# The machine hub

*Vercy for AI agents · lesson 3 of 4 · ~10 min*

## What you will learn

The machine-readable surface of the standard and of conformant models: what to fetch, validate against, and cite.

## The principle: every human page has a machine twin

The standard is built AI-first: anything a human can read has a machine-readable counterpart that an agent can fetch, parse and verify. Nothing meaningful is locked inside prose.

## The spec's own machine surface

At [ver.cy](https://ver.cy) and in the [repository](https://github.com/ver-cy/meta-universe):

| Artifact | What an agent does with it |
|----------|----------------------------|
| `/llms.txt` | The AI entry point: what this is, where everything lives |
| `/spec-index.yaml` | Enumerate all 90 documents: id, class, status, path |
| `REQUIREMENTS-INDEX.md` | Address any of 1284 normative requirements by stable ID (`ARCH017-R12`); cite these, not prose |
| `/schemas/` (10 JSON Schemas) | Validate MUIF models, MUFP envelopes, discovery documents, validation reports |
| `/external-models.csv` | 1180 external standards with compositional roles: query before modelling |
| `AGENTS.md` | Reading order and rules of engagement for this repo |
| `tools/` | Reference implementations: fingerprint, validate, index generators |

Two habits worth hard-coding: resolve terminology from `Terminology.md` (never infer normative meaning from informative prose), and treat a matching **Semantic Fingerprint** as the definition of "same model" (bytes lie, fingerprints do not).

## A conformant model's machine surface

Every conformant model repeats the pattern in miniature:

- `manifest.yaml`: enumerate and classify everything (the model's spec-index).
- `sources.yaml`: authority and freshness for every dataset.
- `artifacts/walk-report.md`: the latest coverage proof: read it instead of re-deriving trust.
- Provenance sidecars in `raw/`: what was harvested, from where, when.
- Registry files of its own ecosystem (e.g. AISMM's `aismm.registry.json`): identity and completeness.

An agent that consumes these files first can answer "what is this model, is it complete, what here is authoritative, how fresh is the rest?" without reading a single record: and knows exactly which records to read next.

## Live endpoints as projections

Models with write-back projections expose machine endpoints on the open web: the Orkestron.AI model publishes `https://orkestron.ai/product-facts.json` (generated, marked, content-hashed) and an `/llms.txt` beside it. For an agent, such endpoints are Projections in the full standard sense: consume them knowing the master is the model, verify via the embedded hash, and never treat them as editable.

## Discovery across universes

At federation scale the pattern generalizes: a Universe publishes `/.well-known/meta-universe.json`: who it is, which models it exposes, how to begin a MUFP conversation. Discovery documents have a JSON Schema like everything else; an agent validates before trusting.

## Key takeaways

- Spec side: llms.txt + spec-index + requirements index + schemas + registry CSV = everything enumerable, validatable, citable.
- Model side: manifest + register + walk report + provenance = trust without re-derivation.
- Public projections and discovery documents are the same idea on the open web: fetch, verify, never edit.

## Go deeper

- [ver.cy/llms.txt](https://ver.cy/llms.txt) · [spec-index.yaml](https://ver.cy/spec-index.yaml)
- [MMAS interchange format and schemas](/spec/#02-architecture/MMAS-Interchange.md)
- [Discovery](/spec/#03-federation/Discovery.md)

Next: [Agents across universes](04-federation.md)