fixtures

Fixtures — golden corpus

Language-neutral conformance corpus. Every implementation runs compress() over each fixture's input and diffs the output against the goldens here. See ../spec/SPEC.md for the contract.

Layout

fixtures/
  corpus/
    <fixture-id>/
      input.json     # the Grid (rows + origin + optional cellMeta + charts)
      meta.json      # human-readable: title + description; used by tooling
      golden/
        anchor.string.txt                       # SPEC §3.2 string form, no trailing newline
        anchor.json                             # SPEC §3.3 JSON form, 2-space indent + trailing newline
        anchor.tokenEstimate.txt                # integer + trailing newline
        invertedIndex.string.txt                # SPEC §4.4 string form, no trailing newline
        invertedIndex.json                      # SPEC §4.5 JSON form, 2-space indent + trailing newline
        invertedIndex.tokenEstimate.txt         # integer + trailing newline
        formatAggregation.string.txt            # string form, no trailing newline
        formatAggregation.json                  # JSON form, 2-space indent + trailing newline
        formatAggregation.tokenEstimate.txt
        charts.json                             # SPEC §6.2 echo of grid.charts, 2-space indent + trailing newline
        rawBaseline.tokenEstimate.txt

Each golden file is byte-comparable. anchor.string.txt deliberately has no trailing newline so the file matches result.encodings.anchor.string exactly. The *.tokenEstimate.txt files each contain one integer followed by a single \n so they're sensible in a text editor.

input.json schema

{
  "rows": [["A", "B"], ["C", "D"]],
  "origin": { "row": 1, "col": 1 },
  "cellMeta": null,
  "charts": null
}

cellMeta MAY be omitted or null; v0 ignores it in the output but accepts it to lock in the Phase-2 contract. charts MAY be omitted or null; when present, each descriptor renders as a CHART(...) token appended to every encoding's string form (see SPEC §6) and is echoed verbatim in charts.json.

meta.json schema

{
  "id": "tiny-table",
  "title": "Tiny three-column table",
  "description": "Smoke-test fixture: 3×3 grid with one fully-empty row."
}

id matches the fixture directory name.

Regenerating goldens

From packages/typescript/:

npm run generate-goldens

The script reads every corpus/<id>/input.json, runs compress(), and overwrites the golden files. Hand-editing goldens is not supported — change compress() (or the SPEC), then regenerate.

The conformance test (npm test) refuses to regenerate; it only diffs. So the regenerate step is an explicit, reviewable action.