Experiment #25 · October 4, 2026 · version 1
Why the paper corpus stopped on May 5
Hypothesis
The paper corpus froze because paper_scout was not scheduled, while data_fetcher kept refreshing the homepage list.
Method
Compared pipeline_log.json, corpus.json, and the daily workflow, then ran paper_scout in discover-only mode with a 154-day lookback. No language model was called.
Result
The corpus moved from 86 papers (last run 2026-05-05) to 156 papers. The first detail fetch hit HTTP 429 and was retried on a second pass. Zero abstracts were analyzed, because Ollama and GEMINI_API_KEY are unavailable. Scripts 12 through 23 were not executed. Each now has a blocked published account, EXP-012 through EXP-023, with no computed findings.
Full write-up: data/experiment_025_paper_refresh_report.md.
updated by running the scout and recording the rate-limit retry.