The prototypes — public, and marked for how real they are.

The canonical LinkML grammar — every node and edge on this site is defined in it. SHACL, Turtle and JSON-LD are generated from the source.

What to try: read mira.yaml, then run the Makefile to regenerate SHACL and validate a graph against it.

The interactive schema-demo deck: the six-node grammar built one record at a time on the discovery of DNA’s structure, with every slot and edge explaining itself on hover.

What to try: scroll it end to end, then hover an edge label to read the relation both ways.

AI-assisted MIRAfication of a source. Emits schema-conformant JSON-LD (<name>.mira.jsonld) — the wire format the viewers and the KOI envelope speak — with a verbatim provenance.excerpt on every node (the receipts), dangling and ungrammatical edges reported instead of dropped, and the source → study → grounds → evidence spine enforced.

What to try: MIRAfy a source and diff the Initial AI draft nodes against what a human would advance; the emitted .mira.jsonld is what moves downstream.

A ~314-node discourse graph plus a standalone d3 viewer and the Python transform pipeline that built it. The data is mock — terms are randomized to a microtubule-transport field; use it to validate tooling, never as a scientific source.

What to try: open the d3 viewer and traverse from a Question to its Evidence; the same grammar, rendered and navigable.

The TypeScript MyST parser — authoring discourse-graph nodes inside MyST Markdown and emitting JSON-LD. One authoring tool to one viewer: real, and narrow.

What to try: npm test (vitest) — the MyST → JSON-LD round-trip on the golden fixture.

The KOI cross-organization envelope — operational. Different orgs’ agents and tools read from and write to the same KOI node, and it ingests the JSON-LD that MIRA-extraction emits. Our working MIRA KOI node contains unpublished scientific results by our users, and so isn’t yet public — and even there, each record is a pointer to its artifact, not a copy of it. Still designed, not built: federation across schemas and a public network at scale.

What to try: stand up your own KOI node to push and pull MIRA nodes across the network.

The north-star user story — “push a result to a shared web interface” — with the normative transport/sharing rules R1–R13 and the authoritative open-questions list.

What to try: read AGENTS.md; it is the brief the whole approach is measured against.

The same grammar running in public on two unrelated topics — a healthcare-language graph and a whitepaper-as-a-graph. Both give every node a URL and generate their narrative as a traversal: the prose is a rendering, the graph is the source.

What to try: open either and read a node, then follow its edges to the evidence.

A draft ATProto lexicon binding (science.mira.*) for public discourse-graph nodes — the intended public layer. Self-labeled DRAFT / PROPOSAL: not adopted, not published, not on the network.

What to try: read the lexicons as a concrete direction; treat them as a proposal pending schema-team review, not a shipped surface.