Gemini Drift: Automated Patch-Twin Discovery for NASA Flight Software via Semantic Graph Analysis
Year: 2025
Type: article
Abstract: Gemini Drift: Automated Patch-Twin Discovery for NASA Flight Software via Semantic Graph Analysis introduces a system for automatically detecting “drift variants”—unpatched sibling vulnerabilities that remain when fixes applied to one module in NASA’s Core Flight System (cFS) fail to propagate to semantically similar code elsewhere. Gemini Drift integrates: A revised patch similarity score that blends structural overlap, sink similarity, and semantic verification. Positive–Unlabeled sibling likelihood modeling to rank candidate variants. Neighborhood risk propagation over the call graph, boosted by heritage reuse across missions. SAVES, a semantic micro-slicing checker that verifies whether patch-intended invariants (bounds, MsgID/length checks, watchdog timing, table validation) are enforced. Mission-criticality gates adjust thresholds depending on whether software is human-rated, flagship, robotic, or CubeSat. A retrospective study on the transition from cFS 6.7→6.8 shows Gemini Drift achieves ~0.45 precision@10 on unseen patches (≈0.70 in pilots), with <6-hour end-to-end turnaround. Outputs are packaged as non-weaponized, review-ready evidence packs suitable for Flight Software Board triage. By systematizing patch-twin discovery, Gemini Drift reframes vulnerability management in mission software from ad-hoc expert diffing into a reproducible, safety-aware, and scalable process—tightening patch discipline while reducing review burden.
Author: Davis, Bee Rosa
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7083460576
FWCI:
Cites:
Cited by:
Related to: