"GraphLedger: Repository-Level Vulnerability Detection via Graph-Based Compression and Assumption Validation," by Thi-Hong-Cuc Le, Hoang-Quoc-Bao Hua and Dr. Xuan-Bach Le, has been accepted into the Research Track of ASE 2026, CORE Rank A*, at 115/1236. The conference is held in Munich, Germany, in November 2026. Finding vulnerabilities at repository scale means reasoning across files, across procedures, and along long dependency chains. Both dominant approaches struggle here: static analysis over-approximates and buries the user in false positives, while LLM-based methods choke on large repository context and can produce claims that sound plausible but rest on nothing. GraphLedger combines Adaptive Assumption-Corroborating Compression with an Assumption Ledger for structured hypothesis verification. The compression step shrinks repository context while preserving the dependency structure that matters for vulnerabilities. The LLM then proposes explicit vulnerability hypotheses over that compressed context, and the ledger checks each assumption against evidence in the repository, returning Verified, Rejected or Inconclusive. On CWE-Bench-Java, the compression achieves 0.71 token reduction while retaining 0.90 of dependency edges and 0.88 of taint paths. On a documented, repository-disjoint 100-repository evaluation partition, GraphLedger reaches 0.77 precision, 0.76 recall and 0.76 F1, cutting the false discovery rate from 0.41 to 0.23 relative to VulAgent under a controlled shared-backend comparison. Paper: https://lexuanbach.github.io/publication/ASE2026.pdf Supplement: https://lexuanbach.github.io/publication/ASE2026-supplement.pdf Artifact: https://doi.org/10.5281/zenodo.21356624 DOI: https://doi.org/10.1145/3832783.3834420