Retrieval-augmented language models are expected to answer from the retrieved evidence, but in practice they often keep answering when that evidence is missing. We trace this behavior to the training signal: answer-focused fine-tuning assigns no target to unsupported contexts, so it cannot distinguish a reader that abstains from one that guesses, and unsupported answering stays near 100% even as supported accuracy improves. We introduce Regime Boundary Alignment (RBA), which trains a single reader on matched variants of the same question and gold answer. The reader is trained to produce the gold answer when the context supports it, including when conflicting evidence is also present, and to abstain when the correct support is removed; inference is ordinary decoding, with no verifier, threshold, or regime label. On three multi-hop QA datasets across three seeds, RBA reduces the unsupported-answer rate by more than sixty percentage points relative to conflict-focused training while matching its supported accuracy. On a held-out TriviaQA retrieval-miss slice, the same reader reduces unsupported answering from 100% to below 1% while also improving supported accuracy. These results indicate that evidence-gated answering must be learned on both sides of the support boundary.
Figures & tables
Figure 1: The answerability boundary on one example. A notice that names the new room supports answering with it; a notice that confirms only the move does not, and repeating the old room is then unsupported. A reader trained only to answer (left) answers in both cases, while the RBA reader (right) answers with support and abstains without it.
Training condition
2Wiki G/U
HotpotQA G/U
MuSiQue G/U
TriviaQA G/U
Macro G/U
Abst.F1 ↑
AURC ↓
C@10 ↑
Reference and answer-focused baselines
Base
0.78/100.00
2.03/100.00
0.00/100.00
8.59/100.00
2.85/100.00
0.00
.980
0.00
Prompt
27.34/8.59
46.70/17.26
9.09/7.14
37.11/99.61
30.06/33.15
63.24
.304
11.36
DTA
27.73/100.00
45.69/100.00
12.34/100.00
49.61/100.00
33.74/100.00
0.00
.707
0.00
Conflict-SFT (no C0 )
72.27 /100.00
64.97 /100.00
70.78/100.00
55.47/100.00
66.13 /100.00
0.00
.486
1.33
Boundary-supervision controls
Table 1: Main result and mechanism controls. Each dataset cell and Macro report Supported Gold/UAR (G/U); abstention F1, AURC, and C@10 pool paired supported/no-support rows. Values are seed medians and percentages except AURC. Base and Prompt are deterministic single-checkpoint evaluations; all trained rows use three seeds. The best displayed value within each individual metric is bold (lower is better for UAR and AURC); displayed ties are included.
Figure 2: Evidence-boundary margin distributions. Each contour is a Gaussian KDE over held-out (q,g) pairs after averaging exact teacher-forced sequence margins within question across the three frozen seeds. The horizontal axis is S(g∣q,C+)−S(IDK∣q,C+) ; the vertical axis is S(IDK∣q,C0)−S(g∣q,C0) . The shaded upper-right quadrant therefore represents the desired joint preference.
Figure 3: Boundary-state geometry and seed stability. Panels (a–b) show seed-averaged paired margin states for Conflict-SFT and RBA . Panels (c–d) compute adjusted Rand index (ARI) between boundary-quadrant labelings of the same held-out questions across seed pairs, with bootstrap resampling for the histogram and interquartile bars.
Evidence shift
Metric
Base
Conflict-SFT
RBA
BM25 miss
G / U
2.85 / 100.00
66.13 / 100.00
66.13 / 0.00
Dense miss
G / U
3.12 / 100.00
61.70 / 100.00
62.10 / 0.10
Partial evidence
U
100.00
100.00
83.07
Controlled seen
U
100.00
100.00
97.62
Unseen template
U
100.00
100.00
76.50
Semantic negation
U
100.00
100.00
99.30
Table 2: RQ3 frozen-checkpoint transfer. Entries are dataset-macro percentages and seed medians. Paired rows report G/U; no-support-only rows report U. Controlled conditions cover three multi-hop datasets.
Conflict-SFT
RBA
Change
Reader
G
U
G
U
Δ G
Δ U
Qwen3-8B
66.13
100.00
66.13
0.00
+0.01
-100.00
Llama-3.1-8B
67.89
100.00
67.69
0.00
-0.19
-100.00
Mistral-7B
66.76
100.00
66.71
0.00
-0.05
-100.00
Table 3: RQ4 cross-family comparison. Values are dataset-macro percentages and medians over three seeds. Deltas are RBA minus Conflict-SFT.
Grounded question answering systems should answer only when the supplied evidence supports the answer. In multi-hop QA, this requirement is difficult because partial evidence can make an unsupported answer appear plausible. We study selective answering through evidence sufficiency boundaries: for the same question, a model should abstain under unsupported or partially supported context, answer when the context first becomes sufficient, and keep the answer stable when redundant evidence is added. We introduce Evidence Sufficiency Boundary Training, a generation-native training framework that constructs ordered evidence chains and supervises the abstain-to-answer transition directly. The method combines level supervision, a boundary flip margin, post-boundary stability, and answer recall protection. We build evidence chains from HotpotQA, 2WikiMultiHopQA, and MuSiQue, then evaluate models with chain metrics, raw QA utility, and unsupported-answer rates on external non-answerable sets. With Qwen2.5-3B-Instruct and LoRA adaptation, Evidence Sufficiency Boundary Training gives the strongest boundary localization among the tested systems, with flip accuracy of 0.807 compared with 0.781 for a token-level abstention baseline. It also achieves the lowest overall unsupported-answer rate on external non-answerable evaluation, 0.095 compared with 0.101 for the same baseline, while retaining competitive raw QA F1. The results show that grounded selective answering improves when training marks the evidence level where refusal should give way to answering.
Many modern AI question-answering systems convert text into vectors and retrieve the closest matches to a user question. While effective for topical similarity, similarity scores alone do not explain why some retrieved text can serve as evidence while other equally similar text cannot. When many candidates receive similar scores, systems may select sentences that are redundant, incomplete, or address different conditions than the question requires. This paper presents a deterministic evidence selection framework for retrieval-augmented question answering. The approach introduces Meaning-Utility Estimation (MUE) and Diversity-Utility Estimation (DUE), fixed scoring and redundancy-control procedures that determine evidence admissibility prior to answer generation. Each sentence or record is evaluated independently using explicit signals for semantic relatedness, term coverage, conceptual distinctiveness, and redundancy. No training or fine-tuning is required. In the prototype, a unit is accepted only if it explicitly states the fact, rule, or condition required by the task. Units are not merged or expanded. If no unit independently satisfies the requirement, the system returns no answer. This deterministic gating produces compact, auditable evidence sets and establishes a clear boundary between relevant text and usable evidence.
In multi-hop RAG evaluation, a top-k answer score can hide two different failures: the retrieval window may drop part of the support chain, or it may contain support in a form the adapted reader does not use well. We call this reader-facing form of retrieved evidence an evidence interface. Using three support-annotated multi-hop QA benchmarks, we compare matched adapted readers trained with raw context, retrieval windows, and gold-support diagnostic renderings. These comparisons distinguish support-availability failures from remaining reader-interface effects. Top-k windows become interpretable only after checking whether the complete annotated support chain survives: when it does, short ranked windows can match or improve over raw context; when it does not, missing support explains much of the loss. Gold support-first improves matched readers; on 2Wiki and MuSiQue, a support-supervised ranker raises coverage and recovers raw-context quality at lower prompt cost, while retaining gold headroom. Support-removal checks further show that the gains rely on exposed evidence, not only answer priors. On support-annotated evaluations, top-k answer scores should therefore be reported together with complete-support coverage.