Studies

Research done with pubmed-stream, kept with its data so the numbers in each study can be checked and the study rerun. A tutorial teaches one step of the tool on a small example; a study asks a research question, freezes what it downloaded, and records what it found. Each study is a folder with its own README, its scripts, its data and its notes. A study may also carry a web page (index.html with its app, stylesheet and a data file built from its files), published with the documentation site at melhzy.github.io/pubmed-stream/studies/<study>/.

Study	Page	Question	Status
abeta56_retraction_impact	Open the page	How far did the retracted 2006 Nature paper on Aβ*56 (PMID 16541076, retracted June 2024) spread through the literature, what exactly was wrong with it, and how much of it did language models learn? Continues tutorial 8.	Citation tree frozen; corpus string counts and PMC Open Access membership done; model probes designed and smoke-tested; a context-control unlearning pilot run

To start a study, copy the layout of an existing one: README.md (question, headline numbers, how to reproduce, status), docs/ (the write-up and sources), data/ (frozen tables, each dated), scripts/ (one script per step, runnable from the study folder), and probes/, notebooks/, figures/ or manuscript/ as needed. Everything a script downloads while reproducing goes in workspace/, which git ignores.