DockProof.Source & setup ↗
The working product · 2:44 · NVIDIA × Nebius

One damaged crate.
Every claim dollar, traced.

From a worksheet using the whole shipment's weight to a shipper-reviewed demand, with the source behind every selected fact and the original files in one packet.

English narration & captions · real browser and API recordingOriginal synthetic shipment and documents

Follow the evidence to $750.

The inherited worksheet uses 600 lb and asks for $3,000. B4's own weight is 150 lb. Its $2,000 invoice value, $100 discount and $175 retained salvage give $1,725 in evidenced loss. The selected class-70 terms give a $750 reference limit and reviewed demand.

The download includes the claim letter, valuation, rule references, and eight original files with SHA-256 fingerprints. The shipper submits through the carrier's filing process; carrier receipt has its own evidence.

A real model-powered document path.

  • Local English OCR reads the scanned delivery page and printed label.
  • NVIDIA Nemotron Super on Nebius Token Factory returns 32 source-matched candidate facts in 11.173 seconds.
  • The reviewer selects all 27 required fields and approves the evidence snapshot.
  • The final 22-file packet retains all eight original records.

The hosted scan case runs fresh browser OCR and loads 30 reviewed candidates from a separate, labeled recorded NVIDIA run. The film shows the live local-server request. Read the model integration notes →

English transcript

A damaged shipment leaves a small shipping team with scattered records and a costly question: how much should we claim? This worksheet uses all six hundred pounds and asks for three thousand dollars. DockProof traces the amount for one damaged crate, B4.

I bring eight synthetic evidence files to the desk: a mixed PDF, an invoice, warehouse records, a printed label, and a damage photo. Local English OCR reads the scanned delivery page and label. Original files stay in the browser, alongside their extracted text.

Now, Extract with Nemotron. This is a real NVIDIA Nemotron Super request on Nebius Token Factory. The model reads the document text and proposes typed values with exact source excerpts. DockProof locates those excerpts for review, including the scanned delivery date.

The reviewer checks the source before accepting each selected fact. The original worksheet carries six hundred pounds for the whole shipment. B4's own record supports one hundred fifty pounds. Both sources stay available. The affected piece, the original worksheet, and the chosen amount each keep their own evidence.

The calculation follows that evidence. Two thousand dollars in goods, less a hundred-dollar discount and one hundred seventy-five dollars in retained salvage, gives seventeen hundred twenty-five dollars in loss. The selected class-seventy terms give a seven-hundred-fifty-dollar reference limit and proposed demand.

After reviewing the sources and terms, I approve this evidence snapshot. DockProof prepares the claim letter, valuation, rule references, and all eight original files with their fingerprints. The shipper can download the packet and use the carrier's filing process. Carrier receipt is a separate recorded event.

Try the same scan case in your browser. It runs local OCR and loads clearly labeled results from a recorded NVIDIA request, ready for your own review. The application and source are public. DockProof: every claim dollar, traced.

Film credits and scope. Original application recording and synthetic records by Blucca. Browser reviewer actions are performed by the recording script. The sequence is edited for English narration; the model's measured response time appears on screen. The English voice is synthesized. Calculation scope: XPO US interstate LTL, new ordinary goods, one visibly damaged piece, direct shipper booking, and documented actual freight class and service terms.