# Witness Evidence Portfolios: Single-Prefill Risk Detection for Closed Multimodal Answers

arXiv: https://arxiv.org/abs/2607.27667 · HTML: https://mlsanity.com/paper/2607.27667/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in VLM benchmarks and visual reasoning, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 27 of its 40 references that are in the map.
- First posted on arXiv 30 Jul 2026.

## Placed in

- Multimodal LLMs / vision-language models / [VLM benchmarks and visual reasoning](https://mlsanity.com/mllm/vlm-benchmarks-and-visual-reasoning/)

## Builds on (most cited first, 27 of 27)

- [Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization](https://mlsanity.com/paper/1610.02391/) (2016)
- [On Calibration of Modern Neural Networks](https://mlsanity.com/paper/1706.04599/) (2017)
- [Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning](https://mlsanity.com/paper/1506.02142/) (2015)
- [Improved Baselines with Visual Instruction Tuning](https://mlsanity.com/paper/2310.03744/) (2023)
- [Axiomatic Attribution for Deep Networks](https://mlsanity.com/paper/1703.01365/) (2017)
- [A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks](https://mlsanity.com/paper/1610.02136/) (2016)
- [MMBench: Is Your Multi-modal Model an All-around Player?](https://mlsanity.com/paper/2307.06281/) (2023)
- [InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency](https://mlsanity.com/paper/2508.18265/) (2025)
- [Attention is not Explanation](https://mlsanity.com/paper/1902.10186/) (2019)
- [Selective Classification for Deep Neural Networks](https://mlsanity.com/paper/1705.08500/) (2017)
- [Object Hallucination in Image Captioning](https://mlsanity.com/paper/1809.02156/) (2018)
- [Attention is not not Explanation](https://mlsanity.com/paper/1908.04626/) (2019)
- [Transformer Interpretability Beyond Attention Visualization](https://mlsanity.com/paper/2012.09838/) (2020)
- [OPERA: Alleviating Hallucination in Multi-Modal Large Language Models via Over-Trust Penalty and Retrospection-Allocation](https://mlsanity.com/paper/2311.17911/) (2023)
- [INSIDE: LLMs' Internal States Retain the Power of Hallucination Detection](https://mlsanity.com/paper/2402.03744/) (2024)
- [AMBER: An LLM-free Multi-dimensional Benchmark for MLLMs Hallucination Evaluation](https://mlsanity.com/paper/2311.07397/) (2023)
- [Mitigating Hallucinations in Large Vision-Language Models with Instruction Contrastive Decoding](https://mlsanity.com/paper/2403.18715/) (2024)
- [Reliable Visual Question Answering: Abstain Rather Than Answer Incorrectly](https://mlsanity.com/paper/2204.13631/) (2022)
- [A Survey of Multimodal Hallucination Evaluation and Detection](https://mlsanity.com/paper/2507.19024/) (2025)
- [Improving Selective Visual Question Answering by Learning from Your Peers](https://mlsanity.com/paper/2306.08751/) (2023)
- [HALP: Detecting Hallucinations in Vision-Language Models without Generating a Single Token](https://mlsanity.com/paper/2603.05465/) (2026)
- [VIB-Probe: Detecting and Mitigating Hallucinations in Vision-Language Models via Variational Information Bottleneck](https://mlsanity.com/paper/2601.05547/) (2026)
- [LEC: Linear Expectation Constraints for Selection-Conditioned Risk Control in Selective Prediction and Routing Systems](https://mlsanity.com/paper/2512.01556/) (2025)
- [Variational Visual Question Answering for Uncertainty-Aware Selective Prediction](https://mlsanity.com/paper/2505.09591/) (2025)
- [HEDGE: Hallucination Estimation via Dense Geometric Entropy for VQA with Vision-Language Models](https://mlsanity.com/paper/2511.12693/) (2025)
- [Reallocating Attention Across Layers to Reduce Multimodal Hallucination](https://mlsanity.com/paper/2510.10285/) (2025)
- [Taming Object Hallucinations with Verified Atomic Confidence Estimation](https://mlsanity.com/paper/2511.09228/) (2025)

## Abstract (arXiv)

Reliable deployment of multimodal large language models (MLLMs) requires deciding whether a confident visual answer should be trusted, reviewed, or routed to a stronger system. Confidence scores capture candidate margins, but not where the estimated signed visual readouts associated with those margins come from or how they are distributed. We study inference-time risk detection for closed visual answers using the same white-box prefill path that produces the answer. Witness Evidence Portfolios (WEP) first estimates, layer by layer, which visual contributions support or contradict the predicted candidate. It summarizes these contributions through two interpretable route families: question-related evidence provenance and signed evidence concentration. Nested grouped validation chooses the more reliable family and a sparse top-k route portfolio, which is fused with candidate confidence. WEP needs no image perturbation, decoding change, backward pass, or external verifier. Across three MLLMs and four binary-answer benchmarks, WEP improves mean error AP by 0.134. All 12 model--dataset gains are positive, and image-cluster bootstrap intervals are strictly positive on 10 pairs. WEP targets white-box closed-answer systems and uses a labeled calibration slice.