Answers are grounded in this paper and cite their sources.
Demonstrates high accuracy in document classification, suggesting improvements over traditional methods.
Document classification in enterprise settings faces a fundamental challenge: documents where the discriminative signal is not concentrated in any fixed-size context window but distributed across the full document content. We call this class Documents with Absent Implicit Classification (DAIC). Existing approaches — fine-tuned BERT variants, Longformer, and direct LLM classifiers — achieve only 40--73% accuracy on DAIC documents. We propose MOSAIC (Multi-Observer Structured Aggregation for Implicit Classification), a multi-observer architecture that reframes document classification as progressive evidence accumulation and hypothesis space reduction. Five specialised filters observe the same document from complementary perspectives, reducing the candidate type set while satisfying four formal properties: monotonicity, correctness, convergence, and epistemic honesty. The final LLM Arbiter reasons over ~1.8 candidates with accumulated evidence — not over the full type space. Evaluated on a real-world corpus of 2,037 administrative inspection documents across 14 document types, MOSAIC achieves 91.25% overall accuracy and 90% accuracy on the DAIC-genuine subset as validated by two independent domain experts. Median latency is 1.0ms with only 2.8% of documents reaching the LLM Arbiter. The contribution is not higher raw accuracy but a system that simultaneously provides competitive accuracy, human-validated performance on the hardest subset, auditable decisions with evidence citation, and the ability to admit uncertainty when evidence is insufficient.
No takes yet. Share an insight, caveat, or question.
Alejandro Jaime (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: