Sample Freight Workflow Audit

This redacted example shows the shape and evidence discipline of the $199 audit. It uses only cited public documents and contains no paid-customer material, private demand signal, or production result.

Effective September 21, 2026

Executive disposition

Conditional automation candidate. Exact extraction was observed for two inspected public rate confirmations. Mixed packets and documents without a reliable load identifier stopped for review. A buyer-specific matched invoice, rate confirmation, BOL, and POD packet has not been validated.

Bounded workflow

  1. Receive one text-readable freight document or an already separated page.
  2. Classify the document as rate confirmation, bill of lading, proof of delivery, or unknown.
  3. Extract only fields supported by visible source evidence.
  4. Stop mixed packets, ambiguous document types, and missing join keys for human review.
  5. Write nothing to a TMS, accounting system, shared drive, or production workflow.

Observed evidence

Public inputObserved resultDisposition
RXO rate confirmation 14404811Rate confirmation; load 14404811; approved rate $1,300.00Matched inspected ground truth
RXO rate confirmation 14336973Rate confirmation; load 14336973; approved rate $600.00Matched inspected ground truth
NHTSA-hosted separated BOL pageBill of lading; no load ID inventedFail closed
Two separated POD pagesProof of delivery; no load ID inventedFail closed
Four mixed publications or packetsUnknown; mixed document types; review requiredRejected as unsplit packets

Exception and control register

TriggerRequired controlDecision owner
More than one document type in a PDFStop; split and re-submit pages before extractionBuyer-designated reviewer
Missing or ambiguous shipment join keyDo not infer a load; record the evidence gapBuyer-designated reviewer
Commercial-goods invoice resembles carrier invoiceRequire freight-specific evidence before classificationZinvyl audit logic, then buyer reviewer
Image-only or photographed pageOutside this fixed scope unless agreed before fundingBuyer and Zinvyl

Prioritized implementation plan

  1. Run a funded, held-out pilot on two or three buyer-approved de-identified text-readable PDFs.
  2. Define exact required fields and page-level evidence citations before measuring accuracy.
  3. Keep mixed-packet detection and missing-key stops as non-bypassable review gates.
  4. Only after field-level acceptance, design read-only integrations; production writes remain a separate contract.

Limits and provenance

The underlying corpus contains six cited public PDFs and seven separated real-world pages. The retained regression suite passed 23 of 23 checks after five defects were found and fixed. These results do not establish buyer-document accuracy, OCR performance, demand, savings, profitability, delivery acceptance, or production reliability.

Open the source-level validation record · Read the complete service contract · Request the bounded $199 audit