Backend
GraphScent: Calibrating Directional Relevance Signals for Bounded-Context Code Localization
arXiv cs.SE0 views
arXiv:2611.17527v1 Announce Type: new Abstract: Running open-weight language models locally on a single consumer GPU requires replacing expanding multi-turn exploration histories with short, stateless code inspections. Stateless repository localization, however, hinges on a foundational question: when a frozen language model evaluates an unedited code entity from a compact view, which structural relationships make its relevance judgment informative about a nearby repair target? Studying Gemma 4 models across 211 real-world Python issues in 86 repositories, we uncover a systematic directional relevance asymmetry: zero-temperature relevance scores on enclosing scopes and upstream callers consistently separate true structural neighbors from hard lexical distractors, whereas scores on downstream callees, peer methods, historical co-change partners, and dense node embeddings remain near chance. Zero-leakage masked-scope probes and regressions controlling for lexical similarity and snippet length confirm that this asymmetry reflects functional role (callers expose symptom entrypoints delegating to defective routines, while downstream helpers appear benign alone) rather than target leakage or surface overlap. Guided by this finding, we introduce GraphScent, a training-free search controller that propagates out-of-repository, relation-conditioned log-likelihood ratios over runtime Typed-AST multigraphs before resolving ties via a single listwise comparison, alongside AST Ripple, a lightweight post-edit check for missed co-edits. Likelihood calibration prevents overconfident smaller models from chasing false-positive callers while lifting exact localization on lexically hard targets from 12.8 to 23.4 percent at 31B within a single-GPU budget.
Source:https://arxiv.org/abs/2611.17527PreviousInterpreting Coding Agents Through Their Behavioral TrajectoriesNextBeyond Source Code: Evaluating Natural Language Specifications for LLM-Based C-to-Rust Translation
Related posts

Your LLM Wiki Compiles Away Your Permissions. Design for It Before the Team Uses It.
0dev.to OpenSource
Three more tools, three more countries: Brunei's new pension fund, Mongolia's salary tax and a Punjabi calendar that starts in March
0dev.to OpenSource
Why Chrome's 'Sanitized HAR' Export Still Leaks API Keys, Query Tokens & GraphQL Variables
0dev.to OpenSource
The runtime was as big as my 2.6 MB PII model, so I wrote a 13 KB inference engine
0dev.to OpenSource
I built Trongate CLI, a command-line toolkit for the Trongate PHP Framework
0dev.to OpenSource
sheet_route: Flutter bottom sheets as real routes, with snap points and go_router pages
0dev.to OpenSource