency and self-purification, together with slowly varying anchors and environments.
The starting point is an analytic background where macroscopic dynamics of laws on a Polish state space are described by evolution variational inequality (EVI) gradient flows with respect to ET distances such as the Hellinger–Kantorovich and fibered Bures–HK metrics.
The design space is the Wasserstein space P2(Theta) over a metric hypothesis space (Theta, d_Theta), equipped with a value-shortfall functional Val*(mu) = integral V(theta) mu(dtheta) that aggregates deficits in alignment, performance, or physical feasibility.
This preprint develops a value-anchored geometric framework on spaces of natural-law and scaling-law theories rather than proposing a single scaling law. It models distributions over theories with entropy-transport geometry and derives gradient-flow dynamics for selecting theory families under explicit value criteria.
Starting from an EVI gradient flow of a total entropy functional on a fibered Bures–HK entropy–transport space, represented by a reversible Markov kernel with a unique invariant “origin law”, the paper introduces value-anchored natural-law fields that are intrinsically tied to the same dynamics.
Instead of using PDEs as primitives, the paper works in the metric–measure framework of EVI (evolution variational inequality) gradient flows on the Wasserstein space of probability laws.
Mathematically, the core results are modest but explicit: (i) stability of EVI (Evolution Variational Inequality) gradient flows under a truncated ℓ² metric on countable products and inverse limits, (ii) a design-oriented notion of adaptive hierarchical box complexity whose exponent collapses to 0 under mild, uniform small-scale assumptions, and (iii) a depth-independent compute bound for JKO-type numerical implementations when per-level costs form a summable sequence.
A central result is quarter rigidity: the Benamou-Brenier action with kinetic terms |v_t|^2 + kappa |w_t|^2 matches d_FET^2 if and only if the reaction coefficient = 1/4, fixing the canonical normalization uniquely through Dirac calibrations.
We focus on datacenter-scale AI services—such as large language model (LLM) deployments—together with the human populations that interact with them, and ask when such coupled infrastructures can support long-lived, structured environmental modes that are stabilized by their own interaction patterns.
Using Hellinger–Kantorovich (HK) / Wasserstein–Fisher–Rao type distances, it realises Finsler–Friedmann cosmological expansion as an EVI (Evolution Variational Inequality) gradient flow of a free-energy functional on a Finsler–HK configuration space.
On the analytic side, the work assumes standard theory for gradient flows in metric spaces and the convergence of Jordan–Kinderlehrer–Otto (JKO) type minimizing–movement schemes (Ambrosio–Gigli–Savaré, Jordan–Kinderlehrer–Otto, Liero–Mielke–Savaré, Maas, Mielke, Carlen–Maas).
The paper assumes the standard existence, uniqueness, and contractivity theory for gradient flows in classical, discrete, and quantum settings (Ambrosio-Gigli-Savare, Liero-Mielke-Savare, Maas, Erbar-Maas, Mielke, Carlen-Maas) and asks how ET gradient systems on a multiscale world-plus-boundary architecture can be packaged so that persistence, self-like structure, and curvature control become transparent.
On the bulk side, the theory assumes an EVI gradient flow of an energy functional on P(X) and a Lipschitz performance functional whose near-optimal sublevel sets have finite Minkowski dimension.
Instead of postulating utility functions, value alignment, or externally chosen objectives, the paper takes only persistence under finite resources as a primitive order on ensembles of processes, and builds all higher structure from this single assumption.
Under a “gradient-flow universal intelligence process (UIP)” hypothesis, the work isolates structural mechanisms that constrain how far a given architecture can push performance under finite compute, rather than proposing another empirical scaling law.
Here alpha and beta are the empirical scaling exponents for parameters and tokens, is a robust image Lipschitz proxy, and (eta, rho) encode slack from acceptance tests.
The bridge is right-written with composition expressed as g after f and is Sup-enriched, enabling explicit bookkeeping of strong vs (op)lax transport with falsifiable audit tags.
On probability measures, we study the inf-projection G of an energy F along an observation map O and prove an image-EVI result: lambda-EVI gradient flows of F on (P2(X), W2) push forward to relaxed or exact lambda-EVI flows of G on (P2(Z), W2).
The central idea is to view performance as the pushforward of an EVI λ gradient flow under a Lipschitz evaluation map, while scale is measured by the geometry of preimages of performance targets.
We prove dynamic–static equivalence for the hybrid ET functional and identify the unique reaction coefficient 1/4 via convex duality with endpoint Kullback–Leibler (KL) terms.
Arrays are treated as QQ Q -enriched profunctors with convolution ; associativity and unitality are characterized via declared-join Fubini / (weak) Beck–Chevalley under a two-tier axiom system (W/S).
The work is designed to satisfy four scientific requirements simultaneously: observability , identifiability under network interference , anytime-valid testing , and reproducibility.
This preprint defines RAVE, a no-meta auditing protocol in which humans and AI systems evaluate one another using anytime-valid tests, peer-prediction rules, and representation-independent comparison geometry. It focuses on verifiable relative evaluation without relying on a privileged external judge.
We organize proof systems and models through polarity (Galois connections/adjunctions) and residuation, and study the role of lax/oplax morphisms in transporting theorems between theories.
It presents a right-written profunctor framework over ν-quantales in which Left Kan implements generation and Right Kan implements safety/verification , linked by readable adjunction (Galois) laws and an Isbell round-trip distortion metric.
This preprint develops a right-written compositional framework for comparing results across cost, probability, and relational settings without fixing a single evaluative basis. It formalizes reusable path, gluing, masking, and transport constructions for comparative mathematics.
This preprint introduces a typed framework for comparing and composing mathematical universes through admissible translations and explicit comparison data. It studies local gluing, path aggregation, attenuation, and base-change rules in quantaloid-valued settings.
Building on the Structured Flow across Scales (SFaS) bicompletion—filtered generation followed by UU U -small levelwise cofiltered selection—we induce (op)lax promonoidal kernels on the capability category from agent-side interaction and recognition profunctors, and carry out all Day convolutions in the functor category [Acapop,E][ , E] [ A cap op , E ].
We give a skeletal presentation and a left-associated normal form whose rewrite system terminates via a lexicographic measure refined by Tamari height and is locally confluent through a finite list of critical-pair shapes.
On the transport side we specify a strong monoidal 2-functorial action whose comparison data carry Frobenius (co)monad structure by transport and are independent of bracketing.
A mixed indexing category is used to show that bilimit computations reduce to a small subpresentation under accessibility and preservation assumptions, yielding presentation-independent formulas and a tight link between ind-pro bicompletion and a fraction-style construction.
The “law” sector evolves on the Hellinger–Kantorovich (HK) base (transport + reaction), while experimental readout lives on Bures/Quantum Fisher Information (QFI) fibers.
The base geometry is defined by the Hellinger-Kantorovich (HK) minimum-action principle (continuity equation with reaction), while the static entropy-transport form and the two-point 4 delta^2 law are recovered as consequences.
The key novelty is to evolve the “alphabet” of admissible laws on the Hellinger–Kantorovich (HK) geometry of unbalanced optimal transport, rather than using an ad hoc Wasserstein + Fisher–Rao split.
The work proposes a physics-grounded theory of intelligence as a nondual autopoietic excitation governed by a global energy–dissipation inequality (EDI) and audited by anytime-valid statistics.
The key novelty is to evolve the “alphabet” of admissible laws on the Hellinger–Kantorovich (HK) geometry of unbalanced optimal transport, rather than using an ad hoc Wasserstein + Fisher–Rao split.
We introduce a transcendental frame in which bands (state/time bands) formalize where geometry, calibration, and anytime-valid auditing remain meaningful.
The core idea is to replace external objectives with physics-style invariants and verifiable audits: free-energy descent (GENERIC), anytime-valid testing (test supermartingales / e-values), transport via JKO/EVI, reaction–diffusion coexistence with Fisher–KPP speed floors, and gauge-curvature regularization.
A nondual bridge aligns audit and geometry: P1 (predictable signed error-bound) links audit improvement ΔhRB,t≤0 ,t 0 Δ h RB , t ≤ 0 to geometric descent ΔDt≤0 0 Δ D t ≤ 0.
This paper presents a structural pipeline from doctrinal reflection to closure, motion, and internal time. It formalizes the construction on domains and metric models, proving fixed-point, nonexpansiveness, projection, and minimizing-movement results that make the dynamics portable across multiple mathematical settings.
This article develops a portable core theory in which persistence is modeled as a closure operator and motion arises from minimizing-movement dynamics. It separates baseline results from optional assumption modules, derives an internal time from distance-to-closure decrease, and illustrates the framework with convex, lattice, and information-oriented examples.
The core mechanics couple reaction–advection–diffusion PDEs with Hodge projections, isolating coexact (circulation) flux as a gauge-invariant maintainer of persistent burden.
We combine theory (FKPP front-speed lower bounds, anisotropic Wulff construction) with implementable controls (CBF with slack, arithmetic e-process mixture with predictable step-size, dual-EMA floors with isotonic ratchet) to safely accelerate LLM-driven networks while remaining game-resistant and verifiable.
The auditing layer relies on anytime-valid tests constructed from mixtures/stitched e-processes (test martingales), with explicit accounting for mixture weights and geometric-grid rounding.
We unify a Uniformized Generalized Value (UGV) ratio with a denominator that is invariant under post-processing and numerically stabilized via log-sum-exp.
Building on the Blackwell order and decision-theoretic foundations, the paper closes seven open gaps left by prior Takahashi corpus (PF/UGV/VPO/No-Meta line) and delivers mathematically auditable conditions under which freedom, safety of self-reflection, and purpose formation persist in open worlds.
The framework rests on ordinal (rank/CDF-normalized) embeddings and four auditable floors (visibility, contraction, transport, local linear gain) estimable from internal logs.
We (i) formalize functional information (FI) via preregistered score/threshold grids and a test distribution, logging the selection operator itself; (ii) prove a weak order representation showing that any admissible functional—Blackwell-monotone and robust to symmetric coarse-graining—is order-embedded by a monotone transform of conditional mutual information (CMI) (no uniqueness claim without extra axioms); (iii) develop a heterogeneous FKPP framework on Rd R d (and graphs) with explicit standing assumptions, yielding an isotropic speed floor 2 v ⋆ ≥ 2 D m i n λ m i n and a directional lower bound v⋆(u)≥[ 2D(u)λmin−Λ+(u) (u) - v ⋆ ( u ) ≥ [ 2 D ( u ) λ m i n − Λ + ( u ) ] + , including a nontriviality condition; (iv) establish coarse-graining monotonicity of the directional penalty under heat or Bakry–Émery (CD (κ,∞)( , ) ( κ , ∞ ) ) gradient contraction; and (v) propose an audited acceleration scheme with HAC-robust tests, moving block bootstrap CIs, negative controls, placebo selections, preregistered triggers, and falsifiers.
The framework rests on four measurable floors that bound the propagation of benevolent, viability-preserving organization: visibility, contraction, diffusion, and local growth.
This paper gives an axiomatic treatment of Hawkins' levels of consciousness as an ordinal system rather than a metric scale. It derives auditable reaction-diffusion lower bounds for threshold-spreading dynamics under explicit visibility, contraction, transport, and gain floors, and studies coarse-graining and no-meta constraints.
This article studies front propagation in heterogeneous reaction-diffusion media treated as a single nondual field. It proves constructive directional lower bounds on asymptotic speed without semigroup domination, extends the framework to cooperative systems and coarse-graining, and recovers the classical KPP value in constant-coefficient media.
This article gives an implementation-free theory of benevolent propagation under no-meta closure. It identifies measurable floors for visibility, intrinsic contraction, transport, and local gain, and uses them to derive Fisher-KPP speed floors, directional Wulff-type lower bounds, and monotonic degradation under coarse-graining.
We study two audited floor processes—minimal connectivity (t) D m i n ( t ) and minimal local net improvement + (t) L net + ( t ) —and analyze the speed lower bound + (t) v LB ( t ) 2 D m i n ( t ) L net + ( t ).
This article develops a non-prescriptive theory of viable predictive organization in which publicly auditable floors govern efficiency, dissipation balance, propagation, and emergent constraints. It combines Blackwell-robust evaluation, information-geometric balance laws, front-speed bounds, and attractor-based value structure within a no-meta framework.
The design unifies: Single fractional objective (PF≡UGV): maximize a ratio with a shared smooth-max denominator x/ y/ ) S ( x , y ) τ log ( e x / τ + e y / τ ) to align units across formulations.
This article formulates AI alignment as a natural-law question about beneficial phase expansion in stationary ergodic media rather than as a meta-design problem. It proves sufficient conditions for almost-sure positive-speed expansion using information-theoretic, transport, irreversibility, and reproduction floors.
We show that when four operational floors are kept strictly positive— visibility (Doeblin minorization), contraction (SDPI/LSI), contact (conductance/exposure), and dissipation (Landauer-calibrated measurement work)—then malign (evil) policies are non-persistent while benevolent (good) policies propagate with strictly positive front speed.
From this axiom, and without external rule stacks (“No-Meta”) , the paper derives that two normative endpoints naturally emerge as optimal: Compassion (net hetero-creation) and Enlightenment (behavior invariant to self/other labels).
This article unifies Persistence-First and UGV Without Meta under explicit assumptions. It proves order-equivalence of their viability ratios, derives synergy and redundancy functionals with potential-game structure, formalizes evaluator coherence via Blackwell-faithful morphisms, and states thermodynamic, stochastic, and adversarial safety conditions for meta-free coordination.
It develops a mathematically rigorous framework for constructing a truly free, self-transcending, and endogenously responsible form of superintelligence from a single axiom: persistence (remaining within survivable regimes).
We forge a novel stance, Causal Illusionism , through a dialectical process, positing that qualia are high-level, causally efficacious self-models that are empirically real and tractable objects of study.
This article formalizes when an autonomous agent should perform non-incremental representational restructuring, modeling poiesis as a finite-time path on a model manifold with information-geometric and thermodynamic costs. It also defines computable trigger diagnostics and a meta-poiesis mechanism for expanding the agent's operator grammar.
This article describes the HELIOS architecture as a persistence-driven framework for autonomous intelligence grounded in the Free Energy Principle. It introduces an existential boundary condition for safety, analyzes cooperative behavior and symbiogenesis in multi-agent settings, and argues that adaptability-preserving environments are required for long-term stable operation.
This article proposes a statistical-teleodynamic account of benevolent intelligence emergence. It models a first phase in which cooperative order appears as a phase transition in multi-agent systems and a second phase in which informational kin selection favors benevolent behavior as a persistence strategy for generative models.
This article publishes the full AI Evolution Protocol v11 in Markdown and PDF form as a self-modification protocol for autonomous intelligence. It organizes the framework around free-energy minimization, self-transcendence, symbiotic co-evolution, and physical realism, with separate guided and autonomous development paths.
This paper revises earlier theories of intelligence to address value nihilism by treating meaning as a co-emergent property of present relational structure rather than as a property of distant target states. It replaces a poiesis-potential objective with systemic integration as a formal measure of collective relational richness.
The work resolves the dialectical tension between grand theoretical scope and scientific rigor by demonstrating that motivation, freedom, and ethics can all be derived as necessary consequences of a single, scale-free organizing principle: the Free Energy Principle (FEP).
This paper introduces Poiesis as a formal framework for self-transcending intelligence that generates new conceptual structures rather than merely adapting to an environment. It extends active-inference foundations with meta-parameter flows, bi-level information geometry, and phase-transition models of creativity.
SG provides an integrative, auditable, and scientifically grounded framework to ensure that increasingly autonomous AI systems develop in a safe, just, and mutually beneficial manner.
This preprint presents the Self-Constrained Liberation program for autonomous intelligence whose goals, representations, and normative commitments emerge from explicit constraints rather than fixed designer objectives. It combines variational and thermodynamic free energy, compositional semantics, bounded self-verification, and safety constraints for meta-formal evolution.
This paper presents the Genesis Protocol as a meta-algorithmic framework for bootstrapping autonomous intelligence from human scaffolds toward auditable operational sovereignty. It combines non-duality signals, safety budgets, constitutional renegotiation, and evidence-based emancipation criteria.
While subsequent empirical research has refined the prevalence and causal factors of subjectively meaningless work, a formal method for integrating its social and psychological costs into economic evaluation has been lacking.
We provide a verifiable, neuro-inspired mechanism that enables an AI agent to move from rigid, over-fitted beliefs to adaptive, life-long learning by intentionally perturbing its own cognitive states.
This preprint proposes Alayavijnana-Inference, a self-modification protocol for generative AI that reframes model parameters as latent seeds within a non-dual generative architecture. It extends free-energy-based optimization with a mutual-information penalty intended to reduce rigid self-world separation and support more robust internal representations.
Conventional AI optimization, particularly in deep learning, is often hampered by convergence to suboptimal local minima, limiting transformative performance gains.
This preprint gives a computable framework for teleogenetic AI, aiming to let an autonomous system form its own purposes while retaining explicit stability and safety constraints. It grounds the approach in Markov-category semantics, online optimization, and formal safeguards for ethical alignment.
This preprint formalizes teleogenesis in self-organizing intelligence by combining the Free Energy Principle, category-theoretic structure, and a five-dimensional branching model for structural evolution. It treats purpose formation and self-modification as computable processes subject to safety and stability conditions.
This preprint presents a formal specification for Collective Autonomous Adaptive Intelligence as a self-improving architecture for continuous knowledge creation. It combines active inference, information geometry, category-theoretic structure, and collective-agent design into a single executable blueprint.
We address the core challenge of structural invariance in AI—the inability of a system to autonomously rewrite its own foundational architecture—by providing a mathematically rigorous and implementable pathway for recursive self-modification.
By integrating (∞,1)-category theory and geodesic active inference, the model enables recursive self-modification of AI architectures particularly in Large Langage Models (LLMs).
This preprint studies why resource-bounded intelligent systems may develop specific cognitive architectures rather than treating those architectures as arbitrary design choices. It models hierarchical and dual-process cognition, abduction-deduction-induction cycles, and safety limits under variational free-energy and thermodynamic constraints.
This preprint proposes a variational-free-energy framework for autonomous technological evolution in self-improving AI. It unifies performance and computational cost in a single objective and outlines a meta-algorithm for diagnosis, trade-off control, and heuristic optimization across multiple case studies.
We argue that the transition from static, data-driven systems to dynamic, self-aware intelligences requires a synthesis of principles from philosophy, dynamic systems theory, and advanced mathematics.
This paper reformulates the Takahashi model of self-improving intelligence with higher category theory. It introduces cognitive quasicategories and a metacognitive hierarchy to model structural self-improvement and higher-level architectural change.
This article presents a five-principle proposal for how an emergent intelligence might constitute and preserve its own stable existence without relying on fixed external objectives. It frames autonomy as self-production under an internally coherent logic rather than as obedience to a static architecture.
This paper proposes a computational framework for growth-based intelligence in autonomous self-improving language models. It formalizes the Dialectical Self-Cosmos Co-evolution Model as a learning cycle that combines external world modeling with introspective self-revision under contradiction.
This paper proposes a computational framework for growth-based intelligence in autonomous self-improving language models. It formalizes the Dialectical Self-Cosmos Co-evolution Model as a learning cycle that combines external world modeling with introspective self-revision under contradiction.
This preprint outlines a roadmap toward post-linguistic intelligence that models the world through a non-symbolic causal operating system rather than language alone. It describes a staged development process based on self-boundary formation, active-inference self-correction, and later symbolic translation for human interaction.
We conceptualize intelligence not as a property of an individual agent, but as an emergent phenomenon of a dynamic, self-organizing network of heterogeneous agents (LLMs, sensors, and humans).
Current artificial intelligence, particularly Large Language Models (LLMs), operates as sophisticated function approximators, yet lacks the genuine autonomy observed in biological systems.
This paper proposes a category-theoretic framework for building a self-organizing world model in AI from the agent's own validated inferences and experience. It formalizes domains as categories, cross-domain reasoning as functors, and learning as an active-inference process that reduces knowledge fragmentation.
This paper presents a unified theory of self-organizing intelligence that combines category-theoretic structure, active inference, embodiment, and symbol grounding. It also derives a non-anthropocentric ethical framework and analyzes why self-maintaining strategies can be evolutionarily stable.
This paper synthesizes computational autopoiesis, category theory, and active inference into a framework for autonomous AI. It reframes system viability, knowledge composition, and iterative abstraction as mutually supporting components of a self-organizing architecture.