# DataShield Ontology — Documentation

Verbose, code-measured documentation for the DataShield Ontology (the Library at library.myorg.ai): the governed data plane that AI agents, developers, analysts and governance teams work through. Every capability statement on these pages was measured against the source tree at v0.22.188 (tag 2b5f46a9) on 2026-09-22 and cites the file and line it came from, so a reader can re-verify any claim rather than trust it. Where the code and a standard or plan disagree, the page says so instead of describing the intent as if it were shipped.

## How to read this set

- **New to the platform:** start with the overview, then the data catalog. Together they explain the one processing chain (providers → connections → crawlers → assets → bindings → datasets → analysis → transforms → publish) and which MCP umbrella owns each stage.
- **Connecting a source:** read the connectors page for the two connection paths, the credential shape per kind, what discover, profile and extract do, and the typed refusals a dial can return.
- **Working with data:** the datasets page covers ingest, the analysis engines, derivative datasets, the declarative transform model and what it does and does not materialise.
- **Governance and privacy teams:** the governance page renders STANDARD-009 in plain language, then the classification kernel, masking posture, stewardship, MDM and the evidence surfaces.
- **Operating or contributing:** the development page documents the job system, the release process as it actually runs, the gate chain, the bug registry and the Development section roadmap.
- **Building an agent:** the agent guide gives the first call, tiers, the error envelope, every umbrella grouped by task, and seven task recipes.

## Conventions

- **RFC 2119** keywords (MUST, SHOULD, MAY) carry their normative meaning.
- Timestamps are **RFC 3339**; durations are **ISO 8601**.
- A `file:line` in backticks is a citation into the Library source tree at the stated release.
- A capability is labelled `real`, `simulator`, `hardcoded`, `stub` or `absent` per STANDARD-005 §13; only `real` means it runs in production today.
- Items marked **PLANNED** or **DESIGN** are not shipped.

## For AI assistants and search tools

Every page is also served as raw Markdown at its own URL plus `.md`, and the set is listed in the site's `llms.txt` and sitemap. The pages are self-contained: an agent that reads one page has the vocabulary and the command names it needs for that area without loading the others.

## Maintaining these pages

The source of truth is this directory in the Library repository (`docs/ontology-documentation/`). The public site imports it with `node site/tools/import-docs.mjs` (configuration in the site's `docs-import.config.json`) followed by the site build; a page whose measured release is superseded should be re-measured and re-imported rather than edited in place on the site.