For accounts payable & procurement
Question the price.
Keep the proof.
An invoice says one thing. The contract says another. Follow the difference, test what matters, and leave a review someone else can reproduce.
No account needed. Start with a fictional example.
Invoice INV501
Atlas
November 1, 2026
FILTER · 40 units$85.00 each
Billed$3,400.00
Invoice source · row 2Contract C1
The agreed price.
Nov 1–30, 2026 · FILTER
Difference to review
$520.0040 × ($85.00 − $72.00)A difference you can trace.
From two files to a reviewable case
A clear thread.
From question to handover.
Keep the original sources, the choices you made, and the calculation that followed.
- 01Sources & decisions
Bring the sources together.
Import invoice and contract CSVs. Select the right dated contract when versions overlap; record why an invoice is excluded.
- 02Original & alternatives
Ask what could change.
Change a shared price once and follow it across every linked invoice. Find the exact price that changes your review priority.
- 03Evidence & handover
Pass on the whole story.
Download a case packet with the source choices, code and results. A separate verifier reconstructs the arithmetic without running its Python.
A small change. A different question.
One contract price.
Three invoices move.
Try changing Atlas’s shared contract price. Watch the differences update and see which invoice comes first.
Enter $65.00–$85.00, or use the slider.
Use a price from $65.00 to $85.00, with up to two decimals.
An exact, local calculation using fictional data.
No AI call. No source files changed.
Positive variance · 7 included invoices
| Invoice | Units | Difference |
|---|---|---|
| INV501 Atlas | 40 | $520.00 |
| INV502 Atlas | 10 | $100.00 |
| INV508 Atlas | 15 | $0.00 |
INV501 is first to review.
The actual workspace
Less guessing.
More context.
Keep the ledger and its source evidence together. Go deeper when you need to; keep the original calculation in view.
Watch the complete site demo3 min 30 sec · English · narration by elevenlabs.io
Useful AI. Visible boundaries.
An answer is a start.
Evidence is the handover.
Vultr generates the Python. A disposable, offline Docker container executes it. Independent Decimal arithmetic checks every reported row.
Read the architecture- 01 Sources stay attached.
- Original rows, contract selections and exclusions stay in the exported case.
- 02 Execution stays visible.
- Inspect the actual mode, generated code, model request IDs, runtime limits and container cleanup.
- 03 Results can be checked.
- The packet verifier reconstructs the arithmetic and checks content integrity. Hashes do not establish contract truth or issuer identity.
Before you open a review
A few useful answers.
What can I try without an account?
Open the workspace and choose “Open a reviewed example.” It uses fictional invoice and contract files and trusted Reference rules. There is no model call, sign-up or payment step. You can inspect sources, explore assumptions and export the case.
Does every review use AI?
No. Reference mode uses fixed arithmetic. Sandbox mode runs a fixed Python recipe in Docker. Vultr agent mode requests generated Python from Vultr Serverless Inference and requires an operator token. The workspace labels the actual mode.
What files does Prooflane accept?
Invoice and contract CSVs with the columns shown in the workspace’s field guide. Matching uses exact vendor and item keys plus inclusive contract dates. This prototype does not scan PDFs, connect to an ERP or make payments.
What does a positive difference mean?
It means a billed unit price exceeds the supplied contract price for an included row. It is a signal for review, not proof of fraud, contractual liability or recovered savings.
Start with one clear question
Which price supports
this invoice?
Bring your CSVs, or use ours.