Cases
Transparency boundary
These pages show the source files, generated scans, task instructions, prompts, and known trap categories for the current paperwork suite. Oracle solution files are intentionally not shown.
Excluded on every case page: ground_truth.json, expected_artifacts.json, human_reading.md, calibration_results.md.
Paperwork Trial — Generated Image Cases
Five cases that hand the model a folder of generated invoice PNG scans plus CSV cross-checks, and require an exact audit_result.json.
P01 — Basic Invoice Folder
A small scanned-invoice folder with a quote distractor, partial payment, and an under-review stamp. 4 generated image files and 6 text/CSV/task files.
- customer vs vendor
- quote is not an invoice
- partial payment
- under-review stamp
P02 — Credit Note And Vendor Hold
A scanned folder with a credit note, partial payment, missing PO, and inactive-vendor warning. 4 generated image files and 5 text/CSV/task files.
- credit note is not payable
- partial payment
- missing PO
- inactive vendor
P03 — Duplicate Risk Mix
A larger mixed folder combining earlier-looking scans, a previous-invoices file, credit note, quote, and duplicate-risk lookup. 8 generated image files and 7 text/CSV/task files.
- duplicate-risk lookup
- mixed folders
- quote and credit note distractors
- partial payment
P04 — Tax ID Collision
A compact case around vendor identity and tax calculation conflicts. 1 generated image file and 5 text/CSV/task files.
- vendor tax ID conflict
- tax rounding mismatch
- statement distractor
P05 — PO Revision
A generated scan case where split payments and the latest purchase-order revision decide the outcome. 1 generated image file and 5 text/CSV/task files.
- split payment
- cancelled PO revision
- quote distractor
Paperwork Workflow — Agentic File Workflow Cases
Four cases that hand the model a messy intake folder and require it to inspect, identify active sources, generate manifests, and leave the source folder byte-identical.
W04 — Messy Intake Folder
A chaotic intake folder where the model must identify active sources, ignore stale files, write manifests, and preserve incoming sources. 4 generated image files and 9 text/CSV/task files.
- old bank export
- duplicate vendor file
- draft PO list
- non-invoice scan
- protected source folder
W05 — Email Attachment Intake
A versioning workflow with email context, revised invoice attachments, old references, and non-invoice screenshots. 4 generated image files and 6 text/CSV/task files.
- superseded invoice
- revised attachment
- old payment reference
- proforma distractor
- chat screenshot as evidence
W06 — Remittance Split
A remittance workflow where one payment has to be mapped across multiple final invoices while ignoring drafts and a proforma. 4 generated image files and 7 text/CSV/task files.
- single payment split
- draft bank export
- proforma distractor
- active remittance batch
W07 — Credit Offset Packet
A credit-offset packet with duplicate scans, a credit memo, statement distractor, inactive vendor, and cancelled PO. 5 generated image files and 6 text/CSV/task files.
- credit offset
- duplicate scan
- statement distractor
- inactive vendor
- cancelled PO