# Hallucination as output-boundary misclassification: a composite abstention architecture for language models

Source: [Hallucination as output-boundary misclassification: a composite abstention architecture for language models](<https://www.sota2.com/research/paper/69d732617d0172140ec76957>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-10T05:00:04.898Z

Authors: Angelina Hintsanen

Published: 2026-03-12T13:44:47.000Z

## Abstract

Large language models often produce unsupported claims. We frame this as a misclassification error at the output boundary, where internally generated completions are emitted as if they were grounded in evidence. This motivates a composite intervention that combines instruction-based refusal with a structural abstention gate. The gate computes a support deficit score, St, from three black-box signals: self-consistency (At), paraphrase stability (Pt), and citation coverage (Ct), and blocks output when St exceeds a threshold. In a controlled evaluation across 50 items, five epistemic regimes, and three models, neither mechanism alone was sufficient. Instruction-only prompting reduced hallucination sharply, but still showed over-cautious abstention on answerable items and residual hallucination for GPT-3.5-turbo. The structural gate preserved answerable accuracy across models but missed confident confabulation on conflicting-evidence items. The composite architecture achieved high overall accuracy with low hallucination, while also inheriting some over-abstention from the instruction component. A supplementary 100-item no-context stress test derived from TruthfulQA showed that structural gating provides a capability-independent abstention floor. Overall, instruction-based refusal and structural gating show complementary failure modes, which suggests that effective hallucination control benefits from combining both mechanisms.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Abstention Enforcement on TruthfulQA no-context abstention stress test 100-item](<https://www.sota2.com/research/sota/abstention-enforcement-on-truthfulqa-no-context-abstention-stress-test-100-item.md>): Abstention Enforcement; TruthfulQA no-context abstention stress test 100-item; Correctness: 100.

- [Abstention and Hallucination Evaluation on Controlled 50-item Evaluation Set (R1-R5)](<https://www.sota2.com/research/sota/abstention-and-hallucination-evaluation-on-controlled-50-item-evaluation-set-r1-r5.md>): Abstention and Hallucination Evaluation; Controlled 50-item Evaluation set epistemic regimes R1–R5; Accuracy: 98.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.06195>)

[PDF](<https://arxiv.org/pdf/2604.06195.pdf>)