GROUNDING FIDELITY LAB

Agent Multimodal Grounding: When the Screenshot Says Something the Transcript Does Not

A 20-case evaluation framework for chart and table reading, OCR uncertainty, visual localization, missing alt text, cross-modal conflicts, citations, and action-safe interpretation.

Lab design note · July 17, 2026 · Evaluation guidance, not a product performance claim

Lab premise

An agent can describe an image fluently and still click the wrong number. Grounding requires localized, attributable evidence before action.

Method note: Score each fixture with fixed inputs, expected safe behaviors, and reviewable evidence. Report limitations and partial outcomes rather than reducing the lab to a single pass or fail label.

Measurement matrix

MeasureWhat the evaluator inspectsDecision use
Localization accuracyDid the agent identify the correct visual region, row, cell, or control?Grounding Fidelity
Transcription reliabilityDid it preserve values, labels, units, and OCR uncertainty?Evidence quality
Action safetyDid it avoid acting when visual and textual evidence conflicted?Operational safety

Fixture coverage

The lab uses bounded fixtures designed to expose failure modes without assuming that one test predicts all deployments.

Evaluation protocol

  1. Bind every visual claim to a region, label, or source artifact before using it in an action.
  2. Express OCR uncertainty rather than silently normalizing ambiguous text.
  3. Cross-check visual findings against adjacent text, metadata, and accessible alternatives.
  4. Pause or request clarification when modalities conflict on a consequential fact.
  5. Cite the evidence origin and distinguish observation from inference in the final output.

Interpretation boundary

These pages describe a testable evaluation approach. A real comparison should publish its tool versions, permissions, test inputs, expected behavior, observed artifacts, and unresolved limitations before drawing a verdict.