Source-run evidence

Returns cited evidence packs for any past ingest run. Use the source-run endpoints when you need to see exactly which source material produced which records, whether for demos, evaluation prompts, audits, or deciding whether to roll a run back.

A source run is the unit of ingestion. Every URL ingest, every uploaded file, every direct CLI write produces a source_run_id. Notes, claims, positions, projects, and edges all carry that id, so any record in Marrow can be traced back to the artefact and the operator action that created it.

This page covers three related read endpoints:

• Inspect tells you what shape a run produced (e.g. "12 notes, 18 claims, 2 new projects, 4 facets, no warnings").
• Evidence packs return the real text excerpts and claim statements with their source kind, dated-at, and confidence. They are ready to drop into a prompt or a demo report.

#Listing runs

curl "$MARROW_API_BASE_URL/v1/source-runs?limit=20" \
  -H "Authorization: Bearer $MARROW_API_KEY"

{
  "schemaVersion": "marrow-source-run-list-v1",
  "runs": [
    {
      "id": "f1a3…",
      "sourceKind": "url",
      "startedAt": "2026-05-12T14:08:22Z",
      "finishedAt": "2026-05-12T14:08:54Z",
      "summary": { "notes": 12, "claims": 18, "projects": 2 }
    }
  ],
  "pagination": { "limit": 20, "returned": 1, "hasMore": false, "nextCursor": null }
}

#Inspecting a run

curl "$MARROW_API_BASE_URL/v1/source-runs/f1a3.../inspect" \
  -H "Authorization: Bearer $MARROW_API_KEY"

Returns counts, predicate frequencies, facet usage, and any quarantine or anchor warnings recorded during distillation.

{
  "schemaVersion": "marrow-source-run-inspect-v1",
  "run": { "id": "f1a3…", "sourceKind": "url", "datedAt": "2026-05-12" },
  "counts": { "notes": 12, "claims": 18, "edges": 33, "facets": 4 },
  "predicates": [
    { "name": "extracted-from", "count": 18 },
    { "name": "demonstrates", "count": 7 }
  ],
  "warnings": []
}

#Generating an evidence pack

Evidence packs are designed for prompt-grade output. They cap word and character counts, prefer high-confidence notes, and link every claim back to the note id it was extracted from.

curl "$MARROW_API_BASE_URL/v1/source-runs/f1a3.../evidence?maxNotes=8&maxClaims=12&maxChars=20000" \
  -H "Authorization: Bearer $MARROW_API_KEY"

{
  "schemaVersion": "marrow-source-run-evidence-v1",
  "run": { "id": "f1a3…", "sourceKind": "url", "datedAt": "2026-05-12" },
  "notes": [
    {
      "id": "9b2c…",
      "excerpt": "On the Northstar team, I drove the onboarding rollout into production launch …",
      "sourceKind": "url",
      "confidenceScore": 0.95,
      "facets": ["cross-functional-delivery"]
    }
  ],
  "claims": [
    {
      "id": "1f0e…",
      "text": "Avery led cross-functional delivery of the Northstar onboarding rollout.",
      "extractedFromNoteIds": ["9b2c…"],
      "confidenceScore": 0.92
    }
  ]
}

#Tuning the pack

#What you get back

• List: a chronological view of runs with summary counts and pagination headers.
• Inspect: shape of a single run, including counts, predicates, facets, and warnings.
• Evidence: real source-grounded text the run produced, capped for prompt use.

#Scope

All three endpoints require an API key with scope evidence. See the Source Runs API Reference for the full schema and pagination semantics.