Perception, Layout, and Validation: Calibrated Confidence for Reliable Straight-Through Processing of Financial Documents
A decomposed perception-layout-validation confidence score with conformal risk control lets VLMs auto-approve 49-72% of financial document fields at target error.
The paper proposes a decomposed confidence layer over VLM-extracted financial document fields using perception, layout, and validation channels plus conformal risk control for reliable straight-through processing. Tested on invoices and ad-buy forms with Qwen3.6-27B and Gemini-3.1-Flash-Lite, AUROC improves from 0.54-0.74 to 0.90-0.99. The method auto-approves 49-72% of fields at sub-10% empirical error versus only 0.1-7.0% for native VLM verbalized confidence.
- Decomposed score spans perception, layout, and validation channels with final conformal risk control
- Validated on three public datasets using Qwen3.6-27B and Gemini-3.1-Flash-Lite VLM families
- AUROC improves from 0.54-0.74 (verbalized confidence) to 0.90-0.99
- Auto-approves 49-72% of fields at target error under 10%, versus 0.1-7.0% natively
Full article198 words · extracted from arxiv.org · click to collapse
Straight-through processing (STP) on extracted key-value fields from financial documents without human review requires a calibrated probability together with a bounded guarantee on the residual error of the auto-approved tier. The emergence of modern Vision Language Models (VLMs) provides an out-of-the-box capability for extracting the key-values, but their verbalized confidence signals are unreliable and weakly track field correctness. This paper introduces a decomposed confidence layer along three interpretable channels, including perception, layout, and validation. Together with a final conformal risk control, the score can be used for reliable STP of financial documents. The method is validated on three public datasets covering real invoices, synthetic invoices, and ad-buy forms, using two different VLM families (Qwen3.6-27B and Gemini-3.1-Flash-Lite). Our decomposed score consistently improves the separation of correct from incorrect extractions, substantially raising the AUROC from 0.54-0.74 for VLM verbalized signals to 0.90-0.99 with contributions from all three designed channels. Crucially for industrial deployment, this enables usable STP. The native VLM confidence signals could clear only 0.1%-7.0% of fields under risk control at a target error of <10%. In contrast, the proposed method auto-approves 49-72% of fields while holding the empirical error of the accepted tier at or below the target.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.20110