Data Extraction proof

Three verticals · reviewed live evidence

Follow every value back to the page.

On three selected one-page authentic public forms populated with privacy-safe demo values, one live Nutrient Data Extraction API response per form exactly matched 26 of 30 predeclared fields: mortgage 6/8, insurance 11/11, and healthcare 9/11. All 30 returned fields included valid page-bounded primary source regions, and four mismatches were retained for review.

Inspect the three proofs See the evidence chain
3selected authentic public forms
26/30predeclared fields matched exactly
30/30primary regions returned and page-bounded
4retained comparison issues

The differentiator

The proof is the chain.

A plausible value is not enough. Each package keeps the document, extracted JSON value, source location, and independent oracle result connected so a developer can inspect where the answer came from.

  1. Document

    A selected authentic public form carries only privacy-safe demo values.

    PDF · page 1

  2. JSON value

    The schema requests a typed field and the live response returns its value.

    $.field → value

  3. Primary source region

    The response returns a page index and primary box that a reviewer can inspect.

    pageIndex + bbox

  4. Oracle status

    A predeclared expected value decides whether the field is exact or retained for review.

    JSON-value match

Reviewed proof pages

Three documents. Three concrete stories.

Every proof opens locally and displays one receipt-bound live response beside its selected authentic public form and predeclared expected values.

Reviewed mortgage extraction proof showing an authentic public RMA form beside eight compared fields and two retained issues
Current reviewed proof · Treasury RMA form · privacy-safe demo values
01 / MortgageReview blocked

Two punctuation mismatches, preserved for review.

Six of eight values matched exactly. The borrower-name region exposes one omitted period; the servicer primary and nested regions clip the decisive final period, so the expected-versus-actual ledger carries that evidence.

Exact values
6/8
Primary regions returned
8/8
Retained issues
2

Scope: verification evidence only—not KYC, underwriting, a lending decision, compliance, or production accuracy.

Open reviewed proof Browse GitHub sample
Reviewed insurance extraction proof showing an authentic public GSA crash report beside eleven exact fields
Current reviewed proof · GSA Standard Form 91 · privacy-safe demo values
02 / Insurance11/11 exact → Continue

Eleven exact fields on one government crash form.

The live response matched every predeclared value and returned a valid, page-bounded primary source region for each field on this selected form.

Exact values
11/11
Primary regions returned
11/11
Retained issues
0

Scope: document intake only—not coverage, liability, fraud, eligibility, payment, adjudication, compliance, or production accuracy.

Open reviewed proof Browse GitHub sample
Reviewed prior-authorization proof showing an authentic public CMS form beside eleven compared fields and two retained issues
Current reviewed proof · CMS GLP-1 Bridge form · privacy-safe demo values
03 / HealthcareReview blocked

A checked box—and the wrong extracted answer.

Nine of eleven values matched exactly. The ledger retains a wrong acknowledgement answer and a duplicated requested-drug value beside their returned primary regions.

Exact values
9/11
Primary regions returned
11/11
Retained issues
2

Scope: intake and source review only—not clinical correctness, medical necessity, coverage, eligibility, approval, compliance, or production accuracy.

Open reviewed proof Browse GitHub sample

Claim boundary

Evidence this narrow stays useful.

These pages demonstrate one extraction chain across three selected one-page authentic public forms populated with privacy-safe demo values. One live response was recorded per form. This is a demonstration, not an accuracy, production, or performance benchmark.

Three of the four mismatches are exposed by returned metadata. The fourth stays visible in the expected-versus-actual ledger and beside the returned region on the full proof page. For mortgage servicer name, the primary and nested regions clip the decisive final period. Grounding makes mistakes reviewable. It does not make them correct.

WORKFLOW SCHEMATIC · CONCEPT: Extract → Source evidence → Decision gate. This is a discussion model for future review, approval, exception, audit, or access controls—not shipped product UI or proof of a complete governed-review workflow.

  • No accuracy, production, or performance benchmark claim
  • No compliance, security, or regulatory certification claim
  • No KYC, underwriting, lending, coverage, liability, fraud, or adjudication decision
  • No clinical correctness, medical necessity, eligibility, or healthcare approval
  • No account-balance reconciliation, personal user approval, or publication approval

Run your own document

Move from proof to first request.

Start in the official guide or Studio, review current pricing, and get the separate Data Extraction API key before making an authenticated request.

Read the guide Open Nutrient Studio Review current pricing Get an API key

Pricing and credit usage can change. Check the current official pricing page before making a live request.