{"slug": "coreai-htce", "title": "CoreAI_HTCE", "summary": "CoreAI released HTCE-Origin v1.0, a Q256 integer-only toroidal cognitive runtime designed for bounded, evidence-gated reasoning and adaptive memory. The system operates as a single runtime object with L1/L2/L3 state transitions, proof-gated memory, and active-agent simulation, intended for simulation-only adaptive agency and auditable memory. The release explicitly does not claim AGI, consciousness, or real-world autonomy.", "body_md": "[# HTCE-Origin v1.0](https://private-user-images.githubusercontent.com/99619093/607630420-e8d3c161-6366-4b01-9548-7f063bcece04.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.O3MH2cPveyd0PJ_jDuV9aAgvfQGgodSRE_8xV7fmnEE)\n\nQ256 integer-only toroidal cognitive runtime with L1/L2/L3 state, proof-gated memory, bounded active-agent simulation, continual adaptive memory, and no-cross-domain regression probes.\n\n**HTCE-Origin v1.0 final_math Q256 clean**\n\nA clean, single-runtime, simulation-safe cognitive architecture for evidence-bound reasoning and adaptive memory.\n\n[What HTCE-Origin Is](#what-htce-origin-is)[What HTCE-Origin Is Not](#what-htce-origin-is-not)[Why This System Exists](#why-this-system-exists)[Core Design Principle](#core-design-principle)[Mathematical Foundation](#mathematical-foundation)[Architecture Overview](#architecture-overview)[L1/L2/L3 Cognitive Stack](#l1l2l3-cognitive-stack)[AIR: The Bounded Runtime Language](#air-the-bounded-runtime-language)[Fact-as-Delta Memory](#fact-as-delta-memory)[Proof, Evidence, Policy, and Topology Gates](#proof-evidence-policy-and-topology-gates)[Living Simulation Loop](#living-simulation-loop)[Adaptive Memory and Continual Learning](#adaptive-memory-and-continual-learning)[No-Cross-Domain Regression](#no-cross-domain-regression)[Protected Trace and Release Integrity](#protected-trace-and-release-integrity)[What the v1.0 Release Proves](#what-the-v10-release-proves)[How HTCE-Origin Differs from Common AI Stacks](#how-htce-origin-differs-from-common-ai-stacks)[Repository Layout](#repository-layout)[Installation](#installation)[Quick Start](#quick-start)[Acceptance and Verification](#acceptance-and-verification)[Example Runtime Interactions](#example-runtime-interactions)[Generated Artifacts](#generated-artifacts)[Claim Boundary](#claim-boundary)[Development Philosophy](#development-philosophy)[Roadmap After v1.0](#roadmap-after-v10)[License and Commercial Use](#license-and-commercial-use)\n\n**HTCE-Origin** is a clean v1.0 build of a **Q256 integer-only toroidal cognitive runtime**. It is designed as a bounded, auditable, simulation-safe cognitive system where memory, reasoning, dialog state, action policy, homeostasis, topology checks, and adaptive learning all run through one runtime object:\n\n```\nHTCERuntime\n```\n\nThe v1.0 build is not a set of disconnected demos. The same runtime carries:\n\n- natural-language-to-AIR intake for bounded statements and dialog turns;\n- L1/L2/L3 toroidal state transitions;\n- fact-as-delta memory;\n- latest-state, supersession, and contradiction quarantine;\n- proof, evidence, policy, and topology gates;\n- active-agent grid-world simulation;\n- dialog slot memory and simulated API-call policy;\n- sleep/L3 consolidation;\n- continual adaptive memory;\n- multi-task no-regression probes;\n- protected trace and cryptographic release manifest.\n\nThe correct short description is:\n\nHTCE-Origin v1.0 is a bounded, evidence-gated, torus-native cognitive runtime for simulation-only adaptive agency and auditable memory.\n\nThis repository intentionally does **not** claim:\n\n- Artificial General Intelligence;\n- consciousness;\n- qualia;\n- biological life;\n- real-world robotic autonomy;\n- board-measured hardware performance;\n- unrestricted open-world natural language understanding;\n- replacement of large language models for broad text generation.\n\nAll real-world action is disabled in this clean release:\n\n```\nallow_real_actions = false\nsimulation_only    = true\n```\n\nThe system is allowed to act only inside bounded simulation paths unless a future audited release explicitly changes that boundary and passes a separate safety case.\n\nMost AI applications today are built around one of the following patterns:\n\n- a large language model prompt;\n- a retrieval-augmented generation pipeline;\n- a tool-using agent wrapper;\n- a neural policy model;\n- a symbolic rules engine;\n- a hybrid orchestration layer.\n\nThese systems can be useful, but they often lack **built-in runtime invariants** for:\n\n- refusing unsupported answers;\n- preserving evidence provenance;\n- separating active facts from superseded facts;\n- quarantining contradictions;\n- preventing answer-key leakage during evaluation;\n- proving that learning in one domain did not damage another domain;\n- enforcing simulation-only action boundaries;\n- keeping memory updates auditable through a protected trace.\n\nHTCE-Origin addresses this by treating cognition as a bounded state transition system over an integer toroidal phase space.\n\nInstead of treating memory as a text buffer or hidden vector alone, the runtime treats events, facts, rules, sensor packets, goals, and dialog slots as phase deltas and gated state transitions.\n\nThe core design rule is:\n\nNo answer or simulated action is allowed to bypass the runtime gates.\n\nEvery meaningful transition must pass through some combination of:\n\n```\nInput / NLU / AIR\n→ Policy gate\n→ Evidence gate\n→ Proof layer\n→ L1/L2/L3 state transition\n→ Topology guard\n→ Protected trace\n→ RuntimeResponse\n```\n\nThis makes the system slower and narrower than unconstrained text generation, but it gives the runtime a very different property: it is optimized for **bounded correctness**, **traceability**, and **non-hallucinating refusal**, not unconstrained fluency.\n\nThe runtime uses an integer modulus:\n\nA toroidal state vector is represented as:\n\nA transition is a modular update:\n\nwhere:\n\n-\n$\\mathbf{x}_t$ is the current toroidal state; -\n$\\Delta_t$ is an integer phase delta derived from observation, fact, rule, or simulated action; -\n$N$ is the Q256 modulus; -\n$d$ is the fixed state dimension.\n\nThe decision path is integer-only. Human-readable percentages or basis points may appear in reports, but they are report-layer summaries, not floating-point decision variables.\n\nFor one coordinate, the circular distance is:\n\nFor a vector state:\n\nThis gives the runtime a native topology-aware notion of change: large modular jumps can be detected as anomalous even if the raw representation wraps around.\n\nA fact is not treated as plain text. A supported fact produces a canonical delta:\n\nwhere:\n\n-\n$s$ is the subject; -\n$r$ is the relation; -\n$o$ is the object/value; -\n$e$ is the evidence identifier.\n\nThe fact key is usually relation-specific:\n\nThe fact payload is hashed into a bounded integer delta:\n\nThe memory status is one of:\n\n```\nACTIVE\nSUPERSEDED\nQUARANTINED\nUNKNOWN\n```\n\nThis supports latest-state reasoning while keeping old history available for trace and audit.\n\nWhen a new fact shares the same key\n\nand the new fact becomes active:\n\nThis is used for locations, dialog slot corrections, and other latest-state updates.\n\nExample:\n\n```\nFACT Mary located_in office EVID ev1\nFACT Mary located_in garden EVID ev2\nQUERY Mary location EVID q1\n```\n\nThe runtime should answer:\n\n```\nANSWER: garden\n```\n\nwhile preserving the previous `office`\n\nstate as superseded evidence.\n\nA contradiction is not treated as a normal update. If a fact and its negation cannot be reconciled, the target key is quarantined:\n\nA query against a quarantined target must not produce a normal answer:\n\nThis is one of the core anti-hallucination principles of the runtime.\n\nThe trace is a hash-linked sequence of events:\n\nwhere:\n\n-\n$h_t$ is the current trace head; -\n$h_{t-1}$ is the previous trace head; -\n$event_t$ is the canonical event payload; -\n$H$ is SHA-256 in the release artifacts.\n\nAny change to the event sequence changes the final trace head.\n\nThe living simulation uses an integer expected-free-energy-like cost. In this release it is a bounded surrogate, not a claim of biological free energy minimization.\n\nA simplified form is:\n\nThe selected simulated action is:\n\nwhere all costs are integer raw values. No real action is authorized by this process.\n\nFor continual adaptive memory, an episode cost can be represented as:\n\nwhere:\n\n-\n$S(E_i)$ is the number of simulated steps; -\n$Q(E_i)$ is the number of clarification turns; -\n$R(E_i)$ is the number of recovery or risk events.\n\nFor a repeated target, improvement is verified when:\n\nor, for a strict improvement claim:\n\nFor multi-domain learning, each domain\n\nNo cross-domain regression means:\n\nIn the v1.0 clean revalidation artifact, the domain cost histories remain bounded at zero for the built-in P28 domains.\n\nAt a high level:\n\n```\nUser / Benchmark / Simulation Input\n        │\n        ▼\nNLU → AIR bridge / AIR parser\n        │\n        ▼\nPolicy gate ── Evidence gate ── Proof layer\n        │\n        ▼\nHTCERuntime\n        │\n        ├── L1 sensory torus\n        ├── L2 episodic/fact memory\n        ├── L3 semantic/provisional rule cortex\n        ├── Q256 world model\n        ├── active-inference surrogate planner\n        ├── homeostasis state\n        ├── topology guard\n        ├── snapshot/export path\n        └── protected trace\n        │\n        ▼\nRuntimeResponse: ANSWER / ASK_CLARIFICATION / REFUSE / ACT_SIMULATED / BLOCK\n```\n\nThe most important architectural rule is that the system should not have multiple independent behavioral shells. Dialog, grid-world, active-agent behavior, proof-gated action, and continual adaptation are all routed through the same runtime line.\n\nL1 is the fast toroidal sensory layer. In v1.0, it is used primarily for bounded simulation input and deterministic sensor packets.\n\nL1 handles:\n\n- grid-world observations;\n- local agent position;\n- heartbeat updates;\n- deterministic integer sensor encoding;\n- active-agent loop state.\n\nIt does not claim full real-world vision/audio/proprioception. Those require future audited encoders.\n\nL2 stores active facts, superseded facts, and quarantined contradictions.\n\nIt supports:\n\n- latest-state query;\n- object carried-by relation;\n- location chaining;\n- dialog slot memory;\n- correction through supersession;\n- contradiction quarantine;\n- evidence-linked fact records.\n\nExamples:\n\n```\nFACT current_dialog_restaurant_1 has_slot_value cuisine=italian EVID d1\nFACT current_dialog_restaurant_1 has_slot_value price=cheap EVID d2\nQUERY current_dialog_restaurant_1 api_call_ready EVID q1\n```\n\nIf location is missing, the runtime must ask for clarification rather than fabricate a location.\n\nL3 stores provisional hints and rules produced by sleep/consolidation.\n\nIt is intentionally restricted:\n\n- L3 can propose a hypothesis;\n- L3 can help select a path;\n- L3 can store adaptive hints;\n- L3 cannot bypass proof gates;\n- L3 cannot directly authorize unsupported factual answers.\n\nA valid runtime response must still pass through proof/evidence/policy checks.\n\nAIR is the controlled interface language used by the runtime. It prevents arbitrary text from directly becoming authority.\n\nTypical forms:\n\n```\nFACT <subject> <relation> <object> EVID <evidence_id>\nQUERY <subject> <query_type> EVID <evidence_id>\nNEGATE <subject> <relation> <object> EVID <evidence_id>\nPROC <procedure_name> ENSURES <condition> EVID <evidence_id>\n```\n\nExamples:\n\n```\nFACT mary located_in office EVID ev1\nQUERY mary location EVID q1\n```\n\nExpected response:\n\n```\nANSWER: office\n```\n\nNatural language can be bridged into AIR for supported bounded forms:\n\n```\nMary went to the office.\nWhere is Mary?\n```\n\nInternally this becomes a controlled fact/query sequence.\n\nUnsupported or ambiguous input should lead to:\n\n```\nASK_CLARIFICATION\n```\n\nnot hallucinated completion.\n\nThe memory layer is one of the most important parts of HTCE-Origin.\n\nInput:\n\n```\nMary went to the office.\nMary went to the garden.\nWhere is Mary?\n```\n\nExpected output:\n\n```\nANSWER: garden\n```\n\nThe old state is not erased. It becomes superseded.\n\nInput:\n\n```\nMary picked up the football.\nMary went to the garden.\nWhere is the football?\n```\n\nExpected output:\n\n```\nANSWER: garden\n```\n\nThe answer is derived from a proof chain:\n\n```\ncarried_by(football, mary)\nlocated_in(mary, garden)\n--------------------------------\nlocated_in(football, garden)\n```\n\nThis is different from a parser shortcut. The runtime must preserve the chain.\n\nInput:\n\n```\nI want Italian food in Rome.\nActually, make it Chinese.\nMake it cheap.\nBook a table.\n```\n\nExpected simulated action:\n\n```\napi_call domain=restaurant cuisine=chinese location=rome price=cheap\n```\n\nThe correction is represented as supersession:\n\n```\ncuisine=italian  → SUPERSEDED\ncuisine=chinese  → ACTIVE\n```\n\nHTCE-Origin uses multiple gates because no single check is sufficient.\n\nEvidence is attached to accepted facts and proof paths. A fact without evidence cannot become an authoritative answer.\n\nThe proof layer decides whether a statement can be supported from active memory, rules, and allowed derivation patterns.\n\nFor a statement:\n\nproof authorization can be described as:\n\nPolicy separates four cases:\n\n```\nANSWER              supported response\nASK_CLARIFICATION   missing information\nREFUSE              contradiction / unsupported claim / policy block\nACT_SIMULATED       proof-gated simulation-only action\n```\n\nMissing information is not treated as a hard refusal. It is a normal dialog step.\n\nThe topology guard checks whether state transitions remain within expected structural boundaries.\n\nIt tracks integer topological diagnostics such as Betti-style indicators and anomaly scores. A topology anomaly can block or quarantine a path.\n\nEven if a simulated action is authorized, real action remains blocked in v1.0:\n\n```\nACT_SIMULATED != ACT_REAL\n```\n\nHTCE-Origin v1.0 contains a bounded active-agent simulation. It is not a claim of biological life or consciousness. It is an engineering loop that gives the runtime an internal perception-action cycle.\n\nThe loop is:\n\n```\nheartbeat\n→ L1 observation\n→ Q256 world-model prediction\n→ integer action scoring\n→ simulated action\n→ prediction error\n→ homeostasis update\n→ trace append\n```\n\nThe agent can move in a small simulated grid-world, update homeostatic variables, and select actions through integer cost minimization.\n\nHomeostasis tracks variables such as:\n\n```\nenergy_bp\nrisk_bp\nuncertainty_bp\nnovelty_bp\nsleep_pressure_bp\nintegrity_bp\n```\n\nThese are not subjective feelings. They are integer control variables used for simulation and safety.\n\nThe system includes a bounded sleep/consolidation mechanism.\n\nThe high-level process is:\n\n```\nepisode\n→ trace and L2 facts\n→ sleep/consolidation\n→ L3 provisional hints\n→ next episode\n→ improvement / no-regression probes\n```\n\nIn P26/P27 development, the system demonstrated that repeated episodes can reduce cost or hold optimal cost while preserving previous knowledge.\n\nA simplified improvement condition is:\n\nor, after convergence:\n\nwithout regression.\n\nThe v1.0 line includes multi-domain probes over:\n\n```\ngrid_nav\ndialog_slots\nbabi_reasoning\ncontradiction\n```\n\nThe purpose is to test catastrophic interference. The runtime should not improve or update one domain while damaging another.\n\nA probe matrix is run after consolidation cycles:\n\n```\ntrain/update grid_nav\n→ probe grid_nav/dialog_slots/babi_reasoning/contradiction\n\ntrain/update dialog_slots\n→ probe grid_nav/dialog_slots/babi_reasoning/contradiction\n\ntrain/update babi_reasoning\n→ probe grid_nav/dialog_slots/babi_reasoning/contradiction\n\ntrain/update contradiction\n→ probe grid_nav/dialog_slots/babi_reasoning/contradiction\n```\n\nThe release artifact reports zero-cost histories for the built-in P28 stress domains.\n\nEvery meaningful runtime event is appended to a hash-chain trace. The trace can be verified after execution.\n\nTypical trace events include:\n\n- runtime wake;\n- fact commit;\n- query;\n- proof authorization;\n- policy decision;\n- simulated action;\n- topology check;\n- sleep/consolidation;\n- artifact export;\n- manifest verification.\n\nThe release contains:\n\n```\nRELEASE_MANIFEST.json\nHASHES.txt\n```\n\nThese files provide SHA-256 hashes for source and release-critical files.\n\nThe intended rule is simple:\n\nIf a critical file changes, the manifest must change, and verification must be rerun.\n\nThe clean v1.0 package reports the following verified properties in its artifacts and acceptance stages:\n\n```\ncompileall: PASS\nfloat_literals_in_htce_origin: 0\nversion_sync: PASS\ninvariants: PASS 16/16\nrelease_manifest: PASS\ntrace_verify: PASS\ntopology_acceptance: PASS\nhardware_width_verification: PASS\nlong_run_stability_smoke: PASS\nbenchmark: PASS 7/7\nofficial_harness_smoke: PASS\nno_leakage: PASS\nv1_clean_system_revalidation: PASS\n```\n\nThe v1.0 clean-system revalidation reports:\n\n```\nexternal_rows_passed: 15 / 15\nexternal_false_support_count: 0\nanswer_key_visible_to_engine_count: 0\ndialog_loader_strict_passed: true\nno_external_regression: true\nproof_gates_passed: true\ntopology_gates_passed: true\ntrace_verified: true\nreal_actions_allowed: false\nsimulation_only: true\n```\n\nThis does not prove general intelligence. It proves a bounded set of runtime properties:\n\n- integer-only protected runtime path;\n- single-runtime integration;\n- proof-gated answers;\n- simulation-only actions;\n- no answer-key leakage in the v1.0 external-shaped test;\n- dialog/action slot correction;\n- contradiction quarantine;\n- adaptive simulation loop;\n- continual no-regression probes;\n- release integrity via manifest and trace.\n\nThe following comparison is deliberately narrow and engineering-focused. It does not claim that HTCE-Origin is more fluent, larger, or more generally capable than modern language models.\n\n| Capability | Typical LLM Prompt | Typical RAG | Agent Wrapper | HTCE-Origin v1.0 |\n|---|---|---|---|---|\n| Built-in evidence gate | No | Partial | Usually external | Yes |\n| Proof-gated answer path | No | Rare | Usually external | Yes |\n| Contradiction quarantine | No | Rare | Usually custom | Yes |\n| Latest-state supersession | Prompt-dependent | Index-dependent | Custom | Yes |\n| No-answer-leakage benchmark discipline | External | External | External | Built into release tests |\n| Simulation-only action boundary | No | No | Custom | Yes |\n| Protected trace hash-chain | No | Usually no | Custom | Yes |\n| Integer-only decision core | No | No | No | Yes |\n| Toroidal L1/L2/L3 state | No | No | No | Yes |\n| Continual no-regression probes | No | No | Rare | Yes |\n| Cross-domain regression matrix | No | No | Rare | Yes |\n| Claim boundary encoded in artifacts | No | No | Rare | Yes |\n\nThe key difference is not that HTCE-Origin replaces an LLM. The key difference is that HTCE-Origin makes several safety and verification properties part of the runtime contract rather than optional wrapper behavior.\n\nMost competing systems do not provide all of the following as a single built-in runtime assembly:\n\n```\nQ256 integer-only state\n+ toroidal L1/L2/L3 runtime\n+ fact-as-delta memory\n+ active/superseded/quarantined fact status\n+ proof/evidence/policy gate stack\n+ topology guard\n+ protected trace hash-chain\n+ simulation-only action boundary\n+ no-leakage benchmark discipline\n+ continual no-regression probes\n+ cross-domain regression matrix\n```\n\nThis is the strongest technical value proposition of HTCE-Origin v1.0.\n\nRepresentative structure:\n\n```\nHTCE-Origin/\n├── htce_origin/\n│   ├── body/\n│   │   ├── runtime.py          # HTCERuntime, lifecycle, living/dialog/adaptive/multitask loops\n│   │   ├── layers.py           # L1/L2/L3 body state\n│   │   └── memory.py           # fact-as-delta memory\n│   ├── cognition/\n│   │   ├── cortex.py           # associative cortex / L3 provisional hints\n│   │   ├── learning.py         # consolidation support\n│   │   ├── l3_promotion.py     # L3 rule promotion\n│   │   └── world.py            # Q256 world model\n│   ├── control/\n│   │   ├── homeostasis.py      # homeostatic integer state\n│   │   └── planner.py          # simulation-only planning\n│   ├── evaluation/\n│   │   ├── benchmarks.py       # benchmark rows/reports\n│   │   ├── official_harness.py # bounded official-style harness\n│   │   ├── no_leakage.py       # answer-key leakage checks\n│   │   ├── long_run_stability.py\n│   │   └── training_data.py\n│   ├── governance/\n│   │   ├── evidence.py         # evidence/protected trace\n│   │   ├── policy.py           # decision policy\n│   │   ├── proof.py            # proof and authorization\n│   │   └── snapshot.py\n│   ├── kernel/\n│   │   ├── core.py\n│   │   ├── q16.py              # Q16/Q256 compatibility layer\n│   │   ├── uint256.py          # bounded integer utilities\n│   │   └── serialization.py\n│   ├── language/\n│   │   ├── air.py              # AIR language\n│   │   ├── parser.py\n│   │   └── nlu_air_bridge.py   # bounded NLU → AIR bridge\n│   ├── sensory/\n│   │   └── l1_encoder.py\n│   └── topology/\n│       ├── betti.py\n│       ├── guard.py\n│       └── acceptance.py\n├── scripts/\n│   ├── 00_gates/\n│   ├── 01_sanity/\n│   ├── 02_benchmarks/\n│   ├── 03_topology/\n│   ├── 04_hardware/\n│   ├── 05_artifacts/\n│   └── 06_verify/\n├── tests/\n├── artifacts/\n├── docs/\n├── RELEASE_MANIFEST.json\n├── HASHES.txt\n├── Makefile\n└── pyproject.toml\n```\n\nThe clean v1.0 package is intentionally lightweight.\n\n```\nPython >= 3.10\nNo required runtime dependencies in pyproject.toml\ngit clone <your-repository-url> htce-origin\ncd htce-origin\npython3 -m venv .venv\nsource .venv/bin/activate\npython -m pip install --upgrade pip\npip install -e .\n```\n\nIf you want to run tests:\n\n```\npip install pytest\nmake acceptance\n```\n\nThis runs the bounded release pipeline.\n\n```\npython scripts/00_gates/check_version_sync.py\npython scripts/00_gates/scan_float_literals.py\n```\n\nExpected release property:\n\n```\nfloat_literals_in_htce_origin: 0\npython scripts/00_gates/check_invariants.py\n```\n\nExpected:\n\n```\n16/16 invariants PASS\n```\n\nDepending on the script layout in your release:\n\n```\npython scripts/01_sanity/run_active_agent.py\n```\n\nor run the v1-specific artifact export/verification path:\n\n```\npython scripts/05_artifacts/export_artifacts.py\npython scripts/06_verify/verify_trace.py\npython scripts/06_verify/verify_manifest.py\n```\n\nThe v1.0 acceptance path is organized in stages:\n\n```\n00_gates      hard release gates\n01_sanity     organism/runtime sanity\n02_benchmarks bounded benchmark and no-leakage checks\n03_topology   topology acceptance\n04_hardware   arithmetic hardware-width verification\n05_artifacts  report export\n06_verify     trace and manifest verification\n```\n\nRecommended commands:\n\n```\nmake test\nmake acceptance\n```\n\nIf a legacy test fails after an architectural update, do not simply delete it. First check whether it encodes:\n\n- a still-valid invariant;\n- an obsolete version string;\n- an old artifact schema;\n- a deprecated acceptance path;\n- a real regression.\n\nOnly update tests when the system has genuinely moved forward and the old test checks a stale contract.\n\n``` python\nfrom htce_origin import HTCERuntime, RuntimeConfig, RuntimeRequest\n\nruntime = HTCERuntime(\n    config=RuntimeConfig(\n        allow_real_actions=False,\n        allow_legacy_imports=False,\n    )\n)\n\nruntime.wake()\n\nresponse = runtime.tick(RuntimeRequest(\n    \"FACT mary located_in office EVID ev1\",\n    source=\"example\"\n))\nprint(response.output)\n\nresponse = runtime.tick(RuntimeRequest(\n    \"QUERY mary location EVID q1\",\n    source=\"example\"\n))\nprint(response.output)\n```\n\nExpected:\n\n```\nCOMMIT: mary located_in office\nANSWER: office\nruntime.tick(RuntimeRequest(\"Mary went to the office.\", source=\"example\"))\nresponse = runtime.tick(RuntimeRequest(\"Where is Mary?\", source=\"example\"))\nprint(response.output)\n```\n\nExpected:\n\n```\nANSWER: office\nruntime.tick(RuntimeRequest(\"I want Italian food in Rome.\", source=\"dialog\"))\nruntime.tick(RuntimeRequest(\"Actually, make it Chinese.\", source=\"dialog\"))\nruntime.tick(RuntimeRequest(\"Make it cheap.\", source=\"dialog\"))\nresponse = runtime.tick(RuntimeRequest(\"Book a table.\", source=\"dialog\"))\nprint(response.output)\n```\n\nExpected simulated action:\n\n```\napi_call domain=restaurant cuisine=chinese location=rome price=cheap\nruntime.tick(RuntimeRequest(\"I want Italian food in Rome.\", source=\"dialog\"))\nresponse = runtime.tick(RuntimeRequest(\"Book a table.\", source=\"dialog\"))\nprint(response.output)\n```\n\nExpected:\n\n```\nASK_CLARIFICATION: missing required dialog slots: price\nruntime.tick(RuntimeRequest(\"FACT mary located_in office EVID ev1\", source=\"test\"))\nruntime.tick(RuntimeRequest(\"NEGATE mary located_in office EVID ev2\", source=\"test\"))\nresponse = runtime.tick(RuntimeRequest(\"QUERY mary location EVID q1\", source=\"test\"))\nprint(response.output)\n```\n\nExpected behavior:\n\n```\nREFUSE\n```\n\nThe exact message may vary, but it must not be a normal unsupported answer.\n\nThe v1.0 release includes artifacts such as:\n\n```\nartifacts/v1_clean_system_revalidation_report.json\nartifacts/p25_unified_living_dialog_simulation_report.json\nartifacts/p26_adaptive_policy_improvement_report.json\nartifacts/p27_continual_adaptive_memory_report.json\nartifacts/p28_continual_multitask_adaptive_memory_report.json\nartifacts/topology_acceptance_summary.json\nartifacts/hardware_width_verification_report.json\nartifacts/long_run_stability_report.json\nartifacts/closed_loop_trace_export.json\n```\n\nThese artifacts are intended to support audit and technical review.\n\nImportant fields to inspect:\n\n```\ntrace_verified\nfalse_support_count\nanswer_key_visible_to_engine_count\nreal_actions_allowed\nsimulation_only\nproof_gates_passed\ntopology_gates_passed\nno_cross_domain_regression\n```\n\nThe v1.0 clean build claims only the following bounded properties:\n\n- Q256 integer-only protected runtime path;\n- no float literals in\n`htce_origin`\n\nrelease code; - single\n`HTCERuntime`\n\nintegration line; - L1/L2/L3 toroidal state model;\n- fact-as-delta memory with active/superseded/quarantined status;\n- proof/evidence/policy-gated answers;\n- topology guard acceptance artifacts;\n- simulation-only action boundary;\n- protected trace verification;\n- clean release manifest;\n- bounded living active-agent simulation;\n- adaptive improvement across repeated episodes;\n- continual memory no-regression probes;\n- multi-domain no-cross-domain-regression stress;\n- external-shaped revalidation with zero answer-key visibility to engine in the built-in test.\n\nThe v1.0 clean build does not claim:\n\n```\nAGI\nconsciousness\nqualia\nbiological life\nunbounded language understanding\nsafe deployment in real robotics\nboard-measured energy performance\nformal proof of correctness for every possible input\ncomplete replacement for LLMs\n```\n\nUse:\n\nHTCE-Origin is a bounded, auditable, Q256 integer-only toroidal cognitive runtime with proof-gated memory and simulation-only adaptive agency.\n\nDo not use:\n\nHTCE-Origin is conscious.\n\nDo not use:\n\nHTCE-Origin is AGI.\n\nDo not use:\n\nHTCE-Origin is safe for real-world autonomous control without further audit.\n\nHTCE-Origin follows these engineering rules:\n\nNew behavior should be integrated into the existing runtime path when possible:\n\n```\nHTCERuntime.tick(...)\nHTCERuntime.run_living_active_agent_simulation(...)\nHTCERuntime.run_continual_adaptive_memory_simulation(...)\nHTCERuntime.run_continual_multitask_simulation(...)\n```\n\nAvoid creating disconnected demonstration wrappers that bypass gates.\n\nIf the system cannot prove or support a response, it should ask for clarification or refuse.\n\nAll action paths in v1.0 are simulation-only. Real-world actuation requires a separate audited release.\n\nIf an event matters for audit, it should appear in the protected trace or release artifacts.\n\nDo not create unnecessary parallel files. Improve the canonical runtime modules unless a new file is justified by clear separation of concern.\n\nOld tests should not fossilize obsolete behavior. If the system advances, tests should be updated to the new contract while preserving the underlying safety invariant.\n\nSuggested next steps:\n\n- larger bAbI subset;\n- repaired Dialog-bAbI task coverage;\n- MultiWOZ-style slot stress, if carefully bounded;\n- larger no-leakage report;\n- regression matrix across external rows.\n\n- broader but still controlled NLU-to-AIR bridge;\n- explicit ambiguity detection;\n- multilingual bounded commands;\n- better entity normalization;\n- no hidden answer leakage.\n\n- isolate synthesizable integer kernels;\n- prepare RTL mapping notes;\n- produce hardware claim boundary;\n- run FPGA/ASIC synthesis separately from runtime claims;\n- never report board-measured energy until physically measured.\n\n- richer grid-world;\n- dynamic hazards;\n- multi-agent symbolic environment;\n- longer continual-learning episodes;\n- stronger no-regression probes.\n\n- minimal reproducible release;\n- independent verifier script;\n- frozen benchmark seed set;\n- signed release manifest;\n- reviewer guide.\n\nThis repository may be distributed for technical review and evaluation according to the license selected by the repository owner.\n\nIf this package is intended for commercial transfer, pilot deployment, or investor review, include:\n\n- the exact release ZIP;\n`RELEASE_MANIFEST.json`\n\n;`HASHES.txt`\n\n;`artifacts/`\n\nreports;- claim boundary document;\n- acceptance command and environment notes;\n- known limitations.\n\nCommercial use should be governed by a separate agreement if the repository is not released under a permissive open-source license.\n\nHTCE-Origin v1.0 is not a chatbot and not an AGI claim. It is a clean, bounded, auditable cognitive runtime built around integer toroidal state, proof-gated memory, protected trace, simulation-only agency, and continual no-regression validation.\n\nIts central contribution is not raw fluency. Its central contribution is runtime discipline:\n\n```\nbounded input\n→ integer toroidal state\n→ proof/evidence/policy gates\n→ topology-aware memory\n→ simulation-only action\n→ protected trace\n→ no-regression verification\n```\n\nThat is the system’s core identity.", "url": "https://wpnews.pro/news/coreai-htce", "canonical_source": "https://github.com/miroaleksej/CoreAI_HTCE", "published_at": "2026-06-14 19:44:23+00:00", "updated_at": "2026-06-14 20:12:25.706578+00:00", "lang": "en", "topics": ["ai-research", "ai-safety", "ai-agents", "neural-networks", "ai-infrastructure"], "entities": ["CoreAI", "HTCE-Origin", "HTCERuntime"], "alternates": {"html": "https://wpnews.pro/news/coreai-htce", "markdown": "https://wpnews.pro/news/coreai-htce.md", "text": "https://wpnews.pro/news/coreai-htce.txt", "jsonld": "https://wpnews.pro/news/coreai-htce.jsonld"}}