UNI Universal Natural Intelligence

Wiki · The Encyclopedia

TA-C - Website, self-driving site and measurement

The Encyclopedia · encyclopedia/appendix-TA/TA-C-website-self-driving.md @ 575fc93d9d31 (main) — opens the published snapshot e850f872196d

How to read this page

The Encyclopedia is the UNI method written out as a reference work: 39 pages, arranged in wings, setting out what the programme is attempting and why it is built the way it is. This is where the ideas are explained in order and in prose, rather than as code, as runbooks, or as dated receipts.

Every chapter is authored against two ledgers and never ahead of them. One records what UNI has built, and the evidence class of each claim. The other records nature's own regularities, kept separate on purpose. That way a fact about biology is never quietly reused as a fact about the software. Where a chapter and a ledger disagree, the chapter is the thing that is wrong. Every chapter closes with an invitation to falsify it, and a recorded negative is published beside the result it qualifies rather than after it.

Read "How to read this work" first. It is the evidence constitution: the classes, the four ledger states, and the rule that a finished chapter is not the same as a working system. Then the calibration ledger, which carries the figures every other chapter is required to use.

What it is not: a description of a person or of a mind. The programme calls itself a developmental active-inference simulation, a bounded peek into a toy world, and its own index prints how much of the developmental ladder has actually been earned — roughly two rungs out of eleven or more. It is also not a report of what is running today. For what ran, and when, go to the evidence record.

A Plain and a Clear version of this page have not been written yet. What follows is the document itself.

Precise — the source document

This is the document. Rendered from the repository at the commit above, with nothing rewritten for the web. A gate re-renders it on every deploy and fails the build if a single byte differs.

This chapter is part of the Track-A appendix: the marketing organism, fenced as engineering and method, never as science. It carries the public web presence for the SolutionWright Universal brand family and the first two phases of a planned self-instrumenting site. The honest position up front: what is shipped here is built software and operational discipline. It is not evidence about the UNI science, it does not demonstrate active inference, and a meaningful part of the measurement layer is deliberately switched OFF until an operator sets one environment variable. The strongest single fact in this chapter is also the most easily misread, so it is stated plainly: the site is built to measure itself, but it is not measuring anything yet, by design.

Everything below cites the claim ledger directly (rows TA19 through TA24, Section 4.3) and authors every shipped result alongside the negative or fence that travels with it.

What shipped, with each result paired to its fence

A full site, build-clean (TA19, Class E). The public site is a Next.js 16 application, 26 routes plus an /api/intake endpoint, with next build and tsc both clean and a safety sweep that enforces zero em-dashes and zero "LEAP" in shipped copy. This is a Class-E claim: test-covered and build-covered, not a runtime capability observation. Class E means DONE in the sense of "the build is green and the sweep passes," not "a user did the thing." The falsifier is operable: next build or tsc fails on main, or the sweep finds an em-dash or the string "LEAP" in shipped copy.

Three domains live on Vercel (TA20, Class A). Three domains were cut over to Vercel with DNS verified live: solutionwright.com was flipped indexable in a Production-only environment for the press push, and iamhitl.com was cut over to the Evidence Explorer with the Printify store relocated to store.iamhitl.com. This is Class A, a live observation, not merely a passing test. Its falsifier is a live DNS or HTTP check showing a domain not resolving to Vercel or not serving the intended app, or indexability that is not Production-scoped. The travelling caution from the archive: production runs through a separate Vercel team reachable only via the dashboard, while the MCP token and CLI are bound to an older account that cannot see production. Drive production through the dashboard, not the CLI.

Self-driving Phases 0 and 1, shipped (TA21, Class A for ingest / Class C for the storage gate) - and the load-bearing OFF fence. On 2026-06-25 the first two phases of the self-driving site shipped to production: a storage-agnostic Postgres adapter (Neon as the hot store, the appliance as a sovereign async mirror) and a no-PII, salted-daily-id first-party measurement path that degrades to CANON on DNT or error. The Phase-0 storage gate is a Class-C dev-gate PASS: 2000 of 2000 writes with identical checksums. The sovereign ingest path was proven end-to-end from the open internet: a valid beacon returns 204 with the row inserted PII-stripped, a Do-Not-Track signal produces no insert, a bad origin returns 403, and 8 smoke routes return 200.

That is a genuine, observed result. It must always be cited next to its fence, because the fence is the whole point of this chapter: measurement is OFF. The live beacon no-ops until NEXT_PUBLIC_INGEST_URL is set in the Vercel environment and a redeploy fires. This is a deliberate degrade-to-CANON safety, not a defect: with no ingest URL the site serves a deterministic content rotation (CANON) and captures nothing. So the calibrated statement is: the measurement plumbing was built and proven able to capture, strip PII, and reject correctly under test, and it is currently capturing nothing in production by design (the parked owner action TA-P4). The falsifier for the shipped claim is sharp: the storage gate fails, measurement leaks PII or fails to degrade to CANON on DNT, or a smoke route returns non-200.

The SEO/GEO program (TA22, Class A). Shipped on SolutionWright: a cross-domain JSON-LD @graph with stable @ids, per-site llms.txt and llms-full.txt, a sitemap and robots welcoming AI crawlers, and a Node link-crawler. The brand-family wiring uses one canonical @id per real-world entity, identical on every page and domain, with parentOrganization and founder pointing to the same organization and Person @id, a sameAs ring across the sibling sites, and Claim rather than the retired ClaimReview. Class A: verifiable live. Falsifier: live JSON-LD @ids are not stable or identical across pages, the machine-facing files are missing on a shipped site, or retired schema types are used. The honest companion bound, recorded as a Track-A negative (TA-N7): AI search crawlers rarely fetch llms.txt and Google has said it will not support it. Ship it as a cheap, correct machine-facing gesture for agentic and IDE tools, never as a ranking lever.

The press-readiness audit harness (TA23, method / Class E). A reusable, four-phase parallel-agent audit: per-site SEO/GEO, a cross-domain JSON-LD entity-graph check, five personas (press-journalist, prospective-client, skeptical-scientist, answer-engine for GEO, and accessibility-mobile), and a GO / HOLD synthesis, producing a prioritized P0/P1/P2/polish scorecard. Its defining discipline is that it fetches the LIVE HTML via curl and never asserts from memory. This is a method claim, a reusable governance pattern, not a capability. Its falsifier (the violation): the audit asserts from memory rather than fetching live HTML, or skips a persona or a phase.

The Reading Room, the evidence data-walk, and i18n (TA24, Class E). The /watch "Reading Room" was rebuilt as an interactive all-ages curated library (mood and audience chips, multi-language, watch-later) over roughly 900 or more films. The evidence data-walk tool presents 44 graded primary-research findings with a three-way mode switch and, by deliberate design, makes NO recommendation: it shows the evidence and the trade-offs and lets the reader decide. Internationalization spans 822 keys across 10 languages on the UNI and Evidence Explorer surfaces. Class E. Falsifier: /watch does not render the curated library or its chips are non-functional, the data-walk emits a recommendation or its counts differ from 44 findings and 3 modes, or the live key and language counts differ from 822 and 10.

The discipline that makes this Track-A, not science

Two patterns are worth naming because they are the reusable assets. First, no-PII first-party measurement: a sendBeacon on visibilitychange, validated against a taxonomy allowlist, PII stripped, stamped with a salted daily HMAC visitor id that creates no cross-day linkage and no tracking cookie, with DNT and GPC producing zero capture, absolutely. Second, deterministic global rotation: floor(epochMs / windowMs) % count gives every visitor the same content in a window with no server state, used both for the hero puzzles and for the CANON fallback. The self-driving design is a framing lens only; no such loop is demonstrated on these pages.

What is NOT claimed in TA-C (Website, self-driving site and measurement)

  • Ceiling. That the site is a working, running, self-driving active-inference agent measuring its own audience is NOT shown. The most we claim is: a full Next.js 16 site is built and build-clean (TA19, Class E); three domains are live on Vercel (TA20, Class A); the self-driving Phases 0 and 1 are shipped with a Class-C storage-gate PASS (2000/2000 writes, identical checksums) and a Class-A end-to-end ingest proof (valid -> 204 PII-stripped, DNT -> no insert, bad-origin -> 403, 8 smoke routes 200), with measurement currently switched OFF in production by design; the SEO/GEO program is live (TA22, Class A); the audit harness is a reusable method (TA23); and the Reading Room, the no-recommendation evidence data-walk, and 822-key/10-language i18n are shipped (TA24, Class E).
  • Fences engaged. Red line 3: never "active inference demonstrated" - the self-driving design is the framing lens, not a demonstrated AIF loop. Red line 8: this is engineering and operational evidence, never a science gate. Red line 7: never raise a claim above its ledger class (Class E here is build-covered, not a runtime capability). Red line 10: no PII (the entire measurement design exists to guarantee this), no featured channel handles. Red line 12 / vocabulary-leak guard: this Track-A chapter uses plain ops vocabulary and does not externalize active-inference or free-energy framework names or internal handles.
  • Negatives that travel with this claim (cite alongside, never strip). TA-P4: measurement is built but OFF until NEXT_PUBLIC_INGEST_URL is set and a redeploy fires (degrade-to-CANON), and Forms-P0 still delivers nowhere - /api/intake forwards to INTAKE_WEBHOOK_URL only if set, it is unset, so leads currently go nowhere, and /falsify has no backend. TA-P5: self-driving Phases 2 through 5 (the bandit engine, public posteriors, the reciprocal three-repo entity graph) are planned, not built - Class-U design until shipped. TA-N7: AI crawlers rarely fetch llms.txt; it is a correct gesture, not a ranking lever. The defining program negative N-LEAK (the 2026-06-24 client-data leak) stands behind the never-deploy-without-confirmation rule.
  • Parked / owed. TA-P4 (set the ingest URL and wire intake plus a press email) and TA-P5 (build and observe Phases 2 through 5 live) are owed before any of this rises above "shipped plumbing, idle by design." No sign-to-park is owed for a Track-A engineering chapter; the owed item is an owner action, not a UNI signature.
  • One-line honest summary a skeptic could not dispute. A real, build-clean site is live on three Vercel domains with a proven-under-test, PII-safe measurement path that is currently capturing nothing in production by deliberate switch, and the "self-driving" engine beyond first-party capture is unbuilt design, not a demonstrated self-driving system.

Falsify this

Set NEXT_PUBLIC_INGEST_URL in the Vercel environment, redeploy, and send a beacon: if a valid event does not insert a PII-stripped row returning 204, or a Do-Not-Track signal does not produce zero capture, or a bad origin does not return 403, or the storage gate does not reproduce 2000 of 2000 writes with identical checksums, then the TA21 claim is falsified at the point where it matters most. Equally: run next build and tsc on main and run the safety sweep - any failure, em-dash, or "LEAP" in shipped copy falsifies TA19.

Sources

  • Curated digests (PII-redacted): curated/website-digest.md (self-driving plan, Phase 0/1 ship status, sovereign-ingest proof, Forms-P0 and measurement-OFF parks, the SEO/GEO program, the client-data-leak negative); curated/uni-precision-digest.md (the press-readiness audit harness and the GEO entity-graph discipline).
  • Claim ledger (single source of truth): encyclopedia/CLAIM-LEDGER.md Section 4.3, rows TA19-TA24; with the travelling parks and negatives TA-P4, TA-P5, TA-N7, and the program-defining N-LEAK.
  • Authoring spec: encyclopedia/MASTER-PLAN.md, section TA-C, and Part I front matter (FM-1 Evidence Constitution, FM-2 A-U rubric, FM-3 red lines, FM-4 not-claimed template).
  • Archive pointers (not read here; local, gitignored, PII): the SolutionWright Universal Website archive and the UNI Precision archive.

sha256 82102c5016aee3bc — of the original file, so what was ingested stays checkable.