Title: Repair Before Veto: Repair-Augmented Constraint Learning for Contextual Decisions

Original: arXiv:2606.02326v1 Announce Type: new Abstract: Hard constraints are usually treated as terminal vetoes: once a candidate violates a requirement, the learned rule rejects it and any repair is handled outside the decision semantics. This misses a common deployed regime in which the system already knows a finite menu of modifications, such as adding a ticket option, changing a configuration, or requesting an available service upgrade. Existing constraint-learning, soft-relaxation, and recourse methods address nearby problems, but they do not learn whether an option should be repaired before being vetoed. We introduce Repair-Augmented Constraint Learning (RACL), a contextual decision framework that lifts known repair operators into the classifier semantics. A candidate is accepted when an affordable repair makes it feasible and preferred enough; otherwise the system returns a structured rejection credit and, when applicable, a repair plan. This repair-before-veto view strictly generalizes no-repair HASSLE-style semantics, reveals an irreducible false-veto gap for terminal-veto rules, separates binary-label non-identifiability from decision-rule learnability, and gives capacity and calibration bounds for the observed-feasibility shared-weight setting. Across controlled and DB1B-derived benchmarks, RACL recovers the intended credit and repair structure. On the hardest raw-data-derived tier, validation-selected RACL reduces false vetoes to 10/4039 (FVR 0.0025), versus about 1064/4039 for the strongest repair-search black-box baseline, while making the FVR/EDR trade-off explicit.

Rewritten: Title: Repair Before Veto: Repair-Augmented Constraint Learning for Contextual Decisions

Original: arXiv:2606.02326v1 Announce Type: new Abstract: Hard constraints are usually treated as terminal vetoes: once a candidate violates a requirement, the learned rule rejects it and any repair is handled outside the decision semantics. This misses a common deployed regime in which the system already knows a finite menu of modifications, such as adding a ticket option, changing a configuration, or requesting an available service upgrade. Existing constraint-learning, soft-relaxation, and recourse methods address nearby problems, but they do not learn whether an option should be repaired before being vetoed. We introduce Repair-Augmented Constraint Learning (RACL), a contextual decision framework that lifts known repair operators into the classifier semantics. A candidate is accepted when an affordable repair makes it feasible and preferred enough; otherwise the system returns a structured rejection credit and, when applicable, a repair plan. This repair-before-veto view strictly generalizes no-repair HASSLE-style semantics, reveals an irreducible false-veto gap for terminal-veto rules, separates binary-label non-identifiability from decision-rule learnability, and gives capacity and calibration bounds for the observed-feasibility shared-weight setting. Across controlled and DB1B-derived benchmarks, RACL recovers the intended credit and repair structure. On the hardest raw-data-derived tier, validation-selected RACL reduces false vetoes to 10/4039 (FVR 0.0025), versus about 1064/4039 for the strongest repair-search black-box baseline, while making the FVR/EDR trade-off explicit.

Rewritten:

Traditionally, hard constraints function as absolute vetoes: if a proposal fails to meet a specific requirement, the system immediately rejects it, leaving any potential fixes to be managed separately from the core decision logic. This approach overlooks a prevalent real-world scenario where systems are equipped with a predefined set of remedial actions—such as offering a ticket alternative, adjusting settings, or upgrading a service. While current methods for constraint learning, soft relaxation, and recourse tackle related issues, they fail to determine whether a proposal should be corrected prior to rejection. To address this, we present Repair-Augmented Constraint Learning (RACL), a decision-making framework that integrates known repair actions directly into the classification process. Under this model, a proposal is approved if a cost-effective modification renders it viable and desirable; if not, the system issues a detailed rejection explanation and, where possible, a specific remediation strategy. This "repair-first" approach extends beyond the no-repair HASSLE framework, highlighting an unavoidable error margin in traditional veto systems, distinguishing the non-identifiability of binary labels from the learnability of decision rules, and establishing performance limits for shared-weight settings with observed feasibility. Tests on both controlled datasets and those derived from DB1B confirm that RACL successfully reconstructs the desired credit and repair mechanisms. In the most challenging raw-data category, the validated RACL model significantly lowered false rejections to 10 out of 4,039 cases (a False Veto Rate of 0.0025), compared to approximately 1,064 out of 4,039 for the leading black-box repair-search baseline, thereby clarifying the balance between false vetoes and expected decision rates.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC