Skip to content

STATUS.md

Latest commit

 

History

History
92 lines (62 loc) · 5.78 KB

STATUS.md

File metadata and controls

92 lines (62 loc) · 5.78 KB

Web4 Implementation Status

Last Updated: April 29, 2026

Headline

Web4 is a working ontology with real proof points and real gaps. R&D, not production. The spec corpus is stable; reference implementations exist; some demonstrations have moved from inferred to measured. As of 2026-04-28, the core primitives are publicly installable.

The strongest single proof point: the same Claude Opus 4.6 you can use today scores 0% on ARC-AGI-3 by default and 94.85% with a Web4-shaped harness around it. The model didn't change. The structure around the model did.

Public scorecard: https://arcprize.org/scorecards/c7dfb4f1-8642-4c9e-ab4d-152f5f8e33b4

Published artifacts (current: v0.1.1, 2026-04-28)

Package Registry Version Install
web4-core (Rust) crates.io 0.1.1 cargo add web4-core
web4-core (Python) PyPI 0.1.1 pip install web4-core
web4-trust-core (Rust) crates.io 0.1.1 cargo add web4-trust-core
web4-trust (Python) PyPI 0.1.1 pip install web4-trust

v0.1.0 yanked from crates.io: the Python web4-core wheel shipped without __init__.py, so import web4_core returned an empty module. Fixed in v0.1.1; v0.1.0 PyPI artifacts remain installable but web4-trust's docstring also incorrectly described tensors as "6-dimensional" (canonical: 3 root dims, fractally extensible via web4:subDimensionOf). Use v0.1.1.

All AGPL-3.0-or-later. Patent grant terms: PATENTS.md. Commercial licensing: [email protected].

web4-core provides LCT (Linked Context Token) primitives + T3/V3 trust tensors + identity coherence + ledger anchoring (InMemoryLedger, LocalLedger). web4-trust-core adds trust persistence and witnessing primitives. The Python wheels are PyO3-built bindings over the same Rust core.

What's working

Layer Status Where
Spec corpus (LCT, T3/V3, MRH, ATP/ADP, R6/R7) Stable web4-standard/core-spec/
web4-core Published v0.1.1 (crates.io + PyPI). LCT, T3/V3, Coherence, Ledger trait + 2 backends (InMemory, Local file). 52 unit tests + 4 doctests. web4-core/
web4-trust-core Published v0.1.1 (crates.io + PyPI). Trust storage, witnessing, decay. web4-trust-core/
Runnable proof of presence python identity_bootstrap.py — bootstraps a host LCT (keypair on disk, hash-chained LocalLedger, public lct.json sidecar); --verify re-checks the chain on re-run. ~30 sec. web4-core/python/examples/identity_bootstrap.py
Cross-language interop demo Python mints an LCT to a hash-chained ledger; a Rust binary reads the same ledger.jsonl and verifies chain integrity + anchor proof. The on-disk format is the contract. web4-core/examples/cross_language_verify/
Reference Python SDK 2,627 tests, mypy --strict clean (not yet on PyPI separately) web4-standard/implementation/
Cognition harness producing 94.85% Open source SAGE
Attack simulation suite 424 vectors / 84 tracks, ~85% detection simulations/
Threat model v2.0 docs/reference/security/THREAT_MODEL.md
Authorization layer PostgreSQL schemas + security mitigations web4-standard/implementation/authorization/
Coordination framework (Phase 2a–2d) Validated web4-standard/implementation/reference/

What's missing (this public repo)

Gap Where it stands Where production lives
Hardware binding reference (TPM 2.0 on real devices) Python AttestationEnvelope shipped; Rust port and on-device integration in progress Hardbound (enterprise, contact [email protected])
Economic attack modeling at scale Empirical defenses only; no real-market testing Open research
Formal Sybil-resistance proofs Empirical defenses only Open research
Production deployment All testing is synthetic Hardbound for regulated environments

Where it landed publicly

Open questions

These are not gaps to fix; they are research questions:

  • Are stake amounts actually deterrent? (no economic modeling at scale)
  • Does witness diversity resist sophisticated cartels?
  • What's the minimal viable Web4 for a public pilot?
  • How does the harness-effect (the 94.85% delta) generalize across reasoning tasks beyond ARC-AGI-3?
  • How does the same harness perform with smaller models?

Pointers for deeper reading

This is the living STATUS.md, kept short and current. Long-form historical detail lives in docs/history/.