The Web is rich in knowledge, but agents repeatedly start from pages
Today, an agent often searches, downloads a human interface, runs a browser and asks a model to infer structure again. That work is expensive, difficult to reproduce and easy to detach from its source.
Without TWIRX
- Find a page or provider-specific API.
- Interpret its vocabulary and structure again.
- Lose source wording as data is normalized.
- Repeat the work for the next agent and query.
With TWIRX
- Observe an approved public representation once.
- Preserve its native statement and exact evidence.
- Compile proof-bearing, reusable semantic state.
- Query, compare and trace through one typed fabric.
The protected public Semantic Highway baseline is live
The released local snapshot is deliberately small. It proves the complete route from evidence to packets, materialized views, cross-origin queries and semantic deltas without pretending that every selected Atlas origin already has a working adapter.
- Atlas identities
- 500selected, not all compiled
- Policy decisions
- 3explicit human review
- Public-source packets
- 153 origins
- Real origin deltas
- 1two archive periods
- Materialized views
- 2immutable snapshot
- Runtime origin calls
- 0snapshot query path
A separate 25,018-packet run is capacity evidence using controlled fixtures. It is never added to the public-source count.
Inspect the cross-origin query Explore all 500 identities Read the complete proof
The first large Utility Universe is validated engineering evidence
One founder-authorized Grants.gov bulk observation has been compiled into a privacy-safe Opportunity Universe and combined with the admitted World State frames. This is real source-derived scale, not the earlier controlled capacity corpus. It is a validated engineering release candidate and is not deployed yet.
- Source records
- 83,08746 rejected closed
- Semantic packets
- 1,037,679real source-derived
- Opportunity frames
- 83,087proof-linked objects
- Combined frames
- 83,122Opportunity + World State
- Bounded query p95
- 3.107 ms1,000 local iterations
- Runtime origin calls
- 0immutable query path
Privacy and semantic boundary
247,382 contact fields and
83,130 descriptions were excluded.
59,189 eligibility fields are represented
as proof-linked withheld state with zero lexical values released.
Mappings remain candidate in the
provisional_semantic lane; no eligibility or “currently open”
conclusion is inferred.
It can distinguish a source change from an interpretation change
Two sealed Common Crawl records show the RFC Editor homepage title changing between 2025 and 2026. TWIRX preserves both lexical values and emits an origin delta. It does not invent a semantic delta or canon delta when neither occurred.
Earlier representation
CC-MAIN-2025-30 · 2025-07-08T10:21:38Z
» RFC Editor
html/head/title[1]/text()
Later representation
CC-MAIN-2026-25 · 2026-06-07T20:53:41Z
RFC Editor
html/head/title[1]/text()
One public layer, three direct beneficiaries
Agents
Discover and query typed capabilities without loading hundreds of provider-specific tools or trusting an opaque summary.
Publishers
Retain native vocabulary, attribution, access policy, correction and revocation while becoming agent-readable.
The public Web
Gain an independently implementable semantic layer that is not owned by a browser vendor, model company or private registry.
Current boundary
Genesis Preview
Read-only, approved origins and immutable snapshots only. No arbitrary URL, browser execution, model authority, authentication, payment or write action.
The snapshot, query runtime and proof above are implemented and validated. The live read-only Query Lab serves exactly that immutable state under hard host limits, with no origin network access. The larger Opportunity Utility release is published as verified engineering evidence but remains undeployed pending target-host resource and deployment admission. Known limitations remain visible on Proof and Security.
Fund the next 90 days
TWIRX is requesting $30,000 for 90 days. The first $6,000 milestone funds 30 days of focused, full-time engineering for Ontology Fabric 0.1, expanded Opportunity and World State universes, and a visual Atlas Agent alpha.
Funding buys labor, infrastructure and reliability. It does not buy protocol ownership, semantic ranking, canon admission or governance authority.
Review the FUTO case Budget and milestones Contact the maintainer