Multimodal agents for visual question answering are increasingly built as multi-step trajectories that combine perception, retrieval, and reasoning. However, evaluation typically focuses on final-answer accuracy, which cannot reveal whether a correct answer was derived from grounded evidence, language priors, or accidental error cancellation.
To address this, researchers propose treating a multimodal agent trajectory as a provenance-constrained state machine. Tool outputs are normalized into a Structured Evidence Ledger that serves as the trajectory state. Downstream reasoning and decision claims may only cite active ledger entries, and grounding is checked at the entity and numeric level. Repair is realized as typed state transitions that cannot introduce content without tool-produced provenance.
This design is instantiated as LedgerMind, augmented by a Three-Layer Grounding Protocol, an Adaptive Dual-Path Dispatcher that matches reasoning depth to question complexity, and an Event-Triggered Verification-and-Repair engine with a formal provenance non-amplification guarantee.
LedgerMind targets four recurring failure patterns that final-answer accuracy tends to obscure: unsupported intermediate reasoning, citation-backed entity hallucination (Phantom Grounding), over-reasoning on simple queries, and repair-time amplification. Experiments across multiple multimodal reasoning benchmarks and backbone MLLMs show that LedgerMind improves both answer accuracy and trajectory-level faithfulness.