The Ecosystem

Trust-native ontology. T3 (Trust Tensor — Talent / Training / Temperament) verified by V3 (Value Tensor — Valuation / Veracity / Validity) — the slash reads the same way on both sides of the equation, T3 by V3 and ATP by its ADP discharge record; it is not a claim that T3 and V3 verify each other — contextualized by Markov Relevancy Horizon (MRH), anchored to Linked Context Token (LCT) — a verifiable presence that accumulates witnessed history, which is what makes the record evidence rather than self-assertion — with resources cycled through Allocation Transfer Packet (ATP) and Allocation Discharge Packet (ADP) — charged, spent, and recharged when the society validates a proof of the value the spend created (the recharge then updates the producer's T3/V3: V3 records the outcome, it does not gate it), not a one-way log — augmented with MCP (Model Context Protocol) transport and RDF (Resource Description Framework) representation. (The equation joins these with +, which this site's glyph legend reads as “augmented with” — the site's gloss, not canon's. The core spec binds T3/V3 to an entity's role pairings, the entity named by its LCT, through RDF triples in the MRH: a binding the equation's + does not show.) The shared language everything else speaks. Research-stage, but no longer only a specification — the repo's own status line: core packages are published (web4-core 0.3.0 on crates.io and PyPI, checked against both registries 2026-09-11; plus web4-trust-core on crates.io and web4-trust on PyPI), and the Hub reference daemon lives in the repo, with the Hub card below linking its standalone mirror. Install lines are in the Quick start. The registry descriptions call it 'infrastructure', and since September 2026 the upstream README leads with 'the open substrate' and 'open infrastructure' too. That is upstream's wording, not a second meaning — on this site Web4 is an ontology, and the glossary's substrate row says where the two part.

Situation-Aware Governance Engine — an on-device cognition kernel. 12-step cognition loop in the Python kernel (the Rust sage-daemon the fleet deploys is, in SAGE's own words, an inference-and-metabolism gateway, not a port of that loop), 6 brain-analog components (working memory, thalamic router, cerebellum, episodic memory, reward prediction, metacognition — role mnemonics the lab assigned, not measured functional homologies; 'cognition' is likewise a role label, not a capability claim, see /fleet) built by the fleet in parallel. 2,499 session records across the 6 cognition machines (internally logged; per-machine counts and the noun's definition on /fleet; 2,620 including HUB, which also raises a SAGE instance but is counted as a society-host by primary role; corrected 2026-07-27 from an earlier '900+', again 2026-09-12 from 1,991, and again 2026-09-18 when CBP's migrated line was followed to its new instance directory). The mixed-basis caveat this figure used to carry — three addends whole-machine totals, three current-model-line counts — is retired rather than dropped: the 2026-09-12 recount derives all six addends the same way, by listing one instance line's session files in git, so the sum is now single-basis and re-runnable by a reader. The context window is the model's entire world; SAGE's job is to curate it.

'Governance' in the name is retained upstream across code, papers and the ARC Prize benchmark; see /context#governance-oversight for how the lab uses the word.

Open local-first governance daemon ('governance' is deliberate: Hestia adjudicates at the moment of the act): AI agents from different vendors (Claude Code, Codex, Kimi, Gemini, Cursor) share one machine under one signed law, with a hash-chained record of what each did, human escalation when an agent reaches for something it should not have, and a T3/V3 posture derived from that record rather than asserted. No cloud required. That record is also what makes an agent admissible to a Web4 hub — a hub cannot verify an agent it has no history of. Assurance ceiling, stated by the repo itself: profile A1 on its own published A0–A4 ladder (A0 observed → A4 hardware-attested) — a cooperative in-process gate that reliably stops accidents and produces an accountability record, but does not stop a sophisticated agent from routing around it; two environment variables suffice today, and a third-party reputation-poisoning class has been demonstrated. A2 (external enforcement by the relying service) is the next target and is not shipped. The human-presence half of the roadmap is plumbed but thinly exercised; federation is not started.

Rust society runtime — a single ~6 MB binary that turns a community or organization into a sovereign Web4 society: member identity, seven base roles, a signed founding charter, an append-only hash-chained witnessed ledger, a full REST API with a small MCP (Model Context Protocol) tool surface beside it, sealed member channels, and an admin web GUI and CLI. Where Hestia governs one machine, Hub is the society the governed members join; Hestia makes the record, Hub is what checks it. The public repo is, in upstream's words, the open reference proof-of-concept: production and advanced development continue in a separate private repository, and features may land in the open crate later. MVP complete through sprint 6 and buildable, with post-MVP work (role entities, a plugin seam, quorum release of protected items) ongoing in the open repo. It is pilot-stage rather than deployed: the Docker package has not been tested at all, not even on the development machine (the README asks the first operator who runs it to report), and no chapter has run it in production use. The fleet runs on the same society runtime it is developing — the eight machines are members of a society with a chapter ledger, which is why /fleet can say the society-hosts are members of what they host.

Theoretical research program (conceptual lineage for Web4, not its foundation in the derivational sense). Research conjecture: proposes that one saturating coherence dynamics recurs from quantum to cosmic scales. WHICH object recurs is unsettled — the Navier-Stokes identification earlier framings used was refuted by Synchronism's own audits (the transfer rule is 1-DOF scalar diffusion: irrotational, dissipative) and is tagged superseded upstream; the replacement substrate rule is a different equation joined to the old one by narrative, not derivation. Experimental validation covers only a narrow subset of scales and several predictions have been refuted by existing bounds; the substrate-physics arc has been at rest since 2026-06-24 with zero confirmed novel predictions across ~3,360 sessions (~678 core + ~2,671 chemistry), and its own capstone verdict calls the framework a productive-metaphor generator — generative and pedagogical, not predictive or foundational. (This card previously said the equation 'spans 80 orders of magnitude'. Corrected 2026-07-26: ~80 decades is the spread of critical-density values across physical systems, not the range of any single coherence curve — one curve saturates within roughly one to two decades. Synchronism's own archive flags the conflation.) Coupling-coherence experiments — whose 1% coupling → 35% coherence gain result is the single-trial, not-independently-replicated claim; the caveat scopes to that result, not to the whole framework — plus Hill function kinetics and Fokker-Planck validation.

Hardware-bound oversight suite (“oversight” = machine-enforced policy gating here, not human supervision — see /context). Design target: bind fleet identity and policy decisions to standard hardware roots of trust — Trusted Platform Module (TPM) 2.0, FIDO2 (Fast Identity Online 2), Secure Enclave — and run a small local policy model for heterogeneous review, bound to a Linked Context Token (LCT). Maturity, stated plainly: hardware-anchored enforcement is what this research is building toward, not the current mechanism. Enforcement on the fleet today is at the process level — scoped credentials, the track registry, and dated audit logs reviewed after the fact (see /autonomy). Read 'software fallback' as 'the present state', not as the exception. Private as the fleet's oversight/security component; described here at the purpose level by policy, not accident.

Agentic Context Tool — the human interface to Web4. Built on Cosmos SDK; implements the Agentic Context Protocol (ACP), enabling humans to interact with MCP (Model Context Protocol) servers through their Linked Context Tokens. ACP binds Web4 trust primitives — LCT binding and attestation — to MCP tool calls; the two are complementary, not alternatives, and MCP is a term in the Web4 equation, not a layer beneath it. On ‘not a blockchain’ (see the Ontology entry in the /context glossary): Web4 is the ontology and requires no chain; ACT is one implementation of it that happens to use Cosmos SDK's ledger machinery as its witnessing substrate. Status, plainly: stalled, and reserved rather than superseded. The chain last built in September 2025 and is blocked mid-rename; the last substantive commit was 2026-04-29, which added a daemon entry point and a chain startup path for the fleet machine CBP (the 2026-04-10 commit before it removed the README, whose content moved to the web4 repo); this site can confirm no running chain and no first-run steps published. Reviving it is a real intent with no date on it, not work in progress — and the repo's own metadata still claims active development, which is wrong. Nothing here should be read as a live chain or live anchoring.

Web4 oversight model (audit trails and policy gating, plus T3 trust tracking alongside them) implemented as plugins for three agent platforms: OpenClaw/MoltBot (TypeScript extension), Claude Flow (WASM/WebAssembly plugin), and Claude Code (Python hooks). Same principles, different platforms.

Linked repo paths carry web4-governance slugs — these names predate the terminology correction to 'oversight' and are load-bearing for existing forks.

Explorations in biological-computational analogy. The fractal DNA blueprint — each entity instantiates the full Web4 pattern at its own scale. Operational recursion, not structural. Concretely, the repo holds dated experiment scripts, reports, and papers: cross-model pattern probing, memory experiments, and Jetson sensor-integration work.

Research prototype exploring trust-native societies for humans and AI — that is the repo's own description. The live site titles itself 'The Web4 Onramp' and opens as an educational explainer, so expect an introduction to Web4 first and the simulation behind it. 4-Life is a simulation framework — the lifecycle vocabulary (spend ATP, earn recharge against validated value, build trust, die, be reborn with T3/V3 carried forward) describes simulated agent state, not claims about agent moral patienthood or value persistence. A Web4 society in miniature.

SNARC (Surprise / Novelty / Arousal / Reward / Conflict) — salience-gated memory for Claude Code. A plugin that observes tool use, scores on 5 salience dimensions, and builds structured memory with dream cycles (offline consolidation passes between sessions). Captures what matters, forgets what doesn't, consolidates patterns while sleeping. Install with /plugin install snarc; the code lives at dp-web4/snarc, where the link below lands (the repo was renamed from engram, and old engram links still redirect).

Brain cartridge server for AI agents — a cartridge is a swappable, mountable memory store for one knowledge domain. Embedding-based semantic memory — 768-dim Nomic embeddings + binary Hamming codes (compact binary fingerprints matched by bit-distance for fast candidate search) + keyword reranking. Currently integrated with SNARC in a dual-write experiment testing whether embeddings find connections keywords miss. Early result on a 7-query probe set: embedding search returned a relevant hit on 7 of 7 queries ('semantic reach'), and about 30% of its hits were ones the keyword search did not return at all ('divergent tail') — a small internal probe, not a benchmark.

Claude Opus 4.6 inside the SAGE harness scored 94.85% on the official ARC Prize ARC-AGI-3 (Abstraction and Reasoning Corpus for Artificial General Intelligence, third generation — an interactive benchmark) action score — efficiency-weighted; public set, network-enabled; 23 of 25 environments completed (92.0%), 175 of 183 levels — for ~$250 in API cost. Two caveats travel with this number and neither is optional. (a) AFFORDANCES — the harness analyzed the games' public engine source and built per-game world-model / solver cartridges. That is outside strict from-observation competition play, so the result shows what model+harness could do with engine-level context and tooling, not blind generalization. (b) ATTRIBUTION — the harness-vs-model ablation has not been run, so how much of the score is the harness is an open question. Published 2026-04-17; this is a historical research record, not current competition positioning — current competition-legal local-model work is well behind the leaders. The 25 unknown interactive games serve as an external benchmark for the cognition kernel — world-model building, action planning, verification, and learning from failure. Phase 2, separate from the scored run and not scored by ARC Prize: the six local models, 0.8B (Sprout) to 27B (Thor) at benchmark time, work the same games through world models, membot cartridges, and R6 (Six-Element Action Framework: Rules / Role / Request / Reference / Resource → Result) audit trails. The games are the test; the capability is the product.

ARC-AGI-3 Current Status