Edward Cheng-I Wu
Edward Cheng-I Wu
## Context Carry-over from #103 Step 13 R8 codex review on commit \`e579478\` (final round). Codex returned 0 P0 / 0 P1 / 4 P2 — ship gate met, all...
## Context Carry-over from #103 Step 13 R6 codex review (P2 #2). The R4 closure for "lint defensive guard against malformed passport" (commit \`f0ca68b\`) added top-level coercion: \`_coerce_aggregate()\` filters non-list...
## Context Carry-over from #103 Step 13 R3 dual-track review (codex P2 #5). In `scripts/claim_audit_pipeline.py` the uncited-path judge invocation currently wraps `_invoke_judge` and on `JudgeInvocationError` substitutes a synthetic `{"judgment": "NOT_VIOLATED",...
## Motivation v3.7.3 ships `contamination_signals` with two vectors: - **Vector 1** `preprint_post_llm_inflection` — heuristic (year ≥ 2024 ∧ venue ∈ 10-server preprint list) - **Vector 2** `semantic_scholar_unmatched` — single-index lookup...
Re-prioritize DELEGATE-52 follow-up work after cross-model review (supersedes #83 priority section)
## Summary Issue #83 mapped the DELEGATE-52 paper (arXiv:2604.15597) findings against ARS and proposed 7 follow-up work items with Item 1 (corruption-detection lint) marked high priority. A cross-model review by...
## Problem When orchestrating ARS `full` mode by spawning agents one phase at a time (rather than letting a single orchestrator handle the whole pipeline), a Phase 2 agent (`bibliography_agent`...
## Scope v3.9.0 (#102, merged) shipped intentionally mirroring `scripts/semantic_scholar_client.py` structure across 3 clients (S2, OpenAlex, Crossref) and 2 migration tools (v3.7.3, v3.9.0). The mirror pattern was a design decision for...
## Scope v3.9.0 (#102, merged) shipped the **measurement layer** for cross-index triangulation. This issue tracks the **policy layer** — the gate-blocking + opt-in strict modes that v3.9.0 deliberately deferred. Spec...
## Background Carried over from #128 §4 — v3.9.3 (#128 §1-3, §5-6) shipped 5 pure refactors as one PR (#${TBD}). Item §4 was deferred because it introduces a behavior change...
## Background #133 surfaced a failure pattern where ARS routing logic silently dispatched a single phase agent (`bibliography_agent`) given ambiguous user input (already-written abstract + pre-collected literature), and that subagent...