Turn supplied evidence rows into a citation table with traceable counts

A citation table becomes difficult to audit when its Markdown, JSON rows, and summary counts drift apart. Rather than maintaining three hand-edited versions, generate them from one explicit packet. The useful outcome is a reviewable representation of what you supplied: stable evidence identifiers, attributable excerpts, and counts whose meaning remains clear. Counting ten references is different from independently verifying ten claims.

I maintain the us-brand-evidence-citation-table Actor. This walkthrough describes its published build 0.2.2, inspected on 16 October 2026. The example below is synthetic and reviewed against that contract; it was not executed for this article. AI-assisted drafting was followed by a separate factual and editorial review.

This Actor accepts input schema 1.0 in the inspected build. Each evidence row has seven required fields. The source URL is attribution text, not a request for the Actor to visit a page. Here is a complete synthetic row using an example.com address. Replace the invented content with material you are entitled to process, preserve the evidence ID, and keep the original source record beside the generated table.

{
  "schemaVersion": "1.0",
  "rows": [
    {
      "evidenceId": "ev-1",
      "brandName": "Example Brand",
      "eventType": "news_item",
      "observedAt": "2026-10-01T08:00:00Z",
      "sourceName": "Synthetic note",
      "sourceUrl": "https://example.com/e1",
      "excerpt": "Synthetic buyer-supplied observation."
    }
  ]
}

The output includes structured citations, rendered citationMarkdown, hostCounts, and sourceCounts. Host counts group the submitted references by host; source counts reflect the supplied source labels. Neither count represents an independently corroborated source inventory. Two labels can point to the same host, and a reachable-looking URL can still be stale or inaccessible. Interpret the counts according to the input you actually provided.

For Markdown, the formatter escapes pipes and backslashes so those characters do not casually become extra table columns or altered escape sequences. That is a defined formatting behavior, not a promise that every downstream renderer is safe for arbitrary embedding. Preview the produced Markdown in your actual report surface, particularly when excerpts contain punctuation, line breaks, or characters with special meaning in that surface.

The row order is derived from observation time, brand name, event type, source name, URL, and evidence ID. Keep that ordering when comparing generated artifacts. The report exposes separate digests for citations, host counts, source counts, Markdown, and the overall result. They let you identify which representation changed without assuming that equal-looking tables have identical bytes.

The output observation timestamp is derived from the latest submitted timestamp. It is not a fresh page check. Similarly, structural confidence values concern the computation over the packet, not the credibility of the excerpt. The freshness information must be read with those limitations rather than turned into a current-source guarantee.

For a small editorial workflow, give the reviewer both the JSON citations and the Markdown preview. Ask them to sample an evidence ID, locate its original record, compare the excerpt, and confirm how it should be cited. Corrections belong in the source packet so the next generated table and its counts stay aligned. Keep any factual acceptance decision as a separate review artifact.

The input is limited to 100 unique evidence IDs and 128 KiB in raw and normalized form. The report and Markdown have separate 64 KiB and 48 KiB limits. Oversized output fails rather than clipping the tail of the bibliography. If you partition a larger collection, label the partitions and avoid adding their host counts as though they were already deduplicated across the entire corpus.

On the observed FREE pricing tier, one result-found event costs $0.01 for the complete useful report or plan, rather than each input row. FREE is the tier name, not a promise of free execution. Automatic Actor-start pricing is also present at $0.005 per GB, with a minimum of one event. Check the Actor page and your account tier before running; this is not a fixed total invoice, and a failed or repeated run may still incur applicable charges.

To try this deliberately, open the Actor page, select build 0.2.2, and submit the JSON as input. Set an explicit spending ceiling appropriate to your account; for example, $0.05 is a ceiling, not a quoted run price or a guarantee of a delivered report. Save the returned run ID and verify its immutable build ID. A local timeout does not cancel the remote run: inspect that same run before considering another attempt. Read the completed run's Dataset report and its separate OUTPUT receipt together; structural output is not independent source verification.