round-trip/ — <name>.json + <name>.md: the markdown adfToLosslessMarkdown must emit for that document, byte for byte, and that losslessMarkdownToAdf must read back to it (docs/decisions.md §The round-trip is the product). Grouped by what the fixture exercises.
normalization/ — <name>.md + <name>.json: markdown input, and the document losslessMarkdownToAdf must build from it, which must in turn emit and read back to itself. The markdown is not canonical.
errors/ — <name>.md: markdown input that must not convert. A <name>.error beside it pins which error.
real-payloads/ — <name>.json: ADF Atlassian's editor wrote, round-tripped ADF→markdown→ADF and ADF→HTML→ADF. No expected markdown or HTML.
real-world/ — <name>.json: anonymized documents from public Jira and Confluence sites, round-tripped as real-payloads/ is. real-world-sources.json names each one's original page, licence and retrieval time. corpus-gathering/run.sh <folder outside the repo> fetches the documents into that folder, then replaces the whole set and that file. It needs Docker and network, and takes minutes (docs/decisions.md §The real-world corpus holds anonymized documents from sites whose licence it records). typesNoFetchedDocumentHolds in src/conformance/corpus.test.ts lists the node types these documents don't show.
commonmark-spec/ — the CommonMark suite run against commonMarkToAdf by three checks. spec.json is the suite; refusals.json pins each refusing example to its error code; exceptions.json pins each known divergence by check, example, kind and the exact divergence, with a reason. kind is mark-model (the permanent count divergence from ADF's mark-per-node model), node-model (a permanent divergence where ADF holds no node for what CommonMark renders) or pending (a parser gap).

JSON is two-space indent, keys sorted, and a document read back must deep-equal the fixture's (docs/decisions.md §Equality is deep). spec.json is the vendored, upstream machine-readable suite, byte-exact from spec.commonmark.org (CommonMark 0.31.2, © John MacFarlane, CC-BY-SA-4.0), and is not re-serialized by the corpus gate.

real-world/ isn't under this repository's MIT licence. Each document adapts the work its entry's url names, replacing its text, names, ids and URLs, and that page credits its creators, whom the documents don't name. It stays under the licence its entry names, CC-BY-3.0 or CC-BY-SA-4.0, without warranty, and an adaptation of a CC-BY-SA-4.0 document is shared under that licence. Only the ONAP wiki states its own licence. The others rest on the Linux Foundation's terms of use, which don't name these sites, so their licence is our reading of those terms.