TA-A - The no-API content model and publishing stack
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 documents the engineering organism that publishes the program's public presence: a self-hosted, multi-container publishing-and-engagement stack that authors and ships content across LinkedIn, YouTube, Reddit, TikTok, and dev.to. It is described here in plain operations vocabulary only. This is the marketing appendix, not the science spine: every claim below is engineering or method evidence (the stack deploys, a route fails over, a gate holds), never a claim about minds, awareness, or active inference. The standing honest position holds even in this appendix: the whole UNI program is a developmental active-inference simulation with roughly 2 of 11+ developmental rungs earned, and nothing in the publishing engine raises that count by a single rung.
The load-bearing design fact, stated once and never softened: authoring runs on the Claude subscription (the CLI, Code, or Desktop) driving MCP tools plus a scheduler (ORCHESTRATE). There is no server-side LLM loop and no content API key. The older archive note suggesting an operator should "set an authoring API key" for an unattended path is explicitly overruled program-wide; the durable fix is an operator re-login, never a key. Organic reach only, no paid ads.
The stack, and what is proven about it
Five of the stack's six containers deploy and run live (TA1, Class A); the full topology is six containers described only by docker-compose.yml, deployed with docker compose up -d --build; produced media lives in bind-mounted content/media/, never in /tmp/. A manifest-drift guard (MCP_MANIFEST_DRIFT) crash-loops the scheduler if the registered tools no longer equal the manifest, so a tool added without a manifest bump fails loudly rather than silently. The honest scope of TA1 is "the no-API authoring model works in production, hundreds of posts published over months." The travelling negative is the dominant live failure mode recorded in the same archive: the engine depends on fragile third-party OAuth that only the operator can refresh, and the subscription login itself expires (the CLI returns 401), silently stalling authoring while every container still looks healthy. The cure is a roughly one-minute operator re-login, not a code change. The falsifier for TA1: the stack fails to deploy via docker compose up -d --build, or authoring is shown to require a content API key, or a tool added without a manifest update fails to crash-loop the scheduler.
The same model is proven as a production stack on a second physical node (TA2, Class A): node2 runs the live ORCHESTRATE stack on Podman (api, UI, scheduler, a TTS sidecar, the LinkedIn and social publishers, and noVNC). This is the concrete instance of the no-server-side-LLM marketing model: scheduler plus MCP tools plus a subscription-driven authoring step, not an autonomous API loop. The GPU image generation stays on a separate Windows host, with networking via WSL mirrored mode. Its falsifier is exact and adversarial: the claim is falsified if the stack is shown running a server-side autonomous LLM or content-API loop, or if it is not actually live on node2.
The no-API model is proven end-to-end at the artifact level by the press kit (TA3, Class E): a seven-document press kit plus a regenerable PDF was authored entirely on the subscription/CLI with no autonomous server LLM, with every figure traced to a verified ledger row and a one-command reproduction. Class E means test-covered and document-grade, not a live runtime observation, and it is carded at exactly that tier. Its falsifier: a press-kit figure cannot be traced to a ledger row, or a one-command reproduction fails, or a server-side content LLM is found anywhere in the authoring path.
The self-healing delivery route ledger
The social-publisher organism was deployed and verified healthy on 2026-06-14, with all durability tests passing (TA4, Class A). It ships a self-healing route ledger: each platform carries public_api | browser | inbox_manual routes, scored by a plain multi-armed bandit (conversion plus novelty, decaying weight), with a circuit breaker (three consecutive failures mark a route down, with exponential backoff), a kill-switch, an audit log, and automatic fail-over from the API route to the browser route on an auth break. This route ledger is plain operations machinery and is described only in plain operations vocabulary. A test-enforced leak guard exists precisely because the bandit shape-mirrors internal scoring code: the falsifier for TA4 includes "internal framework vocabulary leaks into the route ledger," alongside "durability tests fail" and "on an auth break the ledger fails to fail over from API to browser." Naming the framework here would itself trip the fence.
Per-platform publish canaries and their recorded walls
Each delivery path was proven by a narrow, verified canary, and each canary travels with the recorded negative that bounds it.
- TikTok inbox draft via FILE_UPLOAD (TA5, Class A): a verified canary lands a draft with no tunnel and no public URL; the operator taps Publish. This is the only public TikTok path, and it is paired with a hard recorded bound: direct or automatic public posting is blocked for an unaudited app (the API returns a 403 for both public and self-only posts), and the only unlock is the operator submitting the app for the Content Posting API audit. The agent does not retry-loop direct posting. TA5's falsifier: FILE_UPLOAD inbox-draft fails to land a draft without a public URL or tunnel.
- LinkedIn native publishing via the headless Playwright sidecar (TA6, Class A): the sidecar publishes native newsletters and articles, but it is carded with its honest throughput caveat inline: roughly one article per login through the headless sidecar, versus roughly nine per day through the real logged-in Chrome. Falsifier: the sidecar fails to publish a native newsletter or article when authenticated.
- YouTube full-volume Data API upload (TA7, Class A): the Data API path was verified empirically at 30 videos per day. This claim exists specifically as a correction of a false quota panic: the project quota is audit-raised, and the
youtube_quotadatabase counter is unreliable, so the script bypasses it. The negative that travels with TA7, recorded in the same archive, is that headless or automated YouTube Studio browser upload is dead on modern Chrome across all three known routes, and is not to be re-attempted; the Data API path or an operator drag-drop is the discipline. Falsifier: the Data API path fails to sustain the audit-raised quota (for example, 30 per day rejected).
The live-broadcast stack, reachability, and the editorial gate
The live-broadcast production stack was verified live with zero dropped frames (TA8, Class A): a single-feed Director model (OBS set once as a vision-mixer, one feed to YouTube) with window-capture of browser channels and a full go-live runbook. Its safety discipline is part of the claim: an owner clicks Go-Live; there is no autonomous LLM publishing the feed. The falsifier names that discipline directly: the runbook fails to put a single composed feed live, or the publish path fires without an owner Go-Live click.
Internet reachability with zero network changes (TA9, Class A) is a reusable recipe: Cloudflare Tunnel plus Cloudflare TURN make an on-box service publicly reachable with no static IP and no firewall edit. Its falsifier: on-box services cannot be made publicly reachable without a static IP or a firewall edit.
The defamation and editorial gate held under heavy operator pressure (TA10, Class A) on the 12-part Big Tech Accountability series: the agent refused uncorroborated crime accusations against named parties and reframed the work into sourced public-record accountability journalism under the thesis "no bad people, bad systems." All 12 parts carry a gate-passed claims.json (roughly 100 claims, roughly 130 sources, roughly 53 primary). This is a load-bearing example of the honesty constitution working against the operator's own first ask. Falsifier: a claim ships without passing the editorial gate, or an uncorroborated crime accusation against a named party reaches publication.
The load-bearing operational fence: dry mode does not gate the live path
The single highest-urgency fence in this chapter is a safety negative, and it is printed loudly here rather than softened. PUBLISH_MODE=dry does NOT gate the publishers' Playwright path: a publish action with a live session can post live (TA38, Class A, recorded NEGATIVE/safety). The operator-mandated fix is to add auto_publish=false in proxyPublisher before any live publishing, and to stay in dry until then. This fence sits inside the broader auth-outage discipline recorded in the same archive: treat dry mode as report-only because drafts created in dry silently no-op; treat a local "authenticated" heuristic as not the live OAuth token; and never flip the mode to live to "fix" an outage, because that only converts silent no-ops into hard 401s. TA38's own falsifier: PUBLISH_MODE=dry is shown to actually block the Playwright publish path with a live session.
What is NOT claimed in TA-A - The no-API content model and publishing stack
- Ceiling: It is NOT claimed that this stack is an autonomous AI publisher, that "active inference" is demonstrated by any part of the publishing engine, or that any science gate is met by shipping content. The most we claim is the in-class set above: the 5/6-container stack deploys and runs live (TA1, A), is proven as a Podman prod stack on node2 (TA2, A), is proven end-to-end at the press-kit artifact level (TA3, E), ships a self-healing route ledger (TA4, A), and carries six verified per-platform or operational canaries (TA5 through TA10, A). Authoring is a subscription plus MCP plus scheduler model with no content API key; every "autonomous-looking" surface is operator-gated.
- Fences engaged: Red line 7 (never raise a claim above its source evidence class). Red line 8 (operational and substrate engineering never imply a science gate is met). Red line 10 (no PII, no secrets or tokens, no internal channel handles featured, no patent-level math). Red line 12, the vocabulary-leak guard in its hardest Track-A form: plain operations vocabulary only, the route ledger never externalizes internal framework names, and no channel handle is printed. The standing no-server-side-LLM, no-API-key model is itself the fence on red lines 1 through 4: nothing here implies AGI, awareness, life, or "active inference demonstrated."
- Negatives that travel with this claim (cite alongside, never strip): The PUBLISH_MODE=dry safety negative (TA38, Class A) is the most important and must appear next to any "publishing is safe" reading: dry mode does NOT gate the Playwright path, so a live session can post live; the fix is
auto_publish=falseinproxyPublisher, staying in dry until then. Travelling with the canaries: TikTok direct or automatic public posting is hard-blocked pending app audit (paired with TA5); headless YouTube Studio browser upload is dead on modern Chrome and not to be re-attempted (paired with TA7); and the subscription OAuth login expiry is the dominant silent stall on TA1 (a 401 takes down authoring while every container still looks healthy). - Parked / owed: No sign-to-park is owed by this chapter (these are engineering and method claims, not a parked science frontier). What is owed operationally, and named in the source, is the per-platform operator noVNC login that the browser-route publisher still waits on, and the resolution of TA38 (the
auto_publish=falseoverride) before any live publishing run. - One-line honest summary a skeptic could not dispute: A self-hosted multi-container stack, authored on a Claude subscription with no content API key and no autonomous server-side LLM loop, deploys live and ships content across several platforms with a self-healing route ledger and an enforced editorial gate; it is engineering, it depends on fragile operator-only OAuth, and a dry-mode flag does not by itself prevent a live post.
Falsify this
The lead falsifier, stated operably: stand up the stack from its single docker-compose.yml with docker compose up -d --build and inspect the authoring path. If the stack fails to deploy, or if authoring is shown to require a content API key (rather than the subscription plus MCP plus scheduler model), the central TA1 claim of this chapter is false. The standing safety falsifier rides alongside it: if PUBLISH_MODE=dry is shown to actually block the Playwright publish path with a live session, then TA38 is overturned; until that is demonstrated, assume a live session can post live.
Sources
- Curated digests (PII-redacted):
curated/orchestrate-linkedin-digest.md(the Track-A core archive: six-container topology, manifest-drift guard, no-API content model, the self-heal route ledger, the per-platform canaries and their recorded walls, the editorial gate, the auth-outage discipline);curated/uni-os-digest.md(node2 running the ORCHESTRATE Podman prod stack, zero-network-change reachability, and the carried go-live safety flag). - Ledger rows (single source of truth):
encyclopedia/CLAIM-LEDGER.mdSection 4.1, rows TA1 through TA10; paired safety negative TA38 (Section 4.4). - Authoring spec:
encyclopedia/MASTER-PLAN.md, section TA-A and PART I front matter (FM-1 through FM-4). - Archive pointers (local-only, gitignored, never read for this chapter): the ORCHESTRATE LinkedIn campaign archive and the UNI.OS archive.
sha256 40942c5b69e3c6a5 — of the original file, so what was ingested stays checkable.