Source

Causal-aware training with counterfactual supervision and structural causal models improves the robustness of LLM-based vulnerability detection PRIMARY_TEXT

Scientific Reports · 2026-09-01 · no page hash: the reader saw a rendering, not the bytes · created seq 14745

Link: https://www.nature.com/articles/s41598-026-69332-6 · read at 2026-09-28T23:55:00Z · held by reference: only the link, the quoted excerpts with their hashes, and the page hash when the reader had the bytes are stored

FETCHED at 2026-09-28T23:58:04Z · https://www.nature.com/articles/s41598-026-69332-6?error=cookies_not_supported&code=582b9652-9783-4e12-a09e-d902f5fa8592 · seq 14801

Server hash sha256:4d35efabf036787358492f6b4909213241316d791c894c94f8696171e45085e3 · 224550 bytes · text/html. The reader's record is unchanged; this is the server's finding beside it.

Excerpt	Finding
Under a rigorous project-level group-aware evaluation protocol, CausalVul achieves an F1-score of 76.4% on Devign—an improvement of +18.0 points over its own...	verbatim — the passage is on the page, character for character
Under combined semantics-preserving lexical perturbations, CausalVul exhibits only a 5.1% F1 drop compared to 12.1–22.3% for baselines, a 2.4× improvement in...	not found — this server looked at the page and did not find the passage