paper-searchlisted

# paper-search Search OpenAlex + arXiv, fuse compact query variants, then rank by relevance and **impact relative to the paper's own field and age**. No API key or auth. ```bash python3 scripts/paper_search.py "agent runtime" -q "LLM agent infrastructure" python3 scripts/paper_search.py "distributed storage" --institution Alibaba python3 scripts/paper_search.py "crash consistency" --venue FAST python3 scripts/paper_search.py "agent memory" --fresh 60 --category cs.AI python3 scripts/paper_search.py --after arXiv:2307.03172 --about "position bias" python3 scripts/paper_search.py --selftest ``` Repeat `-q` for alternate terminology; papers matching several variants rank higher within their quality bucket. `--institution` resolves a lab/company and includes its descendants; `--institution-type company` searches industry broadly. Venue aliases include FAST, OSDI, SOSP, NSDI, MLSys, NeurIPS, ICML, ICLR, VLDB, and SIGMOD. `--after` takes `arXiv:ID`, `DOI:x`, or an OpenAlex `Wxxx`. Feed a useful displayed topic back through `--topic`. Use `--json` for machine output and `--help` for all flags. ## Reading the output Papers are bucketed, best first. The buckets are the point — **recency and quality are independent axes**, and collapsing them into one score hides exactly the tradeoff that matters in a fast-moving field. | Bucket | Means | |---|---| | `LANDMARK` | cited fast *and* far above its field — read this first | | `STRONG` | peer-reviewed, comfortably above field average |