ParaSuit official ICSE 2026 page:
https://conf.researchr.org/details/icse-2026/icse-2026-research-track/222/Enhancing-Symbolic-Execution-with-Self-Configuring-Parameters

ParaSuit official artifact:
https://github.com/skkusal/ParaSuit

Audited artifact HEAD on 2026-08-13 UTC:
e991924b827db80b117e1f5389fe07820bad13f0

Audited algorithm:
Paper Algorithm 2 and artifact parasuit/value_sample.py. The paper explores
when silhouette <= threshold; the implementation exploits when silhouette >=
threshold. F397 follows the official implementation at equality. F397 uses a
deterministic dependency-free bounded MeanShift adaptation, normalized value
axis, scheduler reward and worker-cost correction; it does not claim sklearn
decision or random-sequence equivalence.

Upstream results are not local results:
The paper evaluates KLEE 2.1 on 12 GNU C programs for 24 hours with five
repetitions. Its reported 25.8% branch-coverage advantage over Symtuner, 11
unique bugs (four exclusive), and value-stage ablation gains belong only to
that study. F397 performs no public-target campaign and does not reuse those
numbers as SymCC results.