Paper Digest¶
Daily Scholar-alert digest. Each entry below links to that day's full report. Use the search box top-right to query across the whole archive.
2026-05-06¶
Two Outstanding deep-reads land today, the second of which is the day's user-curated pick. DirtyFree (Lee, Kwon, Holz — Max Planck Institute for Security and Privacy + Theori, NDSS 2026) collapses the three-stage Data-Oriented Programming exploit chain — heap-leak / arbitrary-address-read / arbitrary-address-write — into a single arbitrary-free…
2026-05-04¶
An empty alert window. No new Scholar-alert threads landed in the last 24 hours, and no user-curated self-emails were forwarded for review either. The most recent Scholar batch resolves to 2026-05-02 — already absorbed into the 2026-05-03 report (NeuroTaint and CoRE) — and the most recent user-curated paper is from 2026-05-01. A broader fallback…
2026-05-03¶
Two Outstanding deep-reads today, both arriving via followed-researcher Scholar alerts (no user-curated self-emails landed inside the window). The first, NeuroTaint (Cai, Tang, Wen, Qin — HKUST + Xidian), reframes information-flow tracking for LLM agents around three orthogonal propagation classes — explicit content propagation, *implicit control…
2026-05-02¶
A quiet alert window today: only one fresh paper makes it through to deep-read, and it is a user-flagged self-email rather than a Scholar alert. The day's Outstanding paper, VulWeaver (Cao, Chen, Hu, Bihuan Chen and collaborators at Fudan University and Huawei), targets the same Java vulnerability-detection terrain that Phoenix mapped…
2026-05-01¶
A four-paper Scholar-alert day with no user-curated picks. The day's clear standout is AnalysisAgent (Bouzenia, Cadar, Pradel — CISPA + Imperial), which formalises automated software analysis as an end-to-end agentic task and presents both a benchmark (AnalysisBench: 35 tool–project pairs across AFL++, KLEE, CSA, cflow, Infer, WALA, SJK on five…
2026-04-30¶
A heavy day, dominated by the user-curated queue: STEP 2b pulled in four self-flagged arXiv preprints, while the Scholar-alert side contributed exactly one EMSE-published static-analysis tool. The four user-picked papers cluster tightly around the same architectural recipe — **multi-agent orchestration with structural feedback grounded in deterministic…
2026-04-29¶
A modest alert window — five raw threads, two papers passed Stage-1 screening. Both come from followed-researcher channels and converge on the same architectural insight despite operating in completely different domains: decouple LLM production from verification, and ground every decision in immutable external evidence.
2026-04-28¶
The Scholar-alert side of this window was empty: the next batch from Google Scholar didn't arrive until late on 04-28 CST (and was caught by the 04-29 report). The user-curated self-email queue, however, contributed exactly one paper — flagged via the new STEP 2b pipeline that picks up self-addressed emails whose subject contains "paper" or "read".
2026-04-27¶
A thin alert window. Gmail's from:scholaralerts-noreply@google.com newer_than:1d query returned exactly one new alert thread — a "Recommended articles" delivery containing a single paper. None of the followed-researcher channels emitted a digest in this window; their previous batch (2026-04-25) was already absorbed into yesterday's 2026-04-26 report.
2026-04-26¶
This window's haul is dominated by three currents. (1) LLM-agentic software analysis is consolidating from a "can it work?" question into a benchmarking question — Pradel's group ships AnalysisBench. (2) Path/program-analysis classics are being made scalable for modern workloads — Hermes (Wu et al.) tackles path-sensitive pointer analysis under…
Pipeline: source-priority triage → Stage-1 Skip/Proceed → Stage-2 deep-read with meta-cognition. See the GitHub repo for the rubric and architecture.