{
  "name": "CAAIL: Cellular Agriculture AI Library",
  "corpusDate": "2026-10-30",
  "status": "beta",
  "placementsUnderReview": true,
  "canonical": "https://github.com/tucca-cellag/caail",
  "site": "https://caail.tufts.edu/",
  "license": "MIT (CAAIL curation). Linked third-party resources keep their own licenses.",
  "scopeNote": "CAAIL is a curated selection of the literature, not a complete record of the field. An empty cell means no INDEXED paper occupies it as of the corpus date (the corpusDate field). That is a limit of the index, not evidence that no such work exists.",
  "openapi": "openapi.json",
  "placementNote": "Placements are proposed by a language-model agent reading the paper's methods section, checked by a separate reviewer agent that must quote the passage it relies on, and entered by a human maintainer; placements made before that pipeline are being re-verified against full texts. Treat a cell as a substantive claim about the paper's method and application area. Where a placement is off, it is typically a closely related cell rather than a paper that does not belong, though re-verification has taken some papers out of the matrix. Cite the paper itself. How this is done, and how far it reaches: https://caail.tufts.edu/curation/",
  "counts": {
    "papersAllSections": 348,
    "papersBySection": {
      "References": 230,
      "Sensory & Flavor Reference Work": 12,
      "Metabolic Reference Work": 9,
      "Foundational Methods Reference Work": 9,
      "Livestock Functional Genomics Reference Work": 12,
      "Reviews & Perspectives": 76
    },
    "papersMatrixEligible": 230,
    "matrixTotalCells": 216,
    "matrixPopulatedCells": 78,
    "matrixEmptyCells": 138,
    "datasetsCurated": 63,
    "datasetsInventoryRows": 172,
    "datasetsTotal": 235,
    "software": 143,
    "databases": 149,
    "catalogTotal": 292,
    "fieldReportEditions": 9
  },
  "endpoints": [
    {
      "path": "index.json",
      "use": "This manifest: corpus date, counts by population, endpoint list."
    },
    {
      "path": "matrix.json",
      "use": "All method×area cells, empties included. Use to ask what has and has not been indexed — then read `matrixNote` in papers-index.json before reading a silence as an absence."
    },
    {
      "path": "papers-index.json",
      "use": "START HERE for papers. Every reference, one compact row: id, title, year, first author, section, methods, areas, DOI, code/data flags. A sixth the size of papers.json, so it survives a fetch tool that truncates. Enumerate here, then fetch full records."
    },
    {
      "path": "papers.json",
      "use": "Full records for the references you selected. Large (~554 KB): if your fetch summarises rather than returning bytes, a \"not found\" from this file is not evidence of absence."
    },
    {
      "path": "catalog-index.json",
      "use": "START HERE for software and databases. One compact row each: slug, name, kind, group, URL, DOI, license tier. A tenth the size of catalog.json."
    },
    {
      "path": "catalog.json",
      "use": "Full records with summaries. Large (~576 KB), two thirds of it the same summary prose as Markdown and as HTML; same truncation caveat as papers.json."
    },
    {
      "path": "datasets.json",
      "use": "Two arrays, together the whole dataset corpus. `entries` = curated dataset entries (portals, atlases, GEMs, reference corpora, and the AI/ML benchmark and evaluation datasets on page \"Benchmarks\"; kind atlas/gem/other). `inventory` = the per-species inventory rows (kind \"inventory\") — the per-study deposits with accession, tissue, assay type and size, keyed by the source page's own column labels. Filter either by `page` (e.g. \"Cow\", \"Benchmarks\"). Use the inventory rows for \"what could I combine my own run with\", and page \"Benchmarks\" for \"what could I evaluate a model against\"."
    },
    {
      "path": "reports.json",
      "use": "Field reports: recurring institutional state-of-field surveys (GFI State of the Industry and the like), one record per edition. When recommending one, use the record with `current: true` for its series and treat any record with a `supersededBy` id as historical only; `seriesEditions` walks the history. The latest edition is derived, so it stays correct as new editions land."
    },
    {
      "path": "topics.json",
      "use": "Subject tree plus an inverted index: topic → items across all content types. Start here for \"what should I use for X\"."
    },
    {
      "path": "taxonomy.json",
      "use": "What each method, area and subject theme means in CAAIL, with exclusion criteria. Read before trusting a placement. Definitions are split by vocabulary under `axes` (`area`, `method`, `theme`) because a label may appear in more than one — \"Bioprocess & Scale-Up\" is both a matrix column and a theme, with different text. `definitions` is the flat matrix lookup (areas + methods only); look a theme up under `axes.theme`. `searchTerms` is the curated search vocabulary per area and method, plus `methodGeneric` for text that applies some AI method (a generic term also listed under a method labels that method too); follow its `contract` when matching."
    },
    {
      "path": "openapi.json",
      "use": "OpenAPI 3.1 description of every endpoint above, generated from the schemas that validate them. Read this instead of guessing a key."
    }
  ]
}