Schema-bound Text Extraction API

$249/month

billed monthly, cancel any time

Checkout for this service opens by emailed invoice link. Tell us where to send it and we reply the same working day.

Free specimen and paid-delivery boundary

A complete batch response for one real public-record corpus. Every sealed input row appears once with five typed field outcomes and its own source-row provenance; unsupported paths would remain unresolved.

Specimen provenance: Captured: 2026-08-21 · Target: all 100 Federal Register document rows in the sealed 2026-08-21 collection run · Evidence method: five-field scalar extraction over canonical retained JSON with exact in-input quotes and null-preserving outcomes

Proof scope: method boundary only. This free specimen does not demonstrate the full paid delivery.

What this free specimen actually shows

• One retained string/null extraction batch with no dropped rows
• Per-field resolved, source-null, and unresolved outcomes
• Recomputed input, row, content, and collection-batch seals

Paid outputs not demonstrated by this specimen

• Number output type
• Integer output type
• Boolean output type
• Monthly metered API provisioning and quota enforcement

Paid delivery contract

You send:

• One authorized UTF-8 JSON text per call
• One caller-defined scalar schema; this specimen requests five fields
• The sealed Federal Register run used here contains no fetched-at-build input

You get back:

• One output row for every input row, including unresolved outcomes
• A typed scalar or null and an exact source quote or null for every field
• Per-row input digest, source row/content seals, snapshot date, and collection time

Limits: This specimen covers one 100-row, 5-field public-record batch. It performs no OCR or URL fetching and makes no inference when the source is null or a path is unsupported. Titles and other person-bearing free text are not projected. This corpus visibly exercises string and null outputs; it does not evidence number, integer, or boolean outputs. It is not legal advice, a completeness claim about the Federal Register, or workflow actuation.

Open the Federal Register clock read-only and select the newest retained successful source receipt. Recompute its body digest and collection batch, then reproduce every document row in that run from the retained response. For each row, recompute the row and content seals, traverse the five exact schema paths, require scalar types, and copy a key-value fragment that occurs byte-for-byte in the canonical input. Reconcile input, output, row-status, and field-status totals before returning. This is bounded evidence of the dated method stated here; it does not establish affiliation, endorsement, authorization, or any current outcome beyond that scope.