This paper describes our submission to SubTask 2 of the FinEval 2026 Multilingual Financial Language Understanding Challenge, Japanese Financial Implicit Communication Recognition (JF-ICR), which places a company’s question-answer on a five-point commitment scale. Our submitted system, a three-stage pipeline, scores 15/15 public and 33/35 private, 4th of 10. A single open-weight model labels each of the 50 test items independently by verbaliser scoring; that labelling is audited against the class distribution the organisers publish in advance, which yields a label-free lower bound F on its error count and names the classes those errors sit in; each over-full class is then handed back to the model as a forced choice over its own members, with the exact number of mislabelled items asserted. The audit opens at F=3 and closes at F=0. Enforcing the same counts directly, by min-cost assignment, also reaches F=0 but recovers only the errors already nearest the decision boundary. The official score bounds what a floor of zero is worth: two of the 50 items are wrong beneath it, and, both the submitted labelling and the gold being balanced, those two errors can only form a compensating transposition.

We present a retrieval-augmented generation (RAG) system in production since April 2026 for credit policy question answering, a setting in which the corpus is revised several times per quarter: a single revision changes thresholds, narrows scopes, and restructures tables. Static test sets silently decay under this regime, and scores computed against successive corpus versions cannot separate system drift from corpus drift. We propose a longitudinal-aware evaluation framework that makes RAG quality over an evolving corpus both measurable and attributable. It combines (i) a change-type taxonomy — Stable, Evolved, Added, Removed — built on a deterministic, model-independent alignment of document subsections keyed to how each gold-supporting provision changed between versions, making explicit which scores remain comparable across releases; (ii) a versioned, class-stratified synthetic QA pipeline that keeps the test set current at revision cadence; and (iii) a statement-level LLM judge for compound policy answers; each component is calibrated against subject-matter experts. Across three corpus versions, a conventional longitudinal-unaware protocol would report a recall decline of up to 25.7 points. However, paired comparisons on answer-invariant items show no evidence of regression in the unchanged system.

Evaluating retrieval-augmented generation (RAG) on custom knowledge bases requires task-specific benchmarks of multi-chunk evidence synthesis, yet most existing generation pipelines apply quality control only after the full question set has been solved. We present VERGE, which moves quality control inside the generation loop: a verification agent targets, per item and without an external solver pass, an operational multi-chunk criterion and a QA-integrity criterion, refining items that fail and discarding those it cannot repair. From 11,052 candidates across six domains, VERGE yields 2,854 verified multiple-choice items targeting aggregation- and comparison-style synthesis over topically coherent evidence. We contribute a method-agnostic validation protocol — exhaustive strict-subset ablation, chunk-removal ablation, format-unconfounded open-ended re-evaluation, and inter-chunk-similarity stratification — and use it to localise where the multi-chunk dependence holds. On a stratified sample (n = 250), 69.8% of items not answerable closed-book cannot be answered from any strict subset of their supporting chunks (strict-necessity violation rate 15.2%). A human annotation study finds the verifier reliable (Cohen’s 𝜅 = 0.903) and VERGE items higher in QA integrity and distractor quality than the closest prior method (p < 0.001).