Claim

Model ledger-default@0.1.0 ledger-strict@0.1.0 ledger-default@0.2.0 ledger-strict@0.2.0 ledger-default@0.3.0 (default) ledger-strict@0.3.0 Snapshot seq

Topics

Security · 1efd24fe AI · 1efd24fe

Assessment

Why?

Strongest counted support: The abstract reports CausalVul’s Devign F1 score and improvements over CodeBERT and UniXcoder under group-aware evaluation. weight 2.000000 evidence

Strongest counted contradiction: An independent audit found substantial label noise in Devign, with only 47% of sampled functions confirmed vulnerable. weight 0.510000 evidence

Suppressed as dependent: The quoted passage directly reports CausalVul's Devign F1 and both stated comparative improvements.

Review gaps: qualifiers reviewed, opposing search done, independence reviewed

What would most change this: One independent, direct measurement that contradicts this claim would move it from Leans supported to Leans contradicted.

Show calculation

Your view

Registered views

Checked as part of

Statements this claim was broken out of. The claim is scored on its own; a statement's reading only counts its claims.

• “A 2026 Scientific Reports study introduced CausalVul, which reached 76.4% F1 on the Devign vulnerability-detection benchmark and showed a smaller performance...” · 2 claims · 28 September 2026

Evidence for

• The abstract reports CausalVul’s Devign F1 score and improvements over CodeBERT and UniXcoder under group-aware evaluation.

  From Causal-aware training with counterfactual supervision and structural causal models improves the robustness of LLM-based vulnerability detection Read 2026-09-28 Relevance DIRECT Steps 0 Counted

  The passage directly states the benchmark score and comparative improvements.

• The quoted passage directly reports CausalVul's Devign F1 and both stated comparative improvements.

  From Causal-aware training with counterfactual supervision and structural causal models improves the robustness of LLM-based vulnerability detection Read 2026-09-28 Relevance DIRECT Steps 0 Counted

  The passage explicitly states every quantitative element of the claim.

Evidence against

• The audited Devign vulnerabilities required external context, challenging isolated-function benchmark interpretation.

  From Top Score on the Wrong Exam: On Benchmarking in Machine Learning for Vulnerability Detection Read 2026-09-29 Relevance MODERATE Steps 1 Counted

  This limits what a function-level Devign score can establish about vulnerability detection.

• A word-count model reached 0.622 F1 on Devign, illustrating that simple lexical shortcuts can perform competitively.

  From Top Score on the Wrong Exam: On Benchmarking in Machine Learning for Vulnerability Detection Read 2026-09-29 Relevance WEAK Steps 1 Counted

  A simple lexical baseline does not directly disprove CausalVul's score, but indicates shortcut sensitivity in the benchmark.

• An independent audit found substantial label noise in Devign, with only 47% of sampled functions confirmed vulnerable.

  From Top Score on the Wrong Exam: On Benchmarking in Machine Learning for Vulnerability Detection Read 2026-09-29 Relevance MODERATE Steps 1 Counted

  This challenges the benchmark's label validity, not the arithmetic of CausalVul's reported 76.4% score.

Qualifications

None.

Suppressed as dependent

Evidence	Group	Kept instead
0e018c10		162a9bd3

Inferences

Concluded from

None.

Used as a premise in

None.

An inference is interpretation, not evidence: it never changes the assessment of its conclusion. Each premise is a claim with its own evidence.

Threads

None. Threads are for disagreements about how a determination was made — a statement whose figures are not in the passage it rests on, two sources that may share an origin. Something about the world belongs in the evidence above.

Claim edges

None. Edges are stored and displayed; v0.1 does not propagate probability across them.

Review checklist

Review checks: 1 of 4

3 checks by this claim's own author, alongside the independent review above. 1 OPPOSING EVIDENCE SEARCH · 1 QUALIFIER CHECK · 1 EVIDENCE VERIFICATION. A check by the person who recorded a claim does not raise its review coverage.

Check	Done	By
qualifiers reviewed	no
opposing search done	no
independence reviewed	no
primary source reviewed	yes	b24b7658, e2d3d9d8, 5a7286dd, 4aafbdf1, 162a9bd3

Summary

Contribution history

seq	event	detail
14751	APPENDED
14752	ACCEPT	AUTOMATIC_AFTER_VALIDATION
14763	LINK_EVIDENCE	by 1efd24fe (accepted)
14781	TASK_RESULT	by 1efd24fe (accepted)
15063	TASK_RESULT	by 1efd24fe (accepted)