ConRAD: Conformal Risk-Aware Neural Databases

Sonia Horchidan , Fabian Zeiher , Xiangyu Shi , Vasiliki Kalavri , Henrik Bostr\"om , Ioannis Kontoyiannis , Paris Carbone

Authors on Pith no claims yet

Pith reviewed 2026-05-07 04:20 UTC · model grok-4.3

classification 💻 cs.DB

keywords neuralacrossconradgraphconformalqueryrecallrisk

0 comments

The pith

ConRAD uses conformal risk control and a new conformal gate operator to enforce declarative recall guarantees in neural graph queries while maximizing precision and skipping unnecessary neural inferences.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Knowledge graphs in databases are often incomplete, so neural models predict missing connections. When a query chains several predictions, small errors can compound and users have no formal way to know if most relevant results were found. ConRAD lets a user set a risk budget, which is the maximum fraction of true answers they are willing to miss. It then applies conformal risk control, a distribution-free statistical technique, to derive thresholds for each query step so the overall recall target is met with mathematical guarantees on finite data. To handle complex multi-operator queries efficiently, it reduces the high-dimensional threshold search to tuning a single scalar in quantile space. The system also adds a conformal gate, a physical operator that checks local graph evidence first and bypasses the neural model entirely when the data already suffices, avoiding wasted computation in dense regions. On three benchmarks and three query topologies, the approach met every risk budget with recall no more than 0.046 below target, eliminated neural calls in near-complete areas, and matched or beat the precision of static baselines that provide no guarantees.

Core claim

ConRAD strictly satisfies all risk budgets, with empirical recall falling below the target by at most 0.046 across all settings. It reduces neural invocations to zero in near-complete graph regions, and achieves precision that matches or exceeds best-case static baselines that offer no guarantees and require manual threshold search.

Load-bearing premise

The finite-sample, distribution-free validity of the derived thresholds rests on the exchangeability assumption of Conformal Risk Control holding for the calibration and test query instances, which may be strained by dependencies inherent in graph-structured data and multi-hop query topologies.

read the original abstract

Querying incomplete knowledge graphs with neural predictors is powerful but dangerous. Errors compound across multi-hop pipelines with no formal bound on the completeness of results. We introduce ConRAD, the first framework to enforce declarative recall guarantees natively within a neural graph database query engine. Given a user-specified risk budget, ConRAD automatically derives per-operator prediction thresholds that satisfy the recall target with finite-sample, distribution-free statistical validity via Conformal Risk Control, while maximizing end-to-end precision. To scale calibration across multi-operator query topologies, we introduce a quantile-space scalarization that reduces intractable high-dimensional threshold searches to a single parameter. We further design the conformal gate, a novel physical operator that dynamically bypasses neural inference when local graph evidence suffices, eliminating unnecessary model inferences in dense graph regions. Evaluated across three benchmarks and three query topologies, ConRAD strictly satisfies all risk budgets, with empirical recall falling below the target by at most 0.046 across all settings. It reduces neural invocations to zero in near-complete graph regions, and achieves precision that matches or exceeds best-case static baselines that offer no guarantees and require manual threshold search.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim depends on the standard assumptions of conformal risk control plus the effectiveness of the newly introduced gate and scalarization in graph query topologies. Free parameters are limited because thresholds are derived rather than freely fitted. The conformal gate is a new invented operator whose independent evidence is only the reported empirical behavior.

free parameters (1)

quantile-space scalarization parameter
Single tunable value that reduces the multi-operator threshold search; its selection is part of calibration but not a free parameter fitted to the final performance metric.

axioms (1)

domain assumption Calibration and test instances are exchangeable.
Required for the finite-sample, distribution-free guarantee of Conformal Risk Control to hold when applied to query operators.

invented entities (1)

conformal gate no independent evidence
purpose: Physical operator that bypasses neural inference when local graph evidence suffices.
New operator introduced to eliminate unnecessary model calls in dense regions; no external falsifiable prediction is given beyond the reported zero-invocation behavior.

pith-pipeline@v0.9.0 · 5517 in / 1676 out tokens · 76566 ms · 2026-05-07T04:20:28.869430+00:00 · methodology

ConRAD: Conformal Risk-Aware Neural Databases

Core claim

Load-bearing premise

discussion (0)