Walk through the full reconciliation flow without uploading anything. Pre-extracted files below were processed in advance — click one to load its extracted data into the workbench. The original PDFs are linked next to each tile if you want to inspect what the extractor saw.
| Shipment | Connote | Carrier | Agent | Receiver | Status | Expected |
|---|
Try these in order to see the extractor handle every format:
Pre-extracted PDF events remain in the inbox when you paste text — only freshly parsed events are replaced.
| Connote | Event | Time | Recipient | Evidence | Source file | Failure reason | Conf. | Notes |
|---|
| Shipment | Connote | Carrier | Agent | Extracted | Source | Match | Conf. | Action | Audit |
|---|
| Shipment | Connote | Event | Time | Recipient | POD status | Source | Conf. | Review | Notes |
|---|
What's the same as production: the manifest schema, the matching algorithm (exact + OCR-aware fuzzy), the review queue contract, and the export shape. What's different in this demo: the rule-based extractor stands in for the production vision model, file uploads are limited to text, and there's no persistence — refresh wipes the workspace.