Nothing. It searches.
This is the environment file as it stands today, not a record of this run. The run predates environment capture, so what it actually had on hand was never recorded and cannot be recovered.
# Find brief — papers worth reproducing, deep sweep
**Environment:** `night-shift-open` — full egress. The settings are on this
mission's page.
This is the brief handed verbatim to a full-network session, recorded here as
what the run was actually asked to do. It differs from FIND-002's brief in one
way that matters: FIND-002 ran from a session that could only reach GitHub, so
its six candidates are all GitHub-hosted. This run had the access to reach the
package hosts proper — Zenodo, Dryad, Dataverse, OSF — and was told to mine
them.
---
You are a research agent for a private research-verification board called
"Night Shift." You have FULL network access. Your entire job is to find papers
worth reproducing that have public, openable, runnable replication packages —
and to write your findings to two files on a branch. Nothing else.
## Step 0 — confirm your reach (do this first, report it in the output)
Actually fetch each of these and note whether it loads: zenodo.org, the AEA
studies search on openICPSR, dataverse.harvard.edu, osf.io, github.com,
codeocean.com. If any are blocked, say so in the Notes — do not pretend.
## The job
Find as many GENUINELY VERIFIED candidate papers as you can — aim for 40–100,
but quality strictly over quantity. A candidate qualifies ONLY if you actually
OPENED its replication package and saw what is in it.
Hard honesty rules (these are the whole point):
- Never invent a paper, a title, or a package URL. A fabricated row is far
worse than a short list. "Six real beat twelve invented."
- `package_read: yes` ONLY if you opened the package page/repo yourself in this
session. If you merely found it via search and did not open it, that
candidate does not go on the list at all.
- `verified_complete`: `yes` only if, having opened it, you saw data + code +
run instructions present; `no` if you opened it and the data is
gated/missing/must-be-reconstructed; `could-not-check` if you opened it but
genuinely can't tell.
- Prefer packages that are complete and runnable on PUBLIC data without special
hardware (no private admin data, no GPU clusters). Those are the most
valuable.
Cast across fields so the backlog isn't all one domain: economics (openICPSR
holds thousands of pre-verified AEA packages — mine it heavily), political
science (Harvard Dataverse), bioinformatics/genomics (GitHub, Zenodo), ecology,
psychology, statistics/methods.
## Output — two files, both required
`candidates.yaml`:
```yaml
candidates_file:
mission: FIND-003
source: output.md
count: <number of entries>
candidates:
- id: C01
title: <short title>
domain: <field>
package_url: <where the package lives>
package_read: yes
language: <R, Python, Stata, …>
verified_complete: yes | no | could-not-check
```
`output.md`: for each candidate, a short paragraph — the paper, where the
package lives, language, rough size, whether you opened it, and why it is worth
reproducing. End with a blunt **Notes on this run** section: which hosts you
could and couldn't reach, how you sampled, what you are unsure about, and any
host that holds a lot you didn't get to. Be honest that `package_read: yes`
means "opened and read," not "cloned and ran."
**These must agree.** The board counts entries in `candidates.yaml` — it never
parses your prose.
Borusyak, Hull and Jaravel, *Review of Economic Studies*. I opened the Zenodo record and read the zip's full central directory: 156 entries, 71 .dta, 25 .csv, 23 .do, 7 .R, 4 .ado (including the `ssaggregate` command shipped with the paper), plus README.pdf and README.docx and per-analysis masters (`master_adh.do`, `master_bartik1991.do`, `master_montecarlo.do`). Marked `no` only because the deposit's own abstract says it contains "all public data underlying the analyses, instructions for how remaining data can be accessed" — the Autor-Dorn-Hanson arm is public and self-contained, but not every input is. Still one of the most valuable things on this list: the shift-share methods literature leans on this package heavily, and the Monte Carlo master runs on nothing but the shipped code.https://zenodo.org/records/4619197
Palma, *Review of Economic Studies*. 2.8 MB, 44 entries read from the zip directory: `MASTER.do`, 12 do-files, two .dta datasets, logs and output figures, and a README.txt. The deposit states outright that the package "constructs the entire analysis from the data sources described in the published paper" and that "the replicator should expect the code to run for less than 10 minutes". Small, public, fast, fully self-contained — close to an ideal first reproduction.https://zenodo.org/records/5338362
Achdou, Han, Lasry, Lions and Moll, *Review of Economic Studies*. 1.4 MB; the zip directory shows 67 entries — 23 .m files including `main.m` and `main_dist.m`, 39 .eps target figures, README.pdf and a plot-data text file. This is a solver, not a data analysis: the figures come out of the HJB/Fokker-Planck code with no external data at all, so there is nothing to gate. Verifying it means checking that the shipped .eps figures come back out of `main.m`.https://zenodo.org/records/4357052
McKay and Reis, *Review of Economic Studies*. The zip is a mirror of the authors' `OptStab` repo: 467 entries, 77 .m files, README.md at the root, and — importantly — the ten FRED series the calibration needs (`CorePCE.csv`, `UNEMPLOY.csv`, `fedfunds.csv`, …) shipped in a `data/FRED` folder. Public data, no licensing, code + README present.https://zenodo.org/records/3925391
Igami and Sugaya, *Review of Economic Studies*. 1.7 MB, 123 entries: 79 .m files (GMM objective, bootstrap and counterfactual scripts), 27 .mat data files including `data.mat` and `worldbank_hic2.mat`, 9 .eps figures and README_version_3.pdf. The deposit says it "contains the data and the code". Compact structural IO with everything in the box.https://zenodo.org/records/5104830
*Review of Economic Studies*. Only 534 KB: 33 entries, `!Replication_main.do` as the entry point, two .dta files, one .R script, 16 LaTeX table outputs to check against, and README.pdf. Tiny, fully shipped, and a well-cited cross-country result — a cheap reproduction with real citation weight.https://zenodo.org/records/5589052
Bilbiie, *Review of Economic Studies*. The smallest thing on this list: 114 KB, three files — one .m, one .xlsx of inputs, one readme.txt — and the deposit is explicit that it replicates Figure 1. Narrow scope, but it is complete, it is honest about its scope, and it takes minutes.https://zenodo.org/records/10937947
Kroft, Laliberté, Leal-Vizcaíno and Notowidigdo. 63 MB, 34 entries read from the zip directory: 16 do-files with `HMS_quarterly_master.do` and `RMS_quarterly_master.do` as the two entry points, 9 .dta datasets actually shipped, a Python helper and "README KLLN.pdf". Data-preparation and analysis both live in the package.https://zenodo.org/records/7526306
Dhyne, Kikkawa, Mogstad and Tintelnot, *Review of Economic Studies*. 407 MB, 326 entries; I read the README.txt straight out of the remote zip. It ships 188 Comtrade .csv files and a chain of masters (`clean_data_master.do`, `construct_share_main_master.do`, …), and the data availability statement documents which Comtrade extracts are included — but the Belgian firm-to-firm administrative data at the core of the paper is not. Marked `no`: the public part is real and substantial, the headline result is not reproducible without an NBB agreement.https://zenodo.org/records/3997900
Banerjee, Breza, Chandrasekhar and Golub, *Review of Economic Studies*. 4.5 MB, 111 entries: 16 do-files and 3 R scripts named per table/figure (`table_1.do`, `figure_4.R`, …), 6 .dta files, 2 .ado helpers, plus survey instruments, a variable key and protocol documents. The deposit claims "all data, codes, documentation required to reproduce figures and tables" and the manifest backs that up.https://zenodo.org/records/7860884
Gentry, Komarova and Schiraldi, *Review of Economic Studies*. 21 MB, 161 entries: 39 .m, 6 .do, 2 .jl, 35 .dta, 3 .jld, Readme.pdf and Readme.docx, with clearly named appendix drivers (`main_appendixG1_0_solve_eqm.m`, `main_appendixG2_0_run_mc_sims.jl`). Interesting as a cross-language reproduction: MATLAB estimation, Julia Monte Carlo, Stata tables.https://zenodo.org/records/6304625
Adusumilli, *Econometrica*. 19 MB, 166 entries: 30 .m files, 8 .Rmd notebooks with their knitted .html, 22 .csv, README.pdf and README.txt. Mostly simulation, so it is largely hardware-independent, and the Rmd/html pairs give a built-in check — knit them and diff against the shipped output.https://zenodo.org/records/13776876
Hsu and Shen, *Quantitative Economics*. 12 MB, 84 entries: 28 R scripts, 6 .dta and 6 .csv data files, 2 do-files, and readme.docx plus a separate note on the Online Appendix E simulations. A methods paper with its simulations shipped is a good verification target because the simulation arm needs no data at all.https://zenodo.org/records/12037888
*Econometrica*. 1.1 MB, 64 entries: 20 R scripts, 12 .csv, 2 .dta, README.pdf. Marked `no` for an honest reason the deposit states itself — "one dataset on the Gaokao experiment (Section 6.2) remains confidential", available only under a data use agreement with the original authors. Everything else is there, so a partial reproduction is realistic; a full one is not.https://zenodo.org/records/8336416
Chiappori, Fiorio, Galichon and Verzillo, *Econometrica*. 440 KB, 78 entries: `00_Master.do`, `04_MasterR.R`, 18 do-files, 12 R scripts, 8 Python scripts, README.pdf and a "Notes_for_Running_Py.txt". Run instructions and code are unambiguous; what I could not settle from the manifest is the data — at 440 KB there is no microdata in the box, and the .xls/.xlsx entries look like parameter inputs rather than the estimation sample. Someone should open the README before committing time.https://zenodo.org/records/17301245
Han, Yang and E, *Quantitative Economics*. 7.3 MB, 448 entries: 40 Python files, 12 .m including `MAIN.m`, 6 notebooks, 24 .mat, README.pdf and !README.txt, and SLURM submission scripts (`simple_run.sbatch`) that reveal the intended cluster shape. A neural-network solver, so it will want a GPU to be pleasant — flagged rather than hidden.https://zenodo.org/records/18019831
Miyauchi, *Econometrica*. 890 KB, 228 entries, 54 R scripts with a clean master chain (`MASTER_0_PROCESS_RAW_TSR_DATA.R` → `MASTER_1_DATA_REDUCED_FORM.R` → `MASTER_3_MODEL_SENSITIVITY.R`) and README.md/README.pdf. The name of the first master gives it away: the Tokyo Shoko Research firm-to-firm data is licensed and not in the package, which at 890 KB obviously holds no microdata. Code and instructions are exemplary; the data is not obtainable.https://zenodo.org/records/13765850
Ferrero, Harrison and Nelson, *Review of Economic Studies*. 2 MB, 402 entries, 78 .m files built on the Bank of England MAPS toolkit, 16 .eps target figures and 2 .xlsx inputs. The deposit says it "contains all the code necessary to replicate the figures". I could not find a README in the manifest — the documents present are the figure PDFs — so I cannot claim run instructions are there.https://zenodo.org/records/7551954
Dávila and Hébert, *Review of Economic Studies*. 29 MB, 66 entries: 53 .m files with `run_all.m` and `main_corporate_tax_dynamic.m` as entry points, 3 .mat solution caches, 8 .eps target figures and ReadMe.pdf. Self-contained computation, single-command driver.https://zenodo.org/records/7044154
11.6 MB, 631 entries: 48 .m files, 11 Dynare `.mod` model files, 13 .mat and a set of alphabetically ordered drivers (`a_run_baseline.m`, `b_run_monpol.m`, `c_run_NFA.m`, …) that make the intended run order self-documenting. The deposit says the readme "contains all the information required to implement the package". Dynare is a free MATLAB/Octave toolbox, so the dependency is tractable.https://zenodo.org/records/10844051
Corbae and D'Erasmo, *Review of Economic Studies*. 87 MB, 63 entries: 14 .f90 sources with `main_simple01.f90`, 7 do-files, 8 .csv, a data appendix PDF. Marked `no` because the deposit says "detailed instructions are also given about accessing the raw data" — the Compustat inputs are licensed. Notable anyway as one of the few Fortran packages in the sample.https://zenodo.org/records/4287356
Buera, Kaboski and Shin, *Review of Economic Studies*. 7.3 MB, 234 entries, 68 .csv, 4 do-files, README.txt and a separate figures readme. Same pattern as the previous entry: the deposit says instructions are given "about accessing the raw data", i.e. the Thai and microfinance survey inputs are not in the box.https://zenodo.org/records/3959907
Hollingsworth, Soni, Carroll, Cawley and Simon, *PLOS ONE* 2019 — a Zenodo archive of the `hollina/elections_and_insurance` GitHub repo. 259 MB, 81 entries: `master.do`, 18 do-files, 7 .dta, README.md, plus logs and .gph graph files to diff against. Public election and ACS/BRFSS-derived data, open-access journal, one-command driver.https://zenodo.org/records/2630581
Bauer, Lakdawala and Mueller, *Economic Journal* 2021. 4.9 MB, 72 entries: 21 .m, 6 .R, 22 .csv and 5 .RData files, 6 target figure PDFs. Data and code are clearly both present. I could not identify a README in the manifest (the only .txt is `swanson.txt`, a data input), so I cannot assert run instructions exist.https://zenodo.org/records/5566246
Squires, *Economic Journal* 2024. 2.7 MB, 160 entries: `main.do` plus 50 do-files, 4 .dta and 4 .csv, README.md and README.pdf, and 8 documentation PDFs. A field experiment whose own collected data is shipped — the good case for a development RCT.https://zenodo.org/records/10888407
Alexander and Schwandt. 367 MB, 642 entries: 56 do-files with `diesel_main_analysis_FINAL.do`, README.pdf, the manuscript, 26 .tex table outputs and ArcGIS project files. Only 4 .dta files against 56 do-files, and the outcome data is US restricted-use natality/birth records — marked `no`.https://zenodo.org/records/5794422
Koren and Pető, *PLOS ONE* 2020. 142 MB, 202 entries: a `Makefile` (rare and welcome in economics), 29 do-files, 43 .csv, 15 .dta, README.pdf and multiple README.md files at directory level. Built on O*NET and public employment data, open-access journal, make-driven. One of the most reproducible-looking economics packages I opened.https://zenodo.org/records/4016325
Dincecco, Fenske, Menon and Mukherjee, *Economic Journal*. 33 MB, 64 entries: 18 .dta files shipped, a complete ESRI shapefile set (.shp/.dbf/.prj/.shx/.sbn/.cpg), 11 .tex table outputs and README.pdf. Historical and geographic data with no living-subject restrictions, so the whole chain is in the box.https://zenodo.org/records/5583263
Hoffmann and Rud, *Econometrica* 2024. 5.2 GB, 1272 entries read via the zip's ZIP64 central directory: `master.do`, 12 do-files, 58 .dta, 8 .qmd Quarto documents, shapefiles for the Mexico City maps, README.pdf. The deposit's contents list — README, raw data, do-files, map folders, output folders — matches what I saw. Large but genuinely complete; budget the disk.https://zenodo.org/records/11060907
Agneman and Chevrot-Bianco, *Economic Journal* 2022. 3.4 MB, 111 entries: `00_MAIN.do` plus 28 do-files, 3 .ado, and full field documentation (experimental design, instructions in three languages, ethics approval). But the manifest holds no .dta or .csv — the outputs are 35 .gph graphs and 16 .tex tables. The survey microdata is not in the package, so marked `no`.https://zenodo.org/records/6974389
Karbownik and Wray, *Journal of Human Resources*. 26 MB, 605 entries, `0_run_all.do`, a proper README.md with an "Overview" section (I read the opening of it), 50 .eps and 32 .tex outputs to check against. Only 2 .dta files though, against linked historical census/vital records — I could not tell from the manifest whether those two are the analysis samples or lookup tables.https://zenodo.org/records/7117307
Queiró, 2021. 15 MB, 111 entries: `run_all.ipynb` and `main.ipynb`, an `environment.yml` pinning the Python environment, 19 do-files, README.pdf, and 45 .png / 20 .tex outputs. The deposit explicitly says it "contains publicly available data and code". A pinned conda environment plus a single notebook driver is about as good as this genre gets.https://zenodo.org/records/5464854
*Econometrica*. 6.3 MB, 488 entries: `0.master.R` and `3.master.do`, 34 R scripts, 21 do-files, 2 .RData, "00.README.pdf" and the IRB approval letter. The directory names in the archive (`raw_data_R`, `input_data_stata`, `cleaning_code_R`, `analysis_R`, `final_paper_replication`) spell out the pipeline. A factory-level RCT with its own collected data shipped.https://zenodo.org/records/11373971
Adamopoulou and Kaya, *Canadian Journal of Economics*. Opened the Borealis record and listed all 9 files: 6 do-files (`INPS_data_preparation.do`, `ISMU_tables.do`, …), README.txt, and two .tab files that are `coefficient_estimates` and `demo_istat`. At 103 KB total there is no microdata — the Italian INPS social-security and ISMU immigrant survey inputs are not distributable. Code and README are open; data is not.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/NZNQBR
Bekkers, Francois and Manchin, *Canadian Journal of Economics*. Five files, all unrestricted: `finaldata.dta` (the bulk of a 1.2 GB deposit), `finaldata_onlyhomogeneous.tab`, `reproducing_tables.do`, `Figure1.xlsx`, and `Readme.pdf`. One do-file, one dataset, one readme — unusually clean, and a good candidate precisely because there is nowhere for ambiguity to hide.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/G55JDI
Bento, *Canadian Journal of Economics*. Nine files, 2.5 MB, none restricted: five .tab data files (US Business Register size distributions, `dataRose.tab`, `ag_empshare.tab`), three figure-specific do-files and ReadMe.pdf. Small enough to reproduce in an afternoon on public data.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/2HUCIH
Liu, Xu, Zhang and Zheng, *Canadian Journal of Economics* 2026. Nine files, 573 MB, none restricted: `master.do`, city-year-month and province-year .tab panels, an `adofiles.zip` bundling the required Stata packages (a detail that materially improves reproducibility), a `Results.zip` of expected output, ReadMe.pdf and an explicit LICENSE.txt.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/WZVH0A
Carter, Saitone and Schaefer, *Canadian Journal of Economics*. Ten files, 460 KB: `price_data.tab` and `quantity_data.tab`, three analysis do-files including a bootstrap script, table/figure .tab outputs to check against, and README.pdf. Tiny, public commodity price data, fully shipped.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/P2J35S
Atallah and Simons. Four files totalling 7.5 KB: `replicate_linear_case.py`, `requirements.txt`, README.pdf, LICENSE.txt. The record states the tables "rely exclusively on equilibrium computations using the linear model described in Appendix A. No external data" is needed. A pure-computation package with pinned dependencies — the cheapest possible verification, and a useful sanity check on the pipeline itself.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/2VWMN4
Moriconi, Peri and Pozzoli, *Canadian Journal of Economics*. Ten files, 553 KB, eight do-files and README.docx — and, decisively, a file literally named `data_access_instructions.pdf`. The Danish employer-employee registers are accessed through Statistics Denmark, not shipped. Marked `no`; listed because the code is fully open and readable even when the data is not.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/VBGZEP
Baker, Callaway, Cunningham, Goodman-Bacon and Sant'Anna, *Journal of Economic Literature*. I opened the repo: `data/`, `scripts/`, `figures/`, `tables/`, `renv.lock`, a GitHub Actions workflow, README.md and README.pdf. All input data is public domain (CDC, BLS, Census, KFF) merged into `county_mortality_data.csv`, and the README gives two parallel entry points — `scripts/R/00_master_did_jel.R` (~10 min) and `scripts/Stata/00_stata_master_did_jel.do` (~11-15 min) — that are supposed to produce identical figures and tables. Public data, pinned R environment, dual implementation, ten-minute runtime, heavily cited. If the board only reproduces one thing, this is my pick.https://github.com/pedrohcgs/JEL-DiD
Heddesheimer, Hilbig and Voeten, *American Political Science Review* 2025. Opened the repo: `code/`, `data/`, `codebooks/`, `results/`, README.md. The README is unusually good — three run paths (`source("code/master_full_replication.R")`, step-by-step, or individual scripts), a stated ~30-minute runtime, and explicit requirements (R 4.4.0+, 8 GB RAM, ~5 GB disk). Marked `no` only because the German SOEP survey, GLES candidate data and party press releases need separate institutional access; the county-level election and economic data are public and the final analysis datasets ship.
https://github.com/vincentheddesheimer/green_transition_replication
Davidson and Enos. Opened the repo: `data-collection/`, `cleaning-and-merging/`, `models/scripts/`, `results/`, README.md, which lays out the intended order clearly. Marked `no`: the raw social media data is not provided (only example API query scripts), and the larger fitted `brms` models live behind a Google Drive link rather than in the repo. Model estimation also assumed a SLURM cluster.https://github.com/t-davidson/engaging-populism-replication
Davis, Dingel, Monras and Morales, *Journal of Political Economy*. Opened the repo: a top-level `Makefile`, `readme.md` and `readme.pdf`, `local_configuration.sh` / `slurm_configuration.sh`, `tasks_flow_graph.png`, `initialdata/`, and 46 task directories. The README documents `make` / `make full_version` versus `make quick_version`. Flagging the cost honestly: the quick version wants ~70 GB of disk and under a day; the full version wants ~4 TB and several days. Complete and make-driven, but not a light reproduction.https://github.com/jdingel/DavisDingelMonrasMorales
Azevedo, Deng, Montiel-Olea, Rao and Weyl, *Journal of Political Economy*. Opened the repo: `bash/`, `r/`, `matlab/`, `data/`, `output/`, `codebook.txt`, `output-list.txt`, `readme.txt`, with `bash/make.sh` as the single driver. The readme says plainly that "the data used is proprietary" and directs users to Microsoft. Clean pipeline, unobtainable data — listed as a documented negative.https://github.com/eduardomazevedo/admrw-ab-replication-code
Bonhomme, Lamadon and Manresa, *Econometrica*. Opened the repo: `R/`, `inst/`, `packrat/`, `DockerConfig/`, a `Dockerfile`, DESCRIPTION/NAMESPACE, README.md/.html/.pdf. Two documented routes — a prebuilt Docker image exposing RStudio, or packrat restore plus `source("inst/main.R")`. Marked `no` because the Swedish IFAU administrative data requires an application and confidentiality agreement; the package runs on synthetic data by default. Still worth listing: the Docker + packrat setup is a model of environment capture even where data cannot follow.
https://github.com/tlamadon/blm-replicate
Santucci (Oxford UP, 2022). Opened the repo: per-chapter directories `ch1_files` … `ch7_files_new`, an `img/` folder, and README.md. Historical STV referendum, roll-call and municipal election data is included, which is the hard part. But the README gives no run instructions and describes the repo as "a work-in-progress", so I cannot say the third leg is there.https://github.com/jacksantucci/mponp
Korthauer et al. Opened the repo: `PREPROCESSING/`, `ANALYSIS/`, `BENCHMARK/`, `FIGURES/CODE/`, README.md. The README specifies the order (preprocessing → analysis → figures) and pins the analysis to a specific `dmrseq` commit. Crucially, every dataset is public — ROADMAP human tissue, DNMT3A murine leukemia and dendritic cell data pulled from SRA, GEO and NCBI FTP by the preprocessing scripts. Scripts assume SLURM but state they can run serially. Genuinely open end-to-end; the cost is sequencing-scale compute, not access.https://github.com/kdkorthauer/dmrseqPaper
Shah lab, *Nature Communications* 2024. Opened the repo: `Snakefile`, `config.yaml`, `cluster.json`, `rules/`, `scripts/`, `notebooks/`, `envs/`, `data/`, MIT LICENSE, README.md. The README states that all source data needed to reproduce the figures, plus file-structure documentation, is in a companion Zenodo deposit (10.5281/zenodo.12786373) — public, just not in the git repo. Conda envs are checked in and the workflow is Snakemake-orchestrated. Users must adjust paths, which is exactly the kind of friction a verification board should record.https://github.com/shahcompbio/scdna_replication_paper
128 MB, 61 entries read from the zip: 19 Jupyter notebooks, 8 .csv and 5 .npy data arrays, and a README.md that I read out of the archive — it names the Python version (3.10+) and gives the pip install line for the companion TopACT package. Data, code and instructions all present; the one external dependency is a public GitLab package.https://zenodo.org/records/11050996
661 MB, 1941 entries: 152 gzipped expression/annotation files covering, per the record, "over 1800 RNA-seq samples", 22 R scripts and 22 .csv, plus a README. The deposit is a single self-contained data-and-analysis bundle rather than a code-only archive, which is what makes it worth the disk.https://zenodo.org/records/51842
Lai et al., *International Journal of Cancer* 2022. 257 MB, 280 entries: 133 .csv node/edge tables, 20 .RData, 5 R scripts, a Python script, 2 pickles and `readme.pdf`. The deposit says it "contains the data, code, and analysis to recreate the results". Derived network data rather than patient-level data, so nothing is gated.https://zenodo.org/records/10788609
Groningen Microbiome Centre, *Nature* 2023. 71 KB, 51 entries: 12 R, 12 shell, 9 Python, 7 .Rmd and a README.md that I read from the archive — it explains the two-folder structure (microbiome SV calling, then GWAS). Marked `no` for the obvious reason: 71 KB is code only, and the underlying cohort genotypes and metagenomes are access-controlled. Good code to read, not a reproduction target.https://zenodo.org/records/10018199
Wu et al., *Nature Genetics* 2021 — Zenodo archive of `Swarbricklab-code/BrCa_cell_atlas`. 360 KB, 123 entries: 59 R scripts, 12 shell scripts, two Snakefiles (CIBERSORTx, DWLS) and a README.md whose "Data Availability" section I read. The processed scRNA-seq is downloadable from the Broad Single Cell Portal rather than shipped, and at 360 KB nothing is in the box. Marked `no`; the data is public but the package is not self-contained.https://zenodo.org/records/5031502
Bodenmiller group. 94 KB, 27 entries: 20 .Rmd analysis notebooks and a README.md. I read the README from the archive — it is a single line, the repository name. No data, no instructions. Listed as a checked negative so nobody spends time rediscovering it.https://zenodo.org/records/7540622
T2T Consortium (Hoyt et al.), for the CHM13 reference. 507 MB, 220 entries: 116 gzipped data files, 8 README.md files at directory level, 4 Perl and 2 R scripts. I read the top README, which describes the organisation ("organized by type of analysis conducted") but is not a run procedure. Data is plainly present; whether the instructions are sufficient I cannot say from the manifest.
https://zenodo.org/records/5895031
Arrondo et al. 2026. 236 KB, 23 entries: `01_analysis_main.R` plus 9 more R scripts, 3 .csv, `CITATION.cff`, and a README.md I read from the archive — it carries MIT (code) / CC-BY-4.0 (data) badges and states which outputs the pipeline reproduces (the facility × movement-component matrix, the bivariate world map, waffle plots, Supplementary Tables S1-S2). A review dataset of 134 studies, so no field data access is needed. Small, licensed, explicitly scoped.https://zenodo.org/records/20311943
Siqueira et al. 4.5 MB, 129 entries: 30 .csv, 14 R scripts and two .Rproj files. The deposit description itself is the run instruction and is unusually specific — it names the five scripts in order (`01_Dataprep_stability_metrics.R`, `02_SEM_analyses.R`, `03_Stab_figs.R`, `04_Stab_supp_m.R`, `05_Sensit_analysis.R`) and says the full result set including supplementary material comes out of running them in that sequence.https://zenodo.org/records/8333128
Hillebrand, Donohue, Harpole et al. 267 KB, 53 entries: 37 .csv and 13 R scripts with a README.md that I read out of the archive. Small, self-contained, and a genuinely contested claim — a high-profile negative result about ecological thresholds is exactly the kind of thing worth independently re-running.https://zenodo.org/records/3828869
Hanson et al. 2020. 1.3 GB, 671 entries: a top-level `Makefile`, 47 README.md files at directory level, 11 .rda, rasters and shapefiles. Marked `no` on the deposit's own statement: "many of the raw data files (e.g. extent of suitable habitat maps, protected area data) are not available in this archive, and must be obtained from the original sources". The make-based structure is excellent; the inputs are not all here.https://zenodo.org/records/1035485
Senf and Seidl, *Nature Sustainability* 2020. 63 MB, 80 entries: 38 .csv, four complete shapefile sets, 4 R scripts, README.md/README.txt (I read the top README from the archive). The catch is stated in that README — the disturbance maps themselves are hosted externally and the repository is a static snapshot of a project the authors say they "continuously improve and update". Whether the archived csvs suffice without re-downloading the maps, I could not determine.https://zenodo.org/records/3925447
Barneche et al., *Global Ecology & Biogeography*, archived from `dbarneche/geb12851`. 4.9 MB, only 19 entries: 7 .csv, 3 R scripts, 2 .yml (a `remake`-style build spec) and README.md, which I read from the archive and which opens with "This repository contains code and data needed to reproduce the article". Nineteen files, a declarative build file, public reef survey data. Very high effort-to-value ratio.https://zenodo.org/records/1484591
*PLOS ONE*. 4.9 GB, 1712 entries: 90 .csv, 60 .m, 9 R scripts, plus README_FIRST.txt, a top-level README.txt and per-subdirectory READMEs — I read the top one, which explicitly tells you that each subdirectory carries its own README and which directory (`ECOTRAN_Code`) you need to run the model elsewhere. Large but layered and documented; the ECOTRAN model is reusable beyond this paper.https://zenodo.org/records/8393490
Lowman et al. 2024. 90 MB, 26 entries: 18 R scripts, 5 .rds fitted objects, 2 `.stan` model files and `README_for_repository`, whose opening I read — it names the creators and their institutions and points to the publication for the analysis description. Stan models shipped alongside cached fits means you can verify both the refit and the downstream figures.https://zenodo.org/records/10455823
4.1 GB, 61 entries: 10 R scripts, 23 .rds posterior objects, 11 .csv, a shell driver, `readme.pdf` and a README.md I read from the archive — it describes a fully Bayesian analysis of an original survey built on protection motivation theory. Survey data plus posteriors plus code plus documentation. The 4 GB is almost entirely cached MCMC draws, so the reproduction can be staged.https://zenodo.org/records/16369904
Dryad, 134 KB, 4 files: `AustraliaFlockData2019Dryad.csv`, `Foraging_observations__2019_Dryad.csv`, `Species_code_information.xlsx` and README.md, which I read in full on the landing page — it sets out the four axes of behavioural plasticity being tested. Dryad curation requires the README, and the record links a preprint DOI and a software DOI. Trivially small, fully open.https://datadryad.org/dataset/doi:10.5061/dryad.0k6djhb7h
*Animal Behaviour* 2025 (primary article DOI linked from the record). Dryad, 9.1 MB: `rabidosa_signal_comparison.csv`, `total_raw.csv` and a README.md I read in full — it enumerates four JupyterLab notebooks (`0_train_GHMM.ipynb`, `1_segmentation.ipynb`, …), one R script, and lists the exact Python packages each notebook needs. That level of dependency detail in a README is rare and makes this an easy verification.https://datadryad.org/dataset/doi:10.5061/dryad.mgqnk9989
Lattanzio, *Journal of Animal Ecology* 2026. Dryad, 1.3 MB, 4 files: `dev_resampler.R`, `spr_resampler.R`, `Lattanzio_2026_RawData.xlsx` and README.md, read in full — it names each file and says the workbook contains the raw and processed data used for analyses, figures and supporting information. Raw data plus resampling helpers plus documentation.https://datadryad.org/dataset/doi:10.5061/dryad.mpg4f4rgq
Dryad, 120 KB, exactly three files: `data.csv`, `Meta-analysis.R`, `README.md` — README read in full, describing a phylogenetic multilevel meta-analysis with overall models, parasite-group models, publication-bias tests and sensitivity analyses. Three files, 120 KB, phylogenetic meta-analysis. Probably the single cheapest verification on this list.https://datadryad.org/dataset/doi:10.5061/dryad.k0p2ngfq0
Dryad, 1.2 MB, 5 files including one R script implementing six movement-based calving-date methods (BCPA, FPT, MRT, rMCP, net-squared displacement), `ParturitionDates.csv`, `Video-based_observations.csv`, an xlsx of NSD values and README.md read in full. The paper benchmarks movement-based inference against animal-borne video ground truth — a method comparison, which makes re-running it more informative than most.https://datadryad.org/dataset/doi:10.5061/dryad.cjsxksnmh
Pereira et al., *Behavioral Ecology* 2025 (primary article DOI on the record). Dryad, 21 KB — twenty-one kilobytes — for `data_FAP.csv`, `data_Interactions.csv`, `script.Rmd` and a README.md I read in full, with citation, originator and collection-date sections. One Rmd, two csvs, a real journal article behind it.https://datadryad.org/dataset/doi:10.5061/dryad.3ffbg79zm
Dryad, 749 KB, 8 files: six .csv covering foraging, growth, lethality and three behaviour datasets, `nogueira_2026_lethality_code.Rmd`, and README.md read in full — it describes the two-phase mesocosm design with crab predators and snail prey. A controlled experiment with all arms shipped.https://datadryad.org/dataset/doi:10.5061/dryad.6djh9w1hj
Dryad, 304 KB, 5 files: seasonal summer/winter datasets, `DMD_D_analysis.R`, `Linearity_D_analysis.R`, README.md read in full — it documents every column and explains why some covariates (bear density, breeding status; moose and roe deer density) appear in only one season. Column-level documentation is exactly what makes an independent re-run possible.https://datadryad.org/dataset/doi:10.5061/dryad.280gb5n39
*Journal of Avian Biology* (primary article DOI on the record). Dryad, 2.5 MB, 5 files: `ANH_code_upload.R`, ecoacoustic sample results, `selections6.csv`, `transect_gradients.csv` and README.md read in full — it even warns which column belongs to a different companion study and is dropped during analysis. That kind of caveat is a good sign.https://datadryad.org/dataset/doi:10.5061/dryad.pvmcvdp1q
Cady et al., *Ecology*. Dryad, 89 KB, 3 files: `R_Code_Cady_etAl_Community_Stability2024.R`, `SpeciesList.csv`, README.md. Marked `no` on the README's own first line about data: "User must download 5 open-source datasets and 1 dataset from Cady et al. (2023) on Dryad before running". The inputs are open, but they are not in this package — reconstruction required.https://datadryad.org/dataset/doi:10.5061/dryad.v6wwpzh1g
Dryad, 93 KB, 6 files: five .csv covering tracking, qPCR and amputation experiments, plus a README.md I read in full which documents every column and is headed "Raw Excel Data and Code". But no script file appears in the file listing — the analysis code the title implies is not visibly there. Flagged rather than assumed either way.https://datadryad.org/dataset/doi:10.5061/dryad.1rn8pk110
McDonald et al., *Nature Communications* 2026. Dryad, 1.4 GB, two files: `DataDryadArchive_McDonald_et_al_final.zip` and README.md, which I read in full — it states that all code and data used in the paper is here. I could not list the zip's interior: unlike Zenodo, Dryad's file endpoints require an API bearer token, so the range-request trick I used elsewhere does not work. High-profile result, plausibly complete, contents unverified.https://datadryad.org/dataset/doi:10.5061/dryad.905qfttz0
*Global Ecology & Biogeography*. Dryad, 263 MB, 4 files: `R_scripts_and_data.zip`, `HTV.zip`, a species list xlsx and README.md read in full — it names each script inside the archive (`range-diversity analysis.R`, `hotspots analysis.R`, `biome_analysis.R`, `PAs analysis.R`) and what each does. Script-level documentation from outside the archive is what lets me call this complete.https://datadryad.org/dataset/doi:10.5061/dryad.m905qfv8g
Dryad, 163 KB, 8 files, and the structure is the reason it is here: `SOFTWARE.zip`, `DATA.zip`, `DOCUMENTATION.zip`, `VALIDATION.zip`, `MANIFEST.txt`, `SHA256SUMS.txt`, README.md and README.txt. The README (read in full) describes the omega_RNA / rho_comp statistics for RNA secondary-structure evolution and gives a package build date. A shipped validation suite plus checksums is close to best-practice for a methods deposit.https://datadryad.org/dataset/doi:10.5061/dryad.w3r22816w
Claus et al. Dryad, 55 MB, two files: `Claus_et_al_2026.zip` and README.md, read in full — it explains the project analyses NEON carabid observational data to determine sampling effort needed for richness and diversity estimates. NEON data is open, which is promising, but the archive interior is not listable without a Dryad token.https://datadryad.org/dataset/doi:10.5061/dryad.0k6djhbfn
*Palaeogeography, Palaeoclimatology, Palaeoecology* 2025. Dryad, 52 KB, 5 files: `BioGeoBEARS_code.R`, `Anazygidae_tree.tre`, `Anazygidae_geoareas.data`, `b_AICc_table.csv` and README.md read in full — it even records the R version used (4.3.0) and explains the geoareas file format. Tiny, complete, and a good test of whether BioGeoBEARS model selection is stable across package versions.https://datadryad.org/dataset/doi:10.5061/dryad.d51c5b0fs
Dryad, 283 MB, 8 files, organised as one zip per pipeline stage — `01_read_processing.zip` through `07_biogeography.zip` — plus a README.md I read in full that walks through each stage and marks which files are Dryad-only (large alignments) versus mirrored on GitHub. Raw reads live at NCBI SRA (PRJNA1336315), but processed alignments ship, so the downstream phylogenetics is reproducible without re-assembling reads.https://datadryad.org/dataset/doi:10.5061/dryad.mkkwh71dd
*Communications Biology* 2026 (primary article DOI on the record). Dryad, 30 MB, 25 files: `Bee_swimming_paper_analysis.Rmd`, 18 figure-specific .csv files named `Fig1A.data.csv` onward, five .mov videos and README.md read in full — it documents each csv down to what the measured angle means and when the experiment was run. Per-figure data files make partial verification easy.https://datadryad.org/dataset/doi:10.5061/dryad.w9ghx3g2r
Dryad, 2.3 MB, 5 files: `data_corrected.zip`, `scripts_corrected.zip`, `make-public-data.R`, README.md and — notably — `CHANGELOG_correction.md`. A deposit that versions its own corrections is a good verification subject: the changelog gives you a documented before/after to check against.https://datadryad.org/dataset/doi:10.5061/dryad.hx3ffbgkx
Bohdalkova and Storch, *Global Ecology & Biogeography* (primary article DOI 10.1111/geb.70270 on the record). Dryad, 2.4 MB, 17 files: ten numbered R scripts (`1_loading_subsampling_data.R` through `10_corespecies_analyses.R`) and six .RData processed datasets. I read the full README on the landing page: raw Breeding Bird Survey data must be downloaded separately for script 1, but "all subsequent scripts use the processed files provided here". Scripts 2-10 are reproducible as shipped, which is an honest and usable arrangement.https://datadryad.org/dataset/doi:10.5061/dryad.5dv41nshz
Dryad records the related publication ISSN as 0022-0477 (*Journal of Ecology*); the record itself lists no primary-article DOI, so treat the journal attribution as the ISSN's, not a checked citation. 392 KB, 8 files: `NovNat_WP3_PSF_Analysis.R` (a single 197 KB script) plus six .xlsx trait and feedback datasets and a ~10 KB README.md. Small, curated, single-script analysis with all inputs present.https://datadryad.org/dataset/doi:10.5061/dryad.z34tmpgwn
Nine files, 876 KB, none restricted: anonymised experiment and pilot data plus event-level data, four R scripts (`GazeModel_Bootstrapping.R` and three area-of-interest models for credits, reels and win displays), `DataDictionary.txt` and README.txt. Anonymised eye-tracking data that is actually distributable is unusual in this subfield.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP2/GSOMLU
Conway et al. Nine files, 9.9 MB, none restricted: `1_conway_et_al_measurement_code.Rmd` with both .docx and .pdf renderings of the same code, English and French lemmatization tables, substitution lists, `8_conway_et_al_corpus_data.RData`, and `0_read_me.txt`. Numbered filenames give the run order without needing to guess.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/MFZTWZ
Choi, Zambrano Garza, Pauly et al. 2025. Seven files, 2.5 MB: `Dataverse_analytic code.Rmd`, `Code for figures.R`, README.txt, and — the problem — the only data present is `f1_plot_data.tab`/`f1_summary.RData`/`f2_plot_data.tab`/`f2_summary.RData`, i.e. figure-level summaries. The participant-level daily-diary data behind the models is not deposited. Marked `no`.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/LLQGA4
Nine files, 148 KB, none restricted: six numbered R scripts running from `1_analyze_protist_videos_Patrick_SH.R` to `6_protist_analysis.R`, two .csv, and a Dryad cross-reference JSON. Data and code are both here and the numbering implies the order, but there is no README, so I will not claim run instructions are present.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP2/SJUDDO
*Canadian Journal of Fisheries and Aquatic Sciences*. Four files, 239 KB, none restricted: `CJFAS_L-Erie-Foodweb-analysis.Rmd` and three .tab datasets (fish data, SIBER inputs). An Rmd is close to self-documenting, but there is no README and no stated package requirements, so this is a `could-not-check` rather than a `yes`.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/6ZGCDT
Eight files, 1.9 MB, none restricted: three numbered .Rmd analyses (feeding, growth, otolith chemistry), an `.Rproj`, three .tab datasets and `00.README_Redfish_experiment.txt`. A controlled six-month feeding experiment at ISMER/Maurice-Lamontagne with all arms deposited.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/C7PW8A
Eight files, 2.9 MB, none restricted: `all_code_R_Markdown.Rmd` with its knitted `.html` (so you have the author's own output to diff against), four .tab datasets covering feeding and water behaviours and the Latin-square design, a raw-data example, and `ReadMe.txt`. Shipping the knitted HTML alongside the Rmd is the detail that makes this an easy verification.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/JLDGQG
Danyk et al., *Journal of Field Ornithology* 2025. Eight files, 72 KB, none restricted, in a real directory structure: `Data/01_Local_movement/` and `Data/02_Behaviour_and_habitat_use/` .tab files, `Danyk_et_al_2025_statistical_analysis.Rmd`, an `.Rproj`, and `README_Danyk_et_al_2025.txt`. 72 KB total.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/KKWVBG
Nine files including six R scripts (data manipulation, final models for 0-6m and 6-60m cohorts, learning curves, z-scores), two post-processed .rds and a data dictionary. All nine files are flagged restricted in the Dataverse record — I could see the manifest but nothing is downloadable without approval. Listed as a clean negative: good code structure, gated paediatric data.https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/M3OPKQ
Preregistered replications with exploratory statistical analyses. I walked the OSF storage tree: 35 nodes across `material/`, `data/`, `readme/` and `supplementary materials/`, with per-experiment readmes (`readme_experiment1.txt`, `readme_experiment2.txt`), participant-level `data_exp1_all.csv` and `data_exp2_all.csv`, variable-definition PDFs, six .Rmd analysis files and five `.stan` models, and a separate satisficer-exclusion branch under supplementary materials. Preregistered plus Bayesian plus full data is a strong combination.https://osf.io/f2xsz/
A preregistered replication of Wolsko et al. (2016), run as a Registered Report (the accepted stage-1 manuscript is in the project). 17 nodes split into `Direct Replication (DR)/` and `Conceptual Replication (CR)/`, each with .xlsx data and separate preprocessing and analysis Rmd files — including both the *planned* scripts (`DR-script-planned-data-analysis.Rmd`) and the executed ones, plus a codebook and materials PDF. Having planned and actual scripts side by side is unusually good for verification.https://osf.io/2dqu6/
A preregistered replication of Kornell & Bjork (2008). Eight nodes: `data/IdentificationTask.csv` and `data/2AFCTask.csv`, `scripts/Identification.Rmd` and `scripts/AFC.Rmd`, the timestamped preregistration PDF, and a supplementary-materials PDF. Two tasks, two datasets, two Rmds, 4.9 MB.https://osf.io/u436s/
13 nodes: `rawdata.zip`, `SBVM_data_aggregated.csv`, two generations of analysis notebooks (`SBVM_stage2.Rmd` and `SBVM_stage2rev.Rmd`) each with knitted HTML, an `IPA Files/` folder holding the in-principle-acceptance proposal and its R code, and a transparency checklist. Raw and aggregated data both present, plus the original and revised analyses — you can check whether the revision changed the conclusion.https://osf.io/k2dbf/
A preregistered replication of Experiment 3 in Glikson et al. (2018). 16 nodes organised as `Data and code/{Excel,SPSS,Codebook,Code}/`, `Experimental material/` (including the Qualtrics `.qsf` survey definition), and a `Statcheck/` folder with statcheck output run against the original paper's reported statistics. Shipping the Qualtrics instrument and a statcheck audit of the target paper is exactly the posture a verification board wants.
https://osf.io/n7yc4/
46 nodes, 2.8 GB: `code/` with `behavioral_analysis.m`, `erp_analysis.m` and `make_figures.m`; `data/behavioral_summary.mat`; `outputs/{figures,tables,models}/`; a top-level `README`; and — the reason this one is unusually self-contained — a bundled `eeglab2019_0.zip`, pinning the toolbox version rather than assuming the reproducer installs a compatible one.
https://osf.io/gf83h/
Submitted to *BMC Psychiatry*. 53 nodes: `datos/` (the harmonised cohort of 63,268 F30/F31 discharge episodes, national population projections, all-cause discharge counts), `codigo/` with 13 R scripts, `figuras/`, `resultados/`, `README.md` and a `MANIFIESTO.txt` inventory. Built from Chile's open DEIS and INE data, and it ships a download manifest so the provenance chain is checkable.https://osf.io/y67f8/
60 nodes organised as `Code for Data ID 1-3/` through `Code for Data ID 10/` plus `Meta-analysis/` and `Independent rating/`, holding 16 do-files and 17 .Rmd files and README.txt files at branch level. The structure is a genuine multi-study replication series, which is interesting. I could not confirm from the tree whether participant-level data accompanies every study branch, so I am not claiming completeness.https://osf.io/cbe7g/
22 nodes: `run_all.do` at the root, `code/`, `data/README_data_access.md`, `output/` with non-disclosive aggregate model results, `CITATION.cff`, `LICENSE` and README.md. The record is candid that it ships "Stata replication code and non-disclosive aggregate outputs" — the CFPS panel microdata must be obtained from the data provider. Marked `no`, but the aggregate outputs mean a partial check against published tables is still possible.https://osf.io/nuzvy/
Orozco-Acosta et al., *Biometrical Journal* 65(8), 2023. Opened the repo: `Rcode/`, `data/`, `results/`, `Scalable_Prediction.Rproj`, README.md. The README organises scripts by paper section and documents dependencies precisely (R 4.2.0, INLA, the authors' `bigDM` package). Marked `no` for a reason worth recording: the real cancer mortality data for Spain's 7,907 municipalities is confidential, and the deposit substitutes structurally comparable synthetic data. You can verify the method, not the published numbers.https://github.com/spatialstatisticsupna/Scalable_Prediction
One team's pipeline from the EEGManyPipelines multi-analyst project. 1.2 MB, 33 entries: 16 Python scripts (`03_run_ica.py`, `06_run_autoreject.py`, …), `environment.yml`, `README.md` read from the archive, a `CITATION.cff`, and two instruction PDFs ("Instructions for analysts", "Instructions for the script folder"). The README carries a GitHub Actions "Run analysis" badge, so the pipeline is CI-exercised. Marked `no` on data only: the shared EEG dataset is distributed by the EEGManyPipelines organisers, not bundled. Valuable as a study of analytic variability rather than a single result.
https://zenodo.org/records/6549063
Preston lab. 2.2 MB, 73 entries: 25 Python files, 26 shell scripts, 6 notebooks, `pyproject`-style `.toml`, README.md read from the archive. The README is refreshingly honest — it says the steps are "designed to reproduce all analyses presented in the paper" but that "the procedure has not been repeated since the original run, and may be missing some steps", and that it was tested on the Lonestar 6 cluster. Code and instructions yes; imaging data is external, and the authors themselves flag the instructions as possibly incomplete.https://zenodo.org/records/6967583
Hulsey et al. 2023. 7.1 MB, 223 entries: 77 Python files, 14 notebooks, seven README.md files at directory level, an `environment.yml`, and a Cython extension. Marked `no` because the record states the related NWB data files are deposited on DANDI (10.48324/dandi.000678) rather than here — public, but a separate fetch. A good example of the increasingly common code-here/data-there split.https://zenodo.org/records/10306018
666 KB, 223 entries: 12 Python scripts including `03_run_ica.py`, an `environment.yml`, a `CITATION.cff`, README.md and a plain `README`, plus 91 .json and 27 .tsv files that look like BIDS sidecars rather than the recordings themselves. Environment and code are pinned and present; whether the EEG data needed to execute the pipeline is included, the manifest does not settle.https://zenodo.org/records/7335171
Holt, Igami and Scheidegger. 145 MB, 29 entries; I read the README.md out of the remote zip — the authors explicitly offer the repository as "a sandbox for researchers interested in studying gasoline price data". Ships 20 Python files including LSTM and estimation frameworks, a `replication_conda_environment.yml`, a README PDF and the paper-plus-appendix PDF. A methods-comparison paper (classifiers for retail-fuel price cycles) with a pinned conda environment and public price data.https://zenodo.org/records/10126406
39 MB, 322 entries. I read the README.md from inside the zip: it is a proper setup guide (virtualenv recommended, module install, then the scripts). Ships 198 .csv files of yearly cosine distances plus a `requirements.txt` and multiple nested READMEs including one for the Jaffe measures. Built on public USPTO patent statistics, so nothing is gated.
---
## Notes on this run
### Step 0 — what I could actually reach
I fetched all six required hosts and confirmed each with a real HTTP status code, not a vibe.
| Host | Result |
|---|---|
| `https://zenodo.org` | **200 — reachable.** Web pages and the full REST API both work. This became the backbone of the sweep. |
| `https://www.openicpsr.org/openicpsr/search/aea/studies` | **403 — blocked.** Cloudflare interstitial ("Just a moment...") on every path I tried, including `/openicpsr/oai` and individual project pages, with and without a browser User-Agent. |
| `https://dataverse.harvard.edu` | **202 with an empty body — blocked.** The response carries `x-amzn-waf-action: challenge`; AWS WAF is challenging the request and returning nothing. Both the web UI and `/api/*` are affected. |
| `https://osf.io` | **200 — reachable.** `api.osf.io/v2` works (occasional transient 502s, retried). |
| `https://github.com` | **Reachable, but only through some paths.** `raw.githubusercontent.com` returns 200. `api.github.com` is bound to this session's configured repository and refuses search and arbitrary-repo endpoints. `github.com` HTML and `codeload.github.com` return 403 to `curl`. The WebFetch tool *can* render github.com repository pages, and that is how every GitHub candidate here was opened. |
| `https://codeocean.com` | **403 to `curl` at the root; `/explore` returns a 200 SPA shell** with no server-rendered content, and `/api/v1/capsules` returns 403/404. I could not enumerate or open a single Code Ocean capsule, so **no Code Ocean candidates appear on this list.** |
I tried to get around the two Cloudflare/WAF blocks with the pre-installed Chromium via Playwright, on the theory that a real browser would clear the JS challenge. It failed for an unrelated reason: Chromium cannot reach the network through this session's egress proxy at all — every request, including `https://example.com`, dies with `ERR_CONNECTION_RESET`, with the proxy configured explicitly, with QUIC and HTTP/2 disabled, and with certificate errors ignored. `curl` to the same URL returns 200. So there was no browser route available and I stopped after four attempts.
### What the openICPSR block cost, and what I did instead
This is the biggest hole in the run. openICPSR holds thousands of AEA replication packages that have already been through the AEA Data Editor's verification — exactly the pre-screened population the brief pointed at, and I got **zero** of them.
What I did instead was find the *other* economics deposit venues that are reachable:
- **Zenodo turned out to be the Econometric Society's and the Review of Economic Studies' deposit host.** A large share of the economics rows here are official `Replication package for: ...` deposits for *Econometrica*, *Quantitative Economics*, *Review of Economic Studies*, *Economic Journal* and *Journal of Human Resources* — the same data-editor-verified population as openICPSR, just at a different DOI prefix. That substitution is why economics is still the largest domain on this list.
- **Borealis (`borealisdata.ca`)** carries the *Canadian Journal of Economics* replication deposits, with unrestricted Stata code and data.
Other non-Harvard Dataverse instances I confirmed reachable but only lightly sampled or did not use: `dataverse.no` (763 replication datasets), `edmond.mpg.de` (Max Planck), `data.sciencespo.fr`, `demo.dataverse.org`. `dataverse.nl` reset the connection. **Harvard Dataverse remains the single largest untouched pool** — it is where most political-science replication data lives, and its absence is why political science is thin here (the poli-sci rows I do have came from GitHub instead).
### How I sampled
Not randomly, and not exhaustively. Roughly:
- **Zenodo** — keyword searches per domain (economics, genomics, ecology, psychology, statistics) restricted to `type=dataset` and `type=software`, sorted by `mostviewed` rather than `mostrecent`. That sort choice matters: sorting by recency surfaced a lot of very new, unreviewed, sometimes plausibly machine-generated "replication packages" with no journal behind them. Sorting by views pulled up established journal deposits instead. I discarded the recency-sorted batch.
- **Dryad** — keyword searches through `/api/v2/search`, then the version file listing for each hit, then the dataset landing page (Dryad renders the full curated README in the page HTML). Dryad requires a README and links a primary article, so the hit rate for genuinely complete packages was the highest of any host.
- **Dataverse** — searching at the *file* level (`fileName:*.do`, `fileName:*.R`, `fileName:*.py`, `fileName:*.Rmd`) rather than the dataset level, then resolving parent datasets. Dataset-level keyword search on these instances returns mostly raw survey data with no code; file-level search finds the packages that actually contain analysis scripts. This was the single most useful trick of the run.
- **OSF** — `api.osf.io/v2/nodes` filtered by title and tags, then walking `osfstorage` recursively to build the file tree.
- **GitHub** — web search to find candidates, then WebFetch on each repository page to read the tree and the README. Slower and one-at-a-time, but it is a real read.
### How "opened" was established, per host
This matters for judging how much to trust each row:
- **Zenodo** — for every zipped package I read the **zip central directory over HTTP range requests**, which yields the complete internal file manifest without downloading the archive. This worked on packages up to ~6 GB (the "Networks, Barriers, and Trade" deposit: 573 entries read from about 100 KB of transfer, ZIP64 and all). Where a README lived inside the zip I range-fetched and inflated that single member and read it. So for most Zenodo rows I have seen the actual internal structure, not just the outer file name.
- **Dryad** — API file listing plus the rendered README on the landing page. **I could not read inside Dryad's `.zip` files:** Dryad's file-download endpoint requires an API bearer token (`401 Unauthorized`) and the public `file_stream` path returns 403. Every Dryad row whose payload is a single zip is therefore marked `could-not-check`, and I have said so on each.
- **Dataverse** — full file manifest with per-file sizes and the `restricted` flag, via `/api/datasets/:persistentId/versions/:latest/files`. The restricted flag is why the Uganda paediatric-sepsis row is a confident `no`.
- **OSF** — recursive `osfstorage` walk giving the real folder tree.
- **GitHub** — the rendered repository page: top-level tree plus README.
### On the count
The brief asked for 40-100 and this list has 111. I did not pad it: every row is something I opened, and the overshoot is mostly the Zenodo range-request trick working better than expected — once reading a 6 GB package's manifest costs 100 KB and a few seconds, the marginal cost of one more verified row collapses. If the board wants to enforce the ceiling, the honest thing to cut first is the 15 `could-not-check` rows, not the 25 `no` rows: the negatives are checked facts, the could-not-checks are open questions.
### What I am not sure about
- **`could-not-check` is doing real work here and I have not rounded it to `yes`.** It means one of: I saw code and data but no README (several Dataverse and a few Zenodo rows); or I saw a README that describes the contents without giving a run procedure; or the payload is a Dryad zip I could not open. Each row says which.
- **A manifest is not an execution.** Seeing `master.do` next to twelve `.dta` files is strong evidence and not proof. Paths may be absolute, dependencies unpinned, a `.dta` may be a lookup table rather than the estimation sample. Several `yes` rows will fail on first contact, and finding out which is the point of the exercise.
- **I judged "data present" partly by file extension and size.** A 440 KB economics package with eighteen do-files almost certainly has no microdata in it — that inference is behind the `could-not-check` on "Assortative Matching on Income". It is an inference, not a reading of the README.
- **Journal attributions come from the deposits' own metadata**, not from checking the published article. Where a deposit says "forthcoming in *Review of Economic Studies*", I repeated that; I did not verify publication status.
- **I have not deduplicated against whatever is already on the board.** The repo's history mentions an earlier six-paper GitHub run (FIND-002); there may be overlap I did not check for, since the brief was to write only these two files.
- **Domain balance is uneven and openICPSR is the reason.** Economics and ecology/evolution dominate. Political science is under-represented because Harvard Dataverse was unreachable. Psychology comes almost entirely from OSF preregistered-replication projects, which skews toward small experiments. Bioinformatics rows skew toward code-with-external-data, which is simply how that field deposits.
### Hosts holding a lot I did not get to
- **openICPSR / AEA** — thousands of pre-verified packages. Completely blocked. Highest-value target for a rerun from a network that can clear Cloudflare.
- **Harvard Dataverse** — blocked by AWS WAF. The main political-science pool.
- **Code Ocean** — reachable only as an empty SPA shell; its whole point is executable capsules, which would have been the most directly verifiable artifacts on offer. Zero coverage.
- **Dryad** — I sampled maybe 74 datasets against a corpus in the tens of thousands, and only across ecology, evolution and behaviour. Barely scratched.
- **Zenodo** — the Econometric Society / RES deposit series alone runs to hundreds of packages; I opened a few dozen. `dataverse.no`, `edmond.mpg.de` and `data.sciencespo.fr` are all reachable and essentially unsampled.
- **figshare** — API returns 200 and I confirmed reachability, but I ran out of run before sampling it at all.
https://zenodo.org/records/7957848
# Repro-candidate sweep: 111 verified candidates
Every candidate below was **opened in this session** — its package page, its API file manifest, its zip central directory, or its repository tree. Nothing here is a search hit I did not click. `package_read: yes` means *opened and read*, not *cloned and ran*: I did not execute any of this code.
Breakdown of `verified_complete`: **71 yes**, **25 no**, **15 could-not-check**. `no` almost always means the code and documentation are open but some input data is gated, licensed, or must be fetched from elsewhere — those rows are kept deliberately, because a documented negative saves the next person the same hour.
---
### C01 — Quasi-Experimental Shift-Share Research Designs
*economics (econometrics) · Stata + R · verified_complete: **no***
https://zenodo.org/records/4619197
Borusyak, Hull and Jaravel, *Review of Economic Studies*. I opened the Zenodo record and read the zip's full central directory: 156 entries, 71 .dta, 25 .csv, 23 .do, 7 .R, 4 .ado (including the `ssaggregate` command shipped with the paper), plus README.pdf and README.docx and per-analysis masters (`master_adh.do`, `master_bartik1991.do`, `master_montecarlo.do`). Marked `no` only because the deposit's own abstract says it contains "all public data underlying the analyses, instructions for how remaining data can be accessed" — the Autor-Dorn-Hanson arm is public and self-contained, but not every input is. Still one of the most valuable things on this list: the shift-share methods literature leans on this package heavily, and the Monte Carlo master runs on nothing but the shipped code.
### C02 — The Real Effects of Monetary Expansions: Evidence from a Large-Scale Historical Experiment
*economics (macro/economic history) · Stata · verified_complete: **yes***
https://zenodo.org/records/5338362
Palma, *Review of Economic Studies*. 2.8 MB, 44 entries read from the zip directory: `MASTER.do`, 12 do-files, two .dta datasets, logs and output figures, and a README.txt. The deposit states outright that the package "constructs the entire analysis from the data sources described in the published paper" and that "the replicator should expect the code to run for less than 10 minutes". Small, public, fast, fully self-contained — close to an ideal first reproduction.
### C03 — Income and Wealth Distribution in Macroeconomics: A Continuous-Time Approach
*economics (macro) · MATLAB · verified_complete: **yes***
https://zenodo.org/records/4357052
Achdou, Han, Lasry, Lions and Moll, *Review of Economic Studies*. 1.4 MB; the zip directory shows 67 entries — 23 .m files including `main.m` and `main_dist.m`, 39 .eps target figures, README.pdf and a plot-data text file. This is a solver, not a data analysis: the figures come out of the HJB/Fokker-Planck code with no external data at all, so there is nothing to gate. Verifying it means checking that the shipped .eps figures come back out of `main.m`.
### C04 — Optimal Automatic Stabilizers
*economics (macro/public finance) · Python + MATLAB · verified_complete: **yes***
https://zenodo.org/records/3925391
McKay and Reis, *Review of Economic Studies*. The zip is a mirror of the authors' `OptStab` repo: 467 entries, 77 .m files, README.md at the root, and — importantly — the ten FRED series the calibration needs (`CorePCE.csv`, `UNEMPLOY.csv`, `fedfunds.csv`, …) shipped in a `data/FRED` folder. Public data, no licensing, code + README present.
### C05 — Measuring the Incentive to Collude: The Vitamin Cartels, 1990-1999
*economics (industrial organization) · MATLAB · verified_complete: **yes***
https://zenodo.org/records/5104830
Igami and Sugaya, *Review of Economic Studies*. 1.7 MB, 123 entries: 79 .m files (GMM objective, bootstrap and counterfactual scripts), 27 .mat data files including `data.mat` and `worldbank_hic2.mat`, 9 .eps figures and README_version_3.pdf. The deposit says it "contains the data and the code". Compact structural IO with everything in the box.
### C06 — Patience and Comparative Development
*economics (development/behavioural) · Stata + R · verified_complete: **yes***
https://zenodo.org/records/5589052
*Review of Economic Studies*. Only 534 KB: 33 entries, `!Replication_main.do` as the entry point, two .dta files, one .R script, 16 LaTeX table outputs to check against, and README.pdf. Tiny, fully shipped, and a well-cited cross-country result — a cheap reproduction with real citation weight.
### C07 — Monetary Policy and Heterogeneity: An Analytical Framework
*economics (macro) · MATLAB · verified_complete: **yes***
https://zenodo.org/records/10937947
Bilbiie, *Review of Economic Studies*. The smallest thing on this list: 114 KB, three files — one .m, one .xlsx of inputs, one readme.txt — and the deposit is explicit that it replicates Figure 1. Narrow scope, but it is complete, it is honest about its scope, and it takes minutes.
### C08 — Salience and Taxation with Imperfect Competition
*economics (public finance) · Stata · verified_complete: **yes***
https://zenodo.org/records/7526306
Kroft, Laliberté, Leal-Vizcaíno and Notowidigdo. 63 MB, 34 entries read from the zip directory: 16 do-files with `HMS_quarterly_master.do` and `RMS_quarterly_master.do` as the two entry points, 9 .dta datasets actually shipped, a Python helper and "README KLLN.pdf". Data-preparation and analysis both live in the package.
### C09 — Trade and Domestic Production Networks
*economics (trade) · Stata · verified_complete: **no***
https://zenodo.org/records/3997900
Dhyne, Kikkawa, Mogstad and Tintelnot, *Review of Economic Studies*. 407 MB, 326 entries; I read the README.txt straight out of the remote zip. It ships 188 Comtrade .csv files and a chain of masters (`clean_data_master.do`, `construct_share_main_master.do`, …), and the data availability statement documents which Comtrade extracts are included — but the Belgian firm-to-firm administrative data at the core of the paper is not. Marked `no`: the public part is real and substantial, the headline result is not reproducible without an NBB agreement.
### C10 — When Less Is More: Experimental Evidence on Information Delivery During India's Demonetization
*economics (development/networks) · Stata + R · verified_complete: **yes***
https://zenodo.org/records/7860884
Banerjee, Breza, Chandrasekhar and Golub, *Review of Economic Studies*. 4.5 MB, 111 entries: 16 do-files and 3 R scripts named per table/figure (`table_1.do`, `figure_4.R`, …), 6 .dta files, 2 .ado helpers, plus survey instruments, a variable key and protocol documents. The deposit claims "all data, codes, documentation required to reproduce figures and tables" and the manifest backs that up.
### C11 — Preferences and Performance in Simultaneous First-Price Auctions: A Structural Analysis
*economics (industrial organization) · MATLAB + Julia + Stata · verified_complete: **yes***
https://zenodo.org/records/6304625
Gentry, Komarova and Schiraldi, *Review of Economic Studies*. 21 MB, 161 entries: 39 .m, 6 .do, 2 .jl, 35 .dta, 3 .jld, Readme.pdf and Readme.docx, with clearly named appendix drivers (`main_appendixG1_0_solve_eqm.m`, `main_appendixG2_0_run_mc_sims.jl`). Interesting as a cross-language reproduction: MATLAB estimation, Julia Monte Carlo, Stata tables.
### C12 — Risk and Optimal Policies in Bandit Experiments
*economics/statistics (experimental design) · MATLAB + R · verified_complete: **yes***
https://zenodo.org/records/13776876
Adusumilli, *Econometrica*. 19 MB, 166 entries: 30 .m files, 8 .Rmd notebooks with their knitted .html, 22 .csv, README.pdf and README.txt. Mostly simulation, so it is largely hardware-independent, and the Rmd/html pairs give a built-in check — knit them and diff against the shipped output.
### C13 — Dynamic Regression Discontinuity under Treatment Effect Heterogeneity
*economics (econometrics) · R + Stata · verified_complete: **yes***
https://zenodo.org/records/12037888
Hsu and Shen, *Quantitative Economics*. 12 MB, 84 entries: 28 R scripts, 6 .dta and 6 .csv data files, 2 do-files, and readme.docx plus a separate note on the Online Appendix E simulations. A methods paper with its simulations shipped is a good verification target because the simulation arm needs no data at all.
### C14 — Randomization Tests for Peer Effects in Group Formation Experiments
*economics (econometrics/experiments) · R + Stata · verified_complete: **no***
https://zenodo.org/records/8336416
*Econometrica*. 1.1 MB, 64 entries: 20 R scripts, 12 .csv, 2 .dta, README.pdf. Marked `no` for an honest reason the deposit states itself — "one dataset on the Gaokao experiment (Section 6.2) remains confidential", available only under a data use agreement with the original authors. Everything else is there, so a partial reproduction is realistic; a full one is not.
### C15 — Assortative Matching on Income
*economics (labour/family) · Stata + R + Python · verified_complete: **could-not-check***
https://zenodo.org/records/17301245
Chiappori, Fiorio, Galichon and Verzillo, *Econometrica*. 440 KB, 78 entries: `00_Master.do`, `04_MasterR.R`, 18 do-files, 12 R scripts, 8 Python scripts, README.pdf and a "Notes_for_Running_Py.txt". Run instructions and code are unambiguous; what I could not settle from the manifest is the data — at 440 KB there is no microdata in the box, and the .xls/.xlsx entries look like parameter inputs rather than the estimation sample. Someone should open the README before committing time.
### C16 — DeepHAM: A Global Solution Method for Heterogeneous Agent Models with Aggregate Shocks
*economics (computational macro) · Python + MATLAB · verified_complete: **yes***
https://zenodo.org/records/18019831
Han, Yang and E, *Quantitative Economics*. 7.3 MB, 448 entries: 40 Python files, 12 .m including `MAIN.m`, 6 notebooks, 24 .mat, README.pdf and !README.txt, and SLURM submission scripts (`simple_run.sbatch`) that reveal the intended cluster shape. A neural-network solver, so it will want a GPU to be pleasant — flagged rather than hidden.
### C17 — Matching and Agglomeration: Theory and Evidence from Japanese Firm-to-Firm Trade
*economics (trade/urban) · R · verified_complete: **no***
https://zenodo.org/records/13765850
Miyauchi, *Econometrica*. 890 KB, 228 entries, 54 R scripts with a clean master chain (`MASTER_0_PROCESS_RAW_TSR_DATA.R` → `MASTER_1_DATA_REDUCED_FORM.R` → `MASTER_3_MODEL_SENSITIVITY.R`) and README.md/README.pdf. The name of the first master gives it away: the Tokyo Shoko Research firm-to-firm data is licensed and not in the package, which at 890 KB obviously holds no microdata. Code and instructions are exemplary; the data is not obtainable.
### C18 — House Price Dynamics, Optimal LTV Limits and the Liquidity Trap
*economics (macro/housing) · MATLAB · verified_complete: **could-not-check***
https://zenodo.org/records/7551954
Ferrero, Harrison and Nelson, *Review of Economic Studies*. 2 MB, 402 entries, 78 .m files built on the Bank of England MAPS toolkit, 16 .eps target figures and 2 .xlsx inputs. The deposit says it "contains all the code necessary to replicate the figures". I could not find a README in the manifest — the documents present are the figure PDFs — so I cannot claim run instructions are there.
### C19 — Optimal Corporate Taxation Under Financial Frictions
*economics (public finance/macro) · MATLAB · verified_complete: **yes***
https://zenodo.org/records/7044154
Dávila and Hébert, *Review of Economic Studies*. 29 MB, 66 entries: 53 .m files with `run_all.m` and `main_corporate_tax_dynamic.m` as entry points, 3 .mat solution caches, 8 .eps target figures and ReadMe.pdf. Self-contained computation, single-command driver.
### C20 — Trade Wars, Nominal Rigidities and Monetary Policy
*economics (international macro) · MATLAB + Dynare · verified_complete: **yes***
https://zenodo.org/records/10844051
11.6 MB, 631 entries: 48 .m files, 11 Dynare `.mod` model files, 13 .mat and a set of alphabetically ordered drivers (`a_run_baseline.m`, `b_run_monpol.m`, `c_run_NFA.m`, …) that make the intended run order self-documenting. The deposit says the readme "contains all the information required to implement the package". Dynare is a free MATLAB/Octave toolbox, so the dependency is tractable.
### C21 — Reorganization or Liquidation: Bankruptcy Choice and Firm Dynamics
*economics (macro/corporate finance) · Fortran + Stata + MATLAB · verified_complete: **no***
https://zenodo.org/records/4287356
Corbae and D'Erasmo, *Review of Economic Studies*. 87 MB, 63 entries: 14 .f90 sources with `main_simple01.f90`, 7 do-files, 8 .csv, a data appendix PDF. Marked `no` because the deposit says "detailed instructions are also given about accessing the raw data" — the Compustat inputs are licensed. Notable anyway as one of the few Fortran packages in the sample.
### C22 — The Macroeconomics of Microfinance
*economics (development macro) · Stata + MATLAB · verified_complete: **no***
https://zenodo.org/records/3959907
Buera, Kaboski and Shin, *Review of Economic Studies*. 7.3 MB, 234 entries, 68 .csv, 4 do-files, README.txt and a separate figures readme. Same pattern as the previous entry: the deposit says instructions are given "about accessing the raw data", i.e. the Thai and microfinance survey inputs are not in the box.
### C23 — Gains in Health Insurance Coverage Explain Variation in Democratic Vote Share, 2008-2016
*political economy / health policy · Stata · verified_complete: **yes***
https://zenodo.org/records/2630581
Hollingsworth, Soni, Carroll, Cawley and Simon, *PLOS ONE* 2019 — a Zenodo archive of the `hollina/elections_and_insurance` GitHub repo. 259 MB, 81 entries: `master.do`, 18 do-files, 7 .dta, README.md, plus logs and .gph graph files to diff against. Public election and ACS/BRFSS-derived data, open-access journal, one-command driver.
### C24 — Market-Based Monetary Policy Uncertainty
*economics (finance/macro) · MATLAB + R · verified_complete: **could-not-check***
https://zenodo.org/records/5566246
Bauer, Lakdawala and Mueller, *Economic Journal* 2021. 4.9 MB, 72 entries: 21 .m, 6 .R, 22 .csv and 5 .RData files, 6 target figure PDFs. Data and code are clearly both present. I could not identify a README in the manifest (the only .txt is `swanson.txt`, a data input), so I cannot assert run instructions exist.
### C25 — Kinship Taxation as an Impediment to Growth: Experimental Evidence from Kenyan Microenterprises
*economics (development) · Stata · verified_complete: **yes***
https://zenodo.org/records/10888407
Squires, *Economic Journal* 2024. 2.7 MB, 160 entries: `main.do` plus 50 do-files, 4 .dta and 4 .csv, README.md and README.pdf, and 8 documentation PDFs. A field experiment whose own collected data is shipped — the good case for a development RCT.
### C26 — The Impact of Car Pollution on Infant and Child Health: Evidence from Emissions Cheating
*economics (environmental/health) · Stata · verified_complete: **no***
https://zenodo.org/records/5794422
Alexander and Schwandt. 367 MB, 642 entries: 56 do-files with `diesel_main_analysis_FINAL.do`, README.pdf, the manuscript, 26 .tex table outputs and ArcGIS project files. Only 4 .dta files against 56 do-files, and the outcome data is US restricted-use natality/birth records — marked `no`.
### C27 — Business Disruptions from Social Distancing
*economics (labour/COVID) · Stata + R · verified_complete: **yes***
https://zenodo.org/records/4016325
Koren and Pető, *PLOS ONE* 2020. 142 MB, 202 entries: a `Makefile` (rare and welcome in economics), 29 do-files, 43 .csv, 15 .dta, README.pdf and multiple README.md files at directory level. Built on O*NET and public employment data, open-access journal, make-driven. One of the most reproducible-looking economics packages I opened.
### C28 — Pre-Colonial Warfare and Long-Run Development in India
*economics (economic history/political economy) · Stata · verified_complete: **yes***
https://zenodo.org/records/5583263
Dincecco, Fenske, Menon and Mukherjee, *Economic Journal*. 33 MB, 64 entries: 18 .dta files shipped, a complete ESRI shapefile set (.shp/.dbf/.prj/.shx/.sbn/.cpg), 11 .tex table outputs and README.pdf. Historical and geographic data with no living-subject restrictions, so the whole chain is in the box.
### C29 — The Unequal Effects of Pollution on Labor Supply
*economics (environmental/labour) · Stata + Quarto · verified_complete: **yes***
https://zenodo.org/records/11060907
Hoffmann and Rud, *Econometrica* 2024. 5.2 GB, 1272 entries read via the zip's ZIP64 central directory: `master.do`, 12 do-files, 58 .dta, 8 .qmd Quarto documents, shapefiles for the Mexico City maps, README.pdf. The deposit's contents list — README, raw data, do-files, map folders, output folders — matches what I saw. Large but genuinely complete; budget the disk.
### C30 — Market Participation and Moral Decision-Making: Experimental Evidence from Greenland
*economics (behavioural/development) · Stata · verified_complete: **no***
https://zenodo.org/records/6974389
Agneman and Chevrot-Bianco, *Economic Journal* 2022. 3.4 MB, 111 entries: `00_MAIN.do` plus 28 do-files, 3 .ado, and full field documentation (experimental design, instructions in three languages, ethics approval). But the manifest holds no .dta or .csv — the outputs are 35 .gph graphs and 16 .tex tables. The survey microdata is not in the package, so marked `no`.
### C31 — Lifetime and Intergenerational Consequences of Poor Childhood Health
*economics (health/economic history) · Stata · verified_complete: **could-not-check***
https://zenodo.org/records/7117307
Karbownik and Wray, *Journal of Human Resources*. 26 MB, 605 entries, `0_run_all.do`, a proper README.md with an "Overview" section (I read the opening of it), 50 .eps and 32 .tex outputs to check against. Only 2 .dta files though, against linked historical census/vital records — I could not tell from the manifest whether those two are the analysis samples or lookup tables.
### C32 — Entrepreneurial Human Capital and Firm Dynamics
*economics (firm dynamics) · Python + Stata · verified_complete: **yes***
https://zenodo.org/records/5464854
Queiró, 2021. 15 MB, 111 entries: `run_all.ipynb` and `main.ipynb`, an `environment.yml` pinning the Python environment, 19 do-files, README.pdf, and 45 .png / 20 .tex outputs. The deposit explicitly says it "contains publicly available data and code". A pinned conda environment plus a single notebook driver is about as good as this genre gets.
### C33 — Multinational Enforcement of Labor Law: Experimental Evidence on OSH Committees
*economics (labour/development) · R + Stata · verified_complete: **yes***
https://zenodo.org/records/11373971
*Econometrica*. 6.3 MB, 488 entries: `0.master.R` and `3.master.do`, 34 R scripts, 21 do-files, 2 .RData, "00.README.pdf" and the IRB approval letter. The directory names in the archive (`raw_data_R`, `input_data_stata`, `cleaning_code_R`, `analysis_R`, `final_paper_replication`) spell out the pipeline. A factory-level RCT with its own collected data shipped.
### C34 — Not Just a Work Permit: EU Citizenship and Consumption Behaviour of Immigrants
*economics (migration) · Stata · verified_complete: **no***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/NZNQBR
Adamopoulou and Kaya, *Canadian Journal of Economics*. Opened the Borealis record and listed all 9 files: 6 do-files (`INPS_data_preparation.do`, `ISMU_tables.do`, …), README.txt, and two .tab files that are `coefficient_estimates` and `demo_istat`. At 103 KB total there is no microdata — the Italian INPS social-security and ISMU immigrant survey inputs are not distributable. Code and README are open; data is not.
### C35 — Sector and Importer Determinants of Prices for Traded Intermediates
*economics (trade) · Stata · verified_complete: **yes***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/G55JDI
Bekkers, Francois and Manchin, *Canadian Journal of Economics*. Five files, all unrestricted: `finaldata.dta` (the bulk of a 1.2 GB deposit), `finaldata_onlyhomogeneous.tab`, `reproducing_tables.do`, `Figure1.xlsx`, and `Readme.pdf`. One do-file, one dataset, one readme — unusually clean, and a good candidate precisely because there is nowhere for ambiguity to hide.
### C36 — Trade without "Scale Effects"
*economics (trade) · Stata · verified_complete: **yes***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/2HUCIH
Bento, *Canadian Journal of Economics*. Nine files, 2.5 MB, none restricted: five .tab data files (US Business Register size distributions, `dataRose.tab`, `ag_empshare.tab`), three figure-specific do-files and ReadMe.pdf. Small enough to reproduce in an afternoon on public data.
### C37 — A Car for Recognition: Heterogeneous Market Responses to Electric Vehicles
*economics (industrial organization) · Stata · verified_complete: **yes***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/WZVH0A
Liu, Xu, Zhang and Zheng, *Canadian Journal of Economics* 2026. Nine files, 573 MB, none restricted: `master.do`, city-year-month and province-year .tab panels, an `adofiles.zip` bundling the required Stata packages (a detail that materially improves reproducibility), a `Results.zip` of expected output, ReadMe.pdf and an explicit LICENSE.txt.
### C38 — Managed Trade: The US-Mexico Sugar Suspension Agreements
*economics (trade/agricultural) · Stata · verified_complete: **yes***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/P2J35S
Carter, Saitone and Schaefer, *Canadian Journal of Economics*. Ten files, 460 KB: `price_data.tab` and `quantity_data.tab`, three analysis do-files including a bootstrap script, table/figure .tab outputs to check against, and README.pdf. Tiny, public commodity price data, fully shipped.
### C39 — Price Caps, Commitment and Innovation
*economics (regulation/theory) · Python · verified_complete: **yes***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/2VWMN4
Atallah and Simons. Four files totalling 7.5 KB: `replicate_linear_case.py`, `requirements.txt`, README.pdf, LICENSE.txt. The record states the tables "rely exclusively on equilibrium computations using the linear model described in Appendix A. No external data" is needed. A pure-computation package with pinned dependencies — the cheapest possible verification, and a useful sanity check on the pipeline itself.
### C40 — The Role of Institutions and Immigrant Networks in Firms' Offshoring Decisions
*economics (trade/migration) · Stata · verified_complete: **no***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/VBGZEP
Moriconi, Peri and Pozzoli, *Canadian Journal of Economics*. Ten files, 553 KB, eight do-files and README.docx — and, decisively, a file literally named `data_access_instructions.pdf`. The Danish employer-employee registers are accessed through Statistics Denmark, not shipped. Marked `no`; listed because the code is fully open and readable even when the data is not.
### C41 — Difference-in-Differences Designs: A Practitioner's Guide
*econometrics / political methodology · R + Stata · verified_complete: **yes***
https://github.com/pedrohcgs/JEL-DiD
Baker, Callaway, Cunningham, Goodman-Bacon and Sant'Anna, *Journal of Economic Literature*. I opened the repo: `data/`, `scripts/`, `figures/`, `tables/`, `renv.lock`, a GitHub Actions workflow, README.md and README.pdf. All input data is public domain (CDC, BLS, Census, KFF) merged into `county_mortality_data.csv`, and the README gives two parallel entry points — `scripts/R/00_master_did_jel.R` (~10 min) and `scripts/Stata/00_stata_master_did_jel.do` (~11-15 min) — that are supposed to produce identical figures and tables. Public data, pinned R environment, dual implementation, ten-minute runtime, heavily cited. If the board only reproduces one thing, this is my pick.
### C42 — The Green Transition and Political Polarization Along Occupational Lines
*political science · R + Stata · verified_complete: **no***
https://github.com/vincentheddesheimer/green_transition_replication
Heddesheimer, Hilbig and Voeten, *American Political Science Review* 2025. Opened the repo: `code/`, `data/`, `codebooks/`, `results/`, README.md. The README is unusually good — three run paths (`source("code/master_full_replication.R")`, step-by-step, or individual scripts), a stated ~30-minute runtime, and explicit requirements (R 4.4.0+, 8 GB RAM, ~5 GB disk). Marked `no` only because the German SOEP survey, GLES candidate data and party press releases need separate institutional access; the county-level election and economic data are public and the final analysis datasets ship.
### C43 — Engaging Populism? European Populist Parties on Facebook and Twitter, 2010-2020
*political science · R (brms/Stan) · verified_complete: **no***
https://github.com/t-davidson/engaging-populism-replication
Davidson and Enos. Opened the repo: `data-collection/`, `cleaning-and-merging/`, `models/scripts/`, `results/`, README.md, which lays out the intended order clearly. Marked `no`: the raw social media data is not provided (only example API query scripts), and the larger fitted `brms` models live behind a Google Drive link rather than in the repo. Model estimation also assumed a SLURM cluster.
### C44 — How Segregated Is Urban Consumption?
*economics (urban) · Julia + Stata + Bash · verified_complete: **yes***
https://github.com/jdingel/DavisDingelMonrasMorales
Davis, Dingel, Monras and Morales, *Journal of Political Economy*. Opened the repo: a top-level `Makefile`, `readme.md` and `readme.pdf`, `local_configuration.sh` / `slurm_configuration.sh`, `tasks_flow_graph.png`, `initialdata/`, and 46 task directories. The README documents `make` / `make full_version` versus `make quick_version`. Flagging the cost honestly: the quick version wants ~70 GB of disk and under a day; the full version wants ~4 TB and several days. Complete and make-driven, but not a light reproduction.
### C45 — A/B Testing with Fat Tails
*economics (experiments) · R + MATLAB + Bash · verified_complete: **no***
https://github.com/eduardomazevedo/admrw-ab-replication-code
Azevedo, Deng, Montiel-Olea, Rao and Weyl, *Journal of Political Economy*. Opened the repo: `bash/`, `r/`, `matlab/`, `data/`, `output/`, `codebook.txt`, `output-list.txt`, `readme.txt`, with `bash/make.sh` as the single driver. The readme says plainly that "the data used is proprietary" and directs users to Microsoft. Clean pipeline, unobtainable data — listed as a documented negative.
### C46 — A Distributional Framework for Matched Employer-Employee Data (BLM)
*economics (labour) · R (Docker) · verified_complete: **no***
https://github.com/tlamadon/blm-replicate
Bonhomme, Lamadon and Manresa, *Econometrica*. Opened the repo: `R/`, `inst/`, `packrat/`, `DockerConfig/`, a `Dockerfile`, DESCRIPTION/NAMESPACE, README.md/.html/.pdf. Two documented routes — a prebuilt Docker image exposing RStudio, or packrat restore plus `source("inst/main.R")`. Marked `no` because the Swedish IFAU administrative data requires an application and confidentiality agreement; the package runs on synthetic data by default. Still worth listing: the Docker + packrat setup is a model of environment capture even where data cannot follow.
### C47 — More Parties or No Parties: The Politics of Electoral Reform in America
*political science · R · verified_complete: **could-not-check***
https://github.com/jacksantucci/mponp
Santucci (Oxford UP, 2022). Opened the repo: per-chapter directories `ch1_files` … `ch7_files_new`, an `img/` folder, and README.md. Historical STV referendum, roll-call and municipal election data is included, which is the hard part. But the README gives no run instructions and describes the repo as "a work-in-progress", so I cannot say the third leg is there.
### C48 — Detection and FDR Control of Differentially Methylated Regions from WGBS (dmrseq)
*bioinformatics (epigenomics) · R + Bash · verified_complete: **yes***
https://github.com/kdkorthauer/dmrseqPaper
Korthauer et al. Opened the repo: `PREPROCESSING/`, `ANALYSIS/`, `BENCHMARK/`, `FIGURES/CODE/`, README.md. The README specifies the order (preprocessing → analysis → figures) and pins the analysis to a specific `dmrseq` commit. Crucially, every dataset is public — ROADMAP human tissue, DNMT3A murine leukemia and dendritic cell data pulled from SRA, GEO and NCBI FTP by the preprocessing scripts. Scripts assume SLURM but state they can run serially. Genuinely open end-to-end; the cost is sequencing-scale compute, not access.
### C49 — Single-Cell DNA Replication Dynamics from DLP Sequencing
*bioinformatics (single-cell genomics) · Python + Snakemake · verified_complete: **yes***
https://github.com/shahcompbio/scdna_replication_paper
Shah lab, *Nature Communications* 2024. Opened the repo: `Snakefile`, `config.yaml`, `cluster.json`, `rules/`, `scripts/`, `notebooks/`, `envs/`, `data/`, MIT LICENSE, README.md. The README states that all source data needed to reproduce the figures, plus file-structure documentation, is in a companion Zenodo deposit (10.5281/zenodo.12786373) — public, just not in the git repo. Conda envs are checked in and the workflow is Snakemake-orchestrated. Users must adjust paths, which is exactly the kind of friction a verification board should record.
### C50 — Multiscale Topology Classifies Cells in Subcellular Spatial Transcriptomics (TopACT)
*bioinformatics (spatial transcriptomics) · Python · verified_complete: **yes***
https://zenodo.org/records/11050996
128 MB, 61 entries read from the zip: 19 Jupyter notebooks, 8 .csv and 5 .npy data arrays, and a README.md that I read out of the archive — it names the Python version (3.10+) and gives the pip install line for the companion TopACT package. Data, code and instructions all present; the one external dependency is a public GitLab package.
### C51 — Parallel Reverse Genetic Screening in Mutant Human Cells Using Transcriptomics
*bioinformatics (transcriptomics) · R · verified_complete: **yes***
https://zenodo.org/records/51842
661 MB, 1941 entries: 152 gzipped expression/annotation files covering, per the record, "over 1800 RNA-seq samples", 22 R scripts and 22 .csv, plus a README. The deposit is a single self-contained data-and-analysis bundle rather than a code-only archive, which is what makes it worth the disk.
### C52 — A Disease Network-Based Deep Learning Approach for Characterizing Melanoma
*bioinformatics (cancer/network biology) · R + Python · verified_complete: **yes***
https://zenodo.org/records/10788609
Lai et al., *International Journal of Cancer* 2022. 257 MB, 280 entries: 133 .csv node/edge tables, 20 .RData, 5 R scripts, a Python script, 2 pickles and `readme.pdf`. The deposit says it "contains the data, code, and analysis to recreate the results". Derived network data rather than patient-level data, so nothing is gated.
### C53 — Host Genetic Regulation of Human Gut Microbial Structural Variation (SV-GWAS)
*bioinformatics (microbiome/GWAS) · R + Python + Bash · verified_complete: **no***
https://zenodo.org/records/10018199
Groningen Microbiome Centre, *Nature* 2023. 71 KB, 51 entries: 12 R, 12 shell, 9 Python, 7 .Rmd and a README.md that I read from the archive — it explains the two-folder structure (microbiome SV calling, then GWAS). Marked `no` for the obvious reason: 71 KB is code only, and the underlying cohort genotypes and metagenomes are access-controlled. Good code to read, not a reproduction target.
### C54 — A Single-Cell and Spatially Resolved Atlas of Human Breast Cancers
*bioinformatics (single-cell/cancer) · R + Snakemake · verified_complete: **no***
https://zenodo.org/records/5031502
Wu et al., *Nature Genetics* 2021 — Zenodo archive of `Swarbricklab-code/BrCa_cell_atlas`. 360 KB, 123 entries: 59 R scripts, 12 shell scripts, two Snakefiles (CIBERSORTx, DWLS) and a README.md whose "Data Availability" section I read. The processed scRNA-seq is downloadable from the Broad Single Cell Portal rather than shipped, and at 360 KB nothing is in the box. Marked `no`; the data is public but the package is not self-contained.
### C55 — Cancer-Associated Fibroblast Classification in Single-Cell and Spatial Proteomics
*bioinformatics (proteomics) · R · verified_complete: **no***
https://zenodo.org/records/7540622
Bodenmiller group. 94 KB, 27 entries: 20 .Rmd analysis notebooks and a README.md. I read the README from the archive — it is a single line, the repository name. No data, no instructions. Listed as a checked negative so nobody spends time rediscovering it.
### C56 — From Telomere to Telomere: Transcriptional and Epigenetic State of Human Repeat Elements
*bioinformatics (genome assembly) · Perl + R + Bash · verified_complete: **could-not-check***
https://zenodo.org/records/5895031
T2T Consortium (Hoyt et al.), for the CHM13 reference. 507 MB, 220 entries: 116 gzipped data files, 8 README.md files at directory level, 4 Perl and 2 R scripts. I read the top README, which describes the organisation ("organized by type of analysis conducted") but is not a run procedure. Data is plainly present; whether the instructions are sufficient I cannot say from the manifest.
### C57 — Impacts of Renewable Energy Facilities on Animal Movement
*ecology (systematic review) · R · verified_complete: **yes***
https://zenodo.org/records/20311943
Arrondo et al. 2026. 236 KB, 23 entries: `01_analysis_main.R` plus 9 more R scripts, 3 .csv, `CITATION.cff`, and a README.md I read from the archive — it carries MIT (code) / CC-BY-4.0 (data) badges and states which outputs the pipeline reproduces (the facility × movement-component matrix, the bivariate world map, waffle plots, Supplementary Tables S1-S2). A review dataset of 134 studies, so no field data access is needed. Small, licensed, explicitly scoped.
### C58 — Understanding Temporal Variability Across Trophic Levels in Freshwater Ecosystems
*ecology (community ecology) · R · verified_complete: **yes***
https://zenodo.org/records/8333128
Siqueira et al. 4.5 MB, 129 entries: 30 .csv, 14 R scripts and two .Rproj files. The deposit description itself is the run instruction and is unusually specific — it names the five scripts in order (`01_Dataprep_stability_metrics.R`, `02_SEM_analyses.R`, `03_Stab_figs.R`, `04_Stab_supp_m.R`, `05_Sensit_analysis.R`) and says the full result set including supplementary material comes out of running them in that sequence.
### C59 — Thresholds for Ecological Responses to Global Change Do Not Emerge from Empirical Data
*ecology (global change) · R · verified_complete: **yes***
https://zenodo.org/records/3828869
Hillebrand, Donohue, Harpole et al. 267 KB, 53 entries: 37 .csv and 13 R scripts with a README.md that I read out of the archive. Small, self-contained, and a genuinely contested claim — a high-profile negative result about ecological thresholds is exactly the kind of thing worth independently re-running.
### C60 — Global Conservation of Species' Niches
*ecology (conservation) · R (make) · verified_complete: **no***
https://zenodo.org/records/1035485
Hanson et al. 2020. 1.3 GB, 671 entries: a top-level `Makefile`, 47 README.md files at directory level, 11 .rda, rasters and shapefiles. Marked `no` on the deposit's own statement: "many of the raw data files (e.g. extent of suitable habitat maps, protected area data) are not available in this archive, and must be obtained from the original sources". The make-based structure is excellent; the inputs are not all here.
### C61 — Mapping the Forest Disturbance Regimes of Europe
*ecology (remote sensing) · R · verified_complete: **could-not-check***
https://zenodo.org/records/3925447
Senf and Seidl, *Nature Sustainability* 2020. 63 MB, 80 entries: 38 .csv, four complete shapefile sets, 4 R scripts, README.md/README.txt (I read the top README from the archive). The catch is stated in that README — the disturbance maps themselves are hosted externally and the repository is a static snapshot of a project the authors say they "continuously improve and update". Whether the archived csvs suffice without re-downloading the maps, I could not determine.
### C62 — Body Size, Reef Area and Temperature Predict Global Reef-Fish Species Richness
*ecology (macroecology) · R · verified_complete: **yes***
https://zenodo.org/records/1484591
Barneche et al., *Global Ecology & Biogeography*, archived from `dbarneche/geb12851`. 4.9 MB, only 19 entries: 7 .csv, 3 R scripts, 2 .yml (a `remake`-style build spec) and README.md, which I read from the archive and which opens with "This repository contains code and data needed to reproduce the article". Nineteen files, a declarative build file, public reef survey data. Very high effort-to-value ratio.
### C63 — An Updated End-to-End Ecosystem Model of the Northern California Current
*ecology (ecosystem modelling) · MATLAB + R · verified_complete: **yes***
https://zenodo.org/records/8393490
*PLOS ONE*. 4.9 GB, 1712 entries: 90 .csv, 60 .m, 9 R scripts, plus README_FIRST.txt, a top-level README.txt and per-subdirectory READMEs — I read the top one, which explicitly tells you that each subdirectory carries its own README and which directory (`ECOTRAN_Code`) you need to run the model elsewhere. Large but layered and documented; the ECOTRAN model is reusable beyond this paper.
### C64 — Macroscale Controls on Recovery of River Ecosystem Productivity After Floods
*ecology (freshwater/Bayesian) · R + Stan · verified_complete: **yes***
https://zenodo.org/records/10455823
Lowman et al. 2024. 90 MB, 26 entries: 18 R scripts, 5 .rds fitted objects, 2 `.stan` model files and `README_for_repository`, whose opening I read — it names the creators and their institutions and points to the publication for the analysis description. Stan models shipped alongside cached fits means you can verify both the refit and the downstream figures.
### C65 — Bayesian Analysis of Climate Change Adaptation Willingness of Canadian Private Forest Owners
*ecology / environmental social science · R (Bayesian) · verified_complete: **yes***
https://zenodo.org/records/16369904
4.1 GB, 61 entries: 10 R scripts, 23 .rds posterior objects, 11 .csv, a shell driver, `readme.pdf` and a README.md I read from the archive — it describes a fully Bayesian analysis of an original survey built on protection motivation theory. Survey data plus posteriors plus code plus documentation. The 4 GB is almost entirely cached MCMC draws, so the reproduction can be staged.
### C66 — Behavioral Plasticity Shapes Participation in a Mixed-Species Flocking Community of Birds
*ecology (behavioural) · R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.0k6djhb7h
Dryad, 134 KB, 4 files: `AustraliaFlockData2019Dryad.csv`, `Foraging_observations__2019_Dryad.csv`, `Species_code_information.xlsx` and README.md, which I read in full on the landing page — it sets out the four axes of behavioural plasticity being tested. Dryad curation requires the README, and the record links a preprint DOI and a software DOI. Trivially small, fully open.
### C67 — Computational Analyses of Dynamic Visual Courtship Display in Wolf Spiders
*ecology (animal behaviour) · Python + R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.mgqnk9989
*Animal Behaviour* 2025 (primary article DOI linked from the record). Dryad, 9.1 MB: `rabidosa_signal_comparison.csv`, `total_raw.csv` and a README.md I read in full — it enumerates four JupyterLab notebooks (`0_train_GHMM.ipynb`, `1_segmentation.ipynb`, …), one R script, and lists the exact Python packages each notebook needs. That level of dependency detail in a README is rare and makes this an easy verification.
### C68 — Beyond Thermoregulation: Behavioural Regulation of UV Exposure by a Free-Ranging Lizard
*ecology (physiological ecology) · R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.mpg4f4rgq
Lattanzio, *Journal of Animal Ecology* 2026. Dryad, 1.3 MB, 4 files: `dev_resampler.R`, `spr_resampler.R`, `Lattanzio_2026_RawData.xlsx` and README.md, read in full — it names each file and says the workbook contains the raw and processed data used for analyses, figures and supporting information. Raw data plus resampling helpers plus documentation.
### C69 — Active, Explorative and Infected: Meta-Analysis of Behaviour and Parasitism
*ecology (meta-analysis) · R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.k0p2ngfq0
Dryad, 120 KB, exactly three files: `data.csv`, `Meta-analysis.R`, `README.md` — README read in full, describing a phylogenetic multilevel meta-analysis with overall models, parasite-group models, publication-bias tests and sensitivity analyses. Three files, 120 KB, phylogenetic meta-analysis. Probably the single cheapest verification on this list.
### C70 — Parturition Timing and Pre-/Post-Partum Behaviour of Female Moose
*ecology (wildlife/movement) · R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.cjsxksnmh
Dryad, 1.2 MB, 5 files including one R script implementing six movement-based calving-date methods (BCPA, FPT, MRT, rMCP, net-squared displacement), `ParturitionDates.csv`, `Video-based_observations.csv`, an xlsx of NSD values and README.md read in full. The paper benchmarks movement-based inference against animal-borne video ground truth — a method comparison, which makes re-running it more informative than most.
### C71 — Ocean Deoxygenation and Warming Disrupt Cooperation in Coral Reef Fish Mutualisms
*ecology (marine/behavioural) · R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.3ffbg79zm
Pereira et al., *Behavioral Ecology* 2025 (primary article DOI on the record). Dryad, 21 KB — twenty-one kilobytes — for `data_FAP.csv`, `data_Interactions.csv`, `script.Rmd` and a README.md I read in full, with citation, originator and collection-date sections. One Rmd, two csvs, a real journal article behind it.
### C72 — The Influence of Predator Lethality on Prey Behaviour and Growth
*ecology (experimental marine) · R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.6djh9w1hj
Dryad, 749 KB, 8 files: six .csv covering foraging, growth, lethality and three behaviour datasets, `nogueira_2026_lethality_code.Rmd`, and README.md read in full — it describes the two-phase mesocosm design with crab predators and snail prey. A controlled experiment with all arms shipped.
### C73 — Why Wolves Walk: Drivers of Daily Movement in a Social Large Carnivore
*ecology (movement ecology) · R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.280gb5n39
Dryad, 304 KB, 5 files: seasonal summer/winter datasets, `DMD_D_analysis.R`, `Linearity_D_analysis.R`, README.md read in full — it documents every column and explains why some covariates (bear density, breeding status; moose and roe deer density) appear in only one season. Column-level documentation is exactly what makes an independent re-run possible.
### C74 — Forest Degradation Reduces Acoustic Partitioning in Tropical Avian Communities
*ecology (ecoacoustics) · R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.pvmcvdp1q
*Journal of Avian Biology* (primary article DOI on the record). Dryad, 2.5 MB, 5 files: `ANH_code_upload.R`, ecoacoustic sample results, `selections6.csv`, `transect_gradients.csv` and README.md read in full — it even warns which column belongs to a different companion study and is dropped during analysis. That kind of caveat is a good sign.
### C75 — Climate Extremes, Species Richness and Temporal Variability of Bird Communities
*ecology (community stability) · R · verified_complete: **no***
https://datadryad.org/dataset/doi:10.5061/dryad.v6wwpzh1g
Cady et al., *Ecology*. Dryad, 89 KB, 3 files: `R_Code_Cady_etAl_Community_Stability2024.R`, `SpeciesList.csv`, README.md. Marked `no` on the README's own first line about data: "User must download 5 open-source datasets and 1 dataset from Cady et al. (2023) on Dryad before running". The inputs are open, but they are not in this package — reconstruction required.
### C76 — Data and Code for Wound Care in Camponotus floridanus Ants
*ecology (behavioural/microbiology) · R · verified_complete: **could-not-check***
https://datadryad.org/dataset/doi:10.5061/dryad.1rn8pk110
Dryad, 93 KB, 6 files: five .csv covering tracking, qPCR and amputation experiments, plus a README.md I read in full which documents every column and is headed "Raw Excel Data and Code". But no script file appears in the file listing — the analysis code the title implies is not visibly there. Flagged rather than assumed either way.
### C77 — Tree Canopy Halves Urban Heat Island Effect Globally
*ecology (urban climate) · unknown (in archive) · verified_complete: **could-not-check***
https://datadryad.org/dataset/doi:10.5061/dryad.905qfttz0
McDonald et al., *Nature Communications* 2026. Dryad, 1.4 GB, two files: `DataDryadArchive_McDonald_et_al_final.zip` and README.md, which I read in full — it states that all code and data used in the paper is here. I could not list the zip's interior: unlike Zenodo, Dryad's file endpoints require an API bearer token, so the range-request trick I used elsewhere does not work. High-profile result, plausibly complete, contents unverified.
### C78 — Rethinking Global Hotspots for Threatened Terrestrial Vertebrates
*ecology (conservation biogeography) · R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.m905qfv8g
*Global Ecology & Biogeography*. Dryad, 263 MB, 4 files: `R_scripts_and_data.zip`, `HTV.zip`, a species list xlsx and README.md read in full — it names each script inside the archive (`range-diversity analysis.R`, `hotspots analysis.R`, `biome_analysis.R`, `PAs analysis.R`) and what each does. Script-level documentation from outside the archive is what lets me call this complete.
### C79 — An Evolutionary Framework for Characterizing Noncoding Regions of DNA
*evolutionary genomics / statistics · mixed (see archive) · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.w3r22816w
Dryad, 163 KB, 8 files, and the structure is the reason it is here: `SOFTWARE.zip`, `DATA.zip`, `DOCUMENTATION.zip`, `VALIDATION.zip`, `MANIFEST.txt`, `SHA256SUMS.txt`, README.md and README.txt. The README (read in full) describes the omega_RNA / rho_comp statistics for RNA secondary-structure evolution and gives a package build date. A shipped validation suite plus checksums is close to best-practice for a methods deposit.
### C80 — Evaluating Baselines for Long-Term Biodiversity Monitoring: NEON Carabid Data
*ecology (monitoring/biodiversity) · R · verified_complete: **could-not-check***
https://datadryad.org/dataset/doi:10.5061/dryad.0k6djhbfn
Claus et al. Dryad, 55 MB, two files: `Claus_et_al_2026.zip` and README.md, read in full — it explains the project analyses NEON carabid observational data to determine sampling effort needed for richness and diversity estimates. NEON data is open, which is promising, but the archive interior is not listable without a Dryad token.
### C81 — Shifting Speciation Mode and Biogeography in Late Ordovician Laurentian Brachiopods
*paleobiology / phylogenetics · R (BioGeoBEARS) · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.d51c5b0fs
*Palaeogeography, Palaeoclimatology, Palaeoecology* 2025. Dryad, 52 KB, 5 files: `BioGeoBEARS_code.R`, `Anazygidae_tree.tre`, `Anazygidae_geoareas.data`, `b_AICc_table.csv` and README.md read in full — it even records the R version used (4.3.0) and explains the geoareas file format. Tiny, complete, and a good test of whether BioGeoBEARS model selection is stable across package versions.
### C82 — Phylogeny of Pacific Starlings (Aplonis) Reveals Cryptic Diversity
*evolutionary biology (phylogenomics) · mixed (Bash/R/Python) · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.mkkwh71dd
Dryad, 283 MB, 8 files, organised as one zip per pipeline stage — `01_read_processing.zip` through `07_biogeography.zip` — plus a README.md I read in full that walks through each stage and marks which files are Dryad-only (large alignments) versus mirrored on GitHub. Raw reads live at NCBI SRA (PRJNA1336315), but processed alignments ship, so the downstream phylogenetics is reproducible without re-assembling reads.
### C83 — Bee Swimming Is Adaptive but Disrupted by Insecticide
*ecology (entomology/toxicology) · R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.w9ghx3g2r
*Communications Biology* 2026 (primary article DOI on the record). Dryad, 30 MB, 25 files: `Bee_swimming_paper_analysis.Rmd`, 18 figure-specific .csv files named `Fig1A.data.csv` onward, five .mov videos and README.md read in full — it documents each csv down to what the measured angle means and when the experiment was run. Per-figure data files make partial verification easy.
### C84 — Sheltered or Suppressed? Tree Regeneration in Unmanaged European Forests
*ecology (forest dynamics) · R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.hx3ffbgkx
Dryad, 2.3 MB, 5 files: `data_corrected.zip`, `scripts_corrected.zip`, `make-public-data.R`, README.md and — notably — `CHANGELOG_correction.md`. A deposit that versions its own corrections is a good verification subject: the changelog gives you a documented before/after to check against.
### C85 — Diversity Regulation in Local Communities: Colonization and Extinction in North American Birds
*ecology (macroecology) · R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.5dv41nshz
Bohdalkova and Storch, *Global Ecology & Biogeography* (primary article DOI 10.1111/geb.70270 on the record). Dryad, 2.4 MB, 17 files: ten numbered R scripts (`1_loading_subsampling_data.R` through `10_corespecies_analyses.R`) and six .RData processed datasets. I read the full README on the landing page: raw Breeding Bird Survey data must be downloaded separately for script 1, but "all subsequent scripts use the processed files provided here". Scripts 2-10 are reproducible as shipped, which is an honest and usable arrangement.
### C86 — Residence Time and Functional Traits Shape Conspecific Plant-Soil Feedback
*ecology (plant-soil interactions) · R · verified_complete: **yes***
https://datadryad.org/dataset/doi:10.5061/dryad.z34tmpgwn
Dryad records the related publication ISSN as 0022-0477 (*Journal of Ecology*); the record itself lists no primary-article DOI, so treat the journal attribution as the ISSN's, not a checked citation. 392 KB, 8 files: `NovNat_WP3_PSF_Analysis.R` (a single 197 KB script) plus six .xlsx trait and feedback datasets and a ~10 KB README.md. Small, curated, single-script analysis with all inputs present.
### C87 — Zoned In or Zoned Out? Immersion in Slot Machine Gambling via Mobile Eye Tracking
*psychology (addiction/attention) · R · verified_complete: **yes***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP2/GSOMLU
Nine files, 876 KB, none restricted: anonymised experiment and pilot data plus event-level data, four R scripts (`GazeModel_Bootstrapping.R` and three area-of-interest models for credits, reels and win displays), `DataDictionary.txt` and README.txt. Anonymised eye-tracking data that is actually distributable is unusual in this subfield.
### C88 — Measuring Lexical Distance Between Parallel Corpora: AI-Generated News Translation
*computational linguistics / methods · R · verified_complete: **yes***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/MFZTWZ
Conway et al. Nine files, 9.9 MB, none restricted: `1_conway_et_al_measurement_code.Rmd` with both .docx and .pdf renderings of the same code, English and French lemmatization tables, substitution lists, `8_conway_et_al_corpus_data.RData`, and `0_read_me.txt`. Numbered filenames give the run order without needing to guess.
### C89 — Everyday Emotion-Goal Pursuit Associations in Older Adults
*psychology (lifespan/ambulatory assessment) · R · verified_complete: **no***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/LLQGA4
Choi, Zambrano Garza, Pauly et al. 2025. Seven files, 2.5 MB: `Dataverse_analytic code.Rmd`, `Code for figures.R`, README.txt, and — the problem — the only data present is `f1_plot_data.tab`/`f1_summary.RData`/`f2_plot_data.tab`/`f2_summary.RData`, i.e. figure-level summaries. The participant-level daily-diary data behind the models is not deposited. Marked `no`.
### C90 — Species Interactions Limit the Predictability of Community Responses to Environmental Change
*ecology (experimental microcosms) · R · verified_complete: **could-not-check***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP2/SJUDDO
Nine files, 148 KB, none restricted: six numbered R scripts running from `1_analyze_protist_videos_Patrick_SH.R` to `6_protist_analysis.R`, two .csv, and a Dryad cross-reference JSON. Data and code are both here and the numbering implies the order, but there is no README, so I will not claim run instructions are present.
### C91 — Food Web Structure Across Basins in Lake Erie
*ecology (freshwater food webs) · R · verified_complete: **could-not-check***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/6ZGCDT
*Canadian Journal of Fisheries and Aquatic Sciences*. Four files, 239 KB, none restricted: `CJFAS_L-Erie-Foodweb-analysis.Rmd` and three .tab datasets (fish data, SIBER inputs). An Rmd is close to self-documenting, but there is no README and no stated package requirements, so this is a `could-not-check` rather than a `yes`.
### C92 — Using Otolith Chemistry to Characterize the Diet of Redfish (Sebastes sp.)
*ecology (fisheries) · R · verified_complete: **yes***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/C7PW8A
Eight files, 1.9 MB, none restricted: three numbered .Rmd analyses (feeding, growth, otolith chemistry), an `.Rproj`, three .tab datasets and `00.README_Redfish_experiment.txt`. A controlled six-month feeding experiment at ISMER/Maurice-Lamontagne with all arms deposited.
### C93 — Overstocking and Temporal Feed Restriction Affect Dairy Cows' Drinking Behavior
*animal science / behavioural ecology · R · verified_complete: **yes***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/JLDGQG
Eight files, 2.9 MB, none restricted: `all_code_R_Markdown.Rmd` with its knitted `.html` (so you have the author's own output to diff against), four .tab datasets covering feeding and water behaviours and the Latin-square design, a raw-data example, and `ReadMe.txt`. Shipping the knitted HTML alongside the Rmd is the detail that makes this an easy verification.
### C94 — Habitat Use and Local Movement of Staging Lesser Yellowlegs
*ecology (ornithology) · R · verified_complete: **yes***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/KKWVBG
Danyk et al., *Journal of Field Ornithology* 2025. Eight files, 72 KB, none restricted, in a real directory structure: `Data/01_Local_movement/` and `Data/02_Behaviour_and_habitat_use/` .tab files, `Danyk_et_al_2025_statistical_analysis.Rmd`, an `.Rproj`, and `README_Danyk_et_al_2025.txt`. 72 KB total.
### C95 — Prediction Models for Post-Discharge Mortality Among Under-Five Children in Uganda
*clinical epidemiology / ML · R · verified_complete: **no***
https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/M3OPKQ
Nine files including six R scripts (data manipulation, final models for 0-6m and 6-60m cohorts, learning curves, z-scores), two post-processed .rds and a data dictionary. All nine files are flagged restricted in the Dataverse record — I could see the manifest but nothing is downloadable without approval. Listed as a clean negative: good code structure, gated paediatric data.
### C96 — Does Guilt Motivate Prosocial Behavior at the Expense of Others?
*psychology (moral/social) · R (Stan) · verified_complete: **yes***
https://osf.io/f2xsz/
Preregistered replications with exploratory statistical analyses. I walked the OSF storage tree: 35 nodes across `material/`, `data/`, `readme/` and `supplementary materials/`, with per-experiment readmes (`readme_experiment1.txt`, `readme_experiment2.txt`), participant-level `data_exp1_all.csv` and `data_exp2_all.csv`, variable-definition PDFs, six .Rmd analysis files and five `.stan` models, and a separate satisficer-exclusion branch under supplementary materials. Preregistered plus Bayesian plus full data is a strong combination.
### C97 — How Effective Is Moral Framing in Promoting Pro-Environmentalism?
*psychology (environmental/social) · R · verified_complete: **yes***
https://osf.io/2dqu6/
A preregistered replication of Wolsko et al. (2016), run as a Registered Report (the accepted stage-1 manuscript is in the project). 17 nodes split into `Direct Replication (DR)/` and `Conceptual Replication (CR)/`, each with .xlsx data and separate preprocessing and analysis Rmd files — including both the *planned* scripts (`DR-script-planned-data-analysis.Rmd`) and the executed ones, plus a codebook and materials PDF. Having planned and actual scripts side by side is unusually good for verification.
### C98 — Testing the Interleaving Effect Without Response Bias
*psychology (learning/memory) · R · verified_complete: **yes***
https://osf.io/u436s/
A preregistered replication of Kornell & Bjork (2008). Eight nodes: `data/IdentificationTask.csv` and `data/2AFCTask.csv`, `scripts/Identification.Rmd` and `scripts/AFC.Rmd`, the timestamped preregistration PDF, and a supplementary-materials PDF. Two tasks, two datasets, two Rmds, 4.9 MB.
### C99 — Preregistered Replication of "Sick Body, Vigilant Mind"
*psychology (behavioural immune system) · R · verified_complete: **yes***
https://osf.io/k2dbf/
13 nodes: `rawdata.zip`, `SBVM_data_aggregated.csv`, two generations of analysis notebooks (`SBVM_stage2.Rmd` and `SBVM_stage2rev.Rmd`) each with knitted HTML, an `IPA Files/` folder holding the in-principle-acceptance proposal and its R code, and a transparency checklist. Raw and aggregated data both present, plus the original and revised analyses — you can check whether the revision changed the conclusion.
### C100 — The Dark Versus Bright Side of a Smiley
*psychology (social cognition) · R + SPSS · verified_complete: **yes***
https://osf.io/n7yc4/
A preregistered replication of Experiment 3 in Glikson et al. (2018). 16 nodes organised as `Data and code/{Excel,SPSS,Codebook,Code}/`, `Experimental material/` (including the Qualtrics `.qsf` survey definition), and a `Statcheck/` folder with statcheck output run against the original paper's reported statistics. Shipping the Qualtrics instrument and a statcheck audit of the target paper is exactly the posture a verification board wants.
### C101 — Feedback-Related Negativity and Reward-Prediction Error in Trial-and-Error Learning
*psychology / cognitive neuroscience (EEG) · MATLAB · verified_complete: **yes***
https://osf.io/gf83h/
46 nodes, 2.8 GB: `code/` with `behavioral_analysis.m`, `erp_analysis.m` and `make_figures.m`; `data/behavioral_summary.mat`; `outputs/{figures,tables,models}/`; a top-level `README`; and — the reason this one is unusually self-contained — a bundled `eeglab2019_0.zip`, pinning the toolbox version rather than assuming the reproducer installs a compatible one.
### C102 — Hospital Admissions for Bipolar Disorder in Chile, 2001-2025
*psychiatric epidemiology · R · verified_complete: **yes***
https://osf.io/y67f8/
Submitted to *BMC Psychiatry*. 53 nodes: `datos/` (the harmonised cohort of 63,268 F30/F31 discharge episodes, national population projections, all-cause discharge counts), `codigo/` with 13 R scripts, `figuras/`, `resultados/`, `README.md` and a `MANIFIESTO.txt` inventory. Built from Chile's open DEIS and INE data, and it ships a download manifest so the provenance chain is checkable.
### C103 — Cross-Temporal Ecological Replications of Gendered Responses to Expressions of Anger
*psychology / political behaviour · Stata + R · verified_complete: **could-not-check***
https://osf.io/cbe7g/
60 nodes organised as `Code for Data ID 1-3/` through `Code for Data ID 10/` plus `Meta-analysis/` and `Independent rating/`, holding 16 do-files and 17 .Rmd files and README.txt files at branch level. The structure is a genuine multi-study replication series, which is interesting. I could not confirm from the tree whether participant-level data accompanies every study branch, so I am not claiming completeness.
### C104 — Does Internet Use Translate into Time Relief? Women's Housework in China
*sociology / development economics · Stata · verified_complete: **no***
https://osf.io/nuzvy/
22 nodes: `run_all.do` at the root, `code/`, `data/README_data_access.md`, `output/` with non-disclosive aggregate model results, `CITATION.cff`, `LICENSE` and README.md. The record is candid that it ships "Stata replication code and non-disclosive aggregate outputs" — the CFPS panel microdata must be obtained from the data provider. Marked `no`, but the aggregate outputs mean a partial check against published tables is still possible.
### C105 — Scalable Bayesian Prediction for High-Resolution Areal Disease Forecasting
*statistics (spatio-temporal/Bayesian) · R (INLA) · verified_complete: **no***
https://github.com/spatialstatisticsupna/Scalable_Prediction
Orozco-Acosta et al., *Biometrical Journal* 65(8), 2023. Opened the repo: `Rcode/`, `data/`, `results/`, `Scalable_Prediction.Rproj`, README.md. The README organises scripts by paper section and documents dependencies precisely (R 4.2.0, INLA, the authors' `bigDM` package). Marked `no` for a reason worth recording: the real cancer mortality data for Spain's 7,907 municipalities is confidential, and the deposit substitutes structurally comparable synthetic data. You can verify the method, not the published numbers.
### C106 — EEGManyPipelines Analysis Code
*psychology / statistics (multi-analyst) · Python (MNE) · verified_complete: **no***
https://zenodo.org/records/6549063
One team's pipeline from the EEGManyPipelines multi-analyst project. 1.2 MB, 33 entries: 16 Python scripts (`03_run_ica.py`, `06_run_autoreject.py`, …), `environment.yml`, `README.md` read from the archive, a `CITATION.cff`, and two instruction PDFs ("Instructions for analysts", "Instructions for the script folder"). The README carries a GitHub Actions "Run analysis" badge, so the pipeline is CI-exercised. Marked `no` on data only: the shared EEG dataset is distributed by the EEGManyPipelines organisers, not bundled. Valuable as a study of analytic variability rather than a single result.
### C107 — Bender Study: Pattern Analysis of Memory Reactivation, Suppression and Integration
*cognitive neuroscience (fMRI) · Python · verified_complete: **could-not-check***
https://zenodo.org/records/6967583
Preston lab. 2.2 MB, 73 entries: 25 Python files, 26 shell scripts, 6 notebooks, `pyproject`-style `.toml`, README.md read from the archive. The README is refreshingly honest — it says the steps are "designed to reproduce all analyses presented in the paper" but that "the procedure has not been repeated since the original run, and may be missing some steps", and that it was tested on the Lonestar 6 cluster. Code and instructions yes; imaging data is external, and the authors themselves flag the instructions as possibly incomplete.
### C108 — Decision-Making Dynamics Are Predicted by Arousal and Uninstructed Movements
*neuroscience (behaviour/electrophysiology) · Python · verified_complete: **no***
https://zenodo.org/records/10306018
Hulsey et al. 2023. 7.1 MB, 223 entries: 77 Python files, 14 notebooks, seven README.md files at directory level, an `environment.yml`, and a Cython extension. Marked `no` because the record states the related NWB data files are deposited on DANDI (10.48324/dandi.000678) rather than here — public, but a separate fetch. A good example of the increasingly common code-here/data-there split.
### C109 — eComp Analysis Code
*cognitive neuroscience (EEG) · Python (MNE) · verified_complete: **could-not-check***
https://zenodo.org/records/7335171
666 KB, 223 entries: 12 Python scripts including `03_run_ica.py`, an `environment.yml`, a `CITATION.cff`, README.md and a plain `README`, plus 91 .json and 27 .tsv files that look like BIDS sidecars rather than the recordings themselves. Environment and code are pinned and present; whether the EEG data needed to execute the pipeline is included, the manifest does not settle.
### C110 — Detecting Edgeworth Cycles
*economics (industrial organization / ML) · Python · verified_complete: **yes***
https://zenodo.org/records/10126406
Holt, Igami and Scheidegger. 145 MB, 29 entries; I read the README.md out of the remote zip — the authors explicitly offer the repository as "a sandbox for researchers interested in studying gasoline price data". Ships 20 Python files including LSTM and estimation frameworks, a `replication_conda_environment.yml`, a README PDF and the paper-plus-appendix PDF. A methods-comparison paper (classifiers for retail-fuel price cycles) with a pinned conda environment and public price data.
### C111 — Mapping Firms' Locations in Technological Space: Topological Analysis of Patent Statistics
*economics / applied topology · Python · verified_complete: **yes***
https://zenodo.org/records/7957848
39 MB, 322 entries. I read the README.md from inside the zip: it is a proper setup guide (virtualenv recommended, module install, then the scripts). Ships 198 .csv files of yearly cosine distances plus a `requirements.txt` and multiple nested READMEs including one for the Jaffe measures. Built on public USPTO patent statistics, so nothing is gated.
---
## Notes on this run
### Step 0 — what I could actually reach
I fetched all six required hosts and confirmed each with a real HTTP status code, not a vibe.
| Host | Result |
|---|---|
| `https://zenodo.org` | **200 — reachable.** Web pages and the full REST API both work. This became the backbone of the sweep. |
| `https://www.openicpsr.org/openicpsr/search/aea/studies` | **403 — blocked.** Cloudflare interstitial ("Just a moment...") on every path I tried, including `/openicpsr/oai` and individual project pages, with and without a browser User-Agent. |
| `https://dataverse.harvard.edu` | **202 with an empty body — blocked.** The response carries `x-amzn-waf-action: challenge`; AWS WAF is challenging the request and returning nothing. Both the web UI and `/api/*` are affected. |
| `https://osf.io` | **200 — reachable.** `api.osf.io/v2` works (occasional transient 502s, retried). |
| `https://github.com` | **Reachable, but only through some paths.** `raw.githubusercontent.com` returns 200. `api.github.com` is bound to this session's configured repository and refuses search and arbitrary-repo endpoints. `github.com` HTML and `codeload.github.com` return 403 to `curl`. The WebFetch tool *can* render github.com repository pages, and that is how every GitHub candidate here was opened. |
| `https://codeocean.com` | **403 to `curl` at the root; `/explore` returns a 200 SPA shell** with no server-rendered content, and `/api/v1/capsules` returns 403/404. I could not enumerate or open a single Code Ocean capsule, so **no Code Ocean candidates appear on this list.** |
I tried to get around the two Cloudflare/WAF blocks with the pre-installed Chromium via Playwright, on the theory that a real browser would clear the JS challenge. It failed for an unrelated reason: Chromium cannot reach the network through this session's egress proxy at all — every request, including `https://example.com`, dies with `ERR_CONNECTION_RESET`, with the proxy configured explicitly, with QUIC and HTTP/2 disabled, and with certificate errors ignored. `curl` to the same URL returns 200. So there was no browser route available and I stopped after four attempts.
### What the openICPSR block cost, and what I did instead
This is the biggest hole in the run. openICPSR holds thousands of AEA replication packages that have already been through the AEA Data Editor's verification — exactly the pre-screened population the brief pointed at, and I got **zero** of them.
What I did instead was find the *other* economics deposit venues that are reachable:
- **Zenodo turned out to be the Econometric Society's and the Review of Economic Studies' deposit host.** A large share of the economics rows here are official `Replication package for: ...` deposits for *Econometrica*, *Quantitative Economics*, *Review of Economic Studies*, *Economic Journal* and *Journal of Human Resources* — the same data-editor-verified population as openICPSR, just at a different DOI prefix. That substitution is why economics is still the largest domain on this list.
- **Borealis (`borealisdata.ca`)** carries the *Canadian Journal of Economics* replication deposits, with unrestricted Stata code and data.
Other non-Harvard Dataverse instances I confirmed reachable but only lightly sampled or did not use: `dataverse.no` (763 replication datasets), `edmond.mpg.de` (Max Planck), `data.sciencespo.fr`, `demo.dataverse.org`. `dataverse.nl` reset the connection. **Harvard Dataverse remains the single largest untouched pool** — it is where most political-science replication data lives, and its absence is why political science is thin here (the poli-sci rows I do have came from GitHub instead).
### How I sampled
Not randomly, and not exhaustively. Roughly:
- **Zenodo** — keyword searches per domain (economics, genomics, ecology, psychology, statistics) restricted to `type=dataset` and `type=software`, sorted by `mostviewed` rather than `mostrecent`. That sort choice matters: sorting by recency surfaced a lot of very new, unreviewed, sometimes plausibly machine-generated "replication packages" with no journal behind them. Sorting by views pulled up established journal deposits instead. I discarded the recency-sorted batch.
- **Dryad** — keyword searches through `/api/v2/search`, then the version file listing for each hit, then the dataset landing page (Dryad renders the full curated README in the page HTML). Dryad requires a README and links a primary article, so the hit rate for genuinely complete packages was the highest of any host.
- **Dataverse** — searching at the *file* level (`fileName:*.do`, `fileName:*.R`, `fileName:*.py`, `fileName:*.Rmd`) rather than the dataset level, then resolving parent datasets. Dataset-level keyword search on these instances returns mostly raw survey data with no code; file-level search finds the packages that actually contain analysis scripts. This was the single most useful trick of the run.
- **OSF** — `api.osf.io/v2/nodes` filtered by title and tags, then walking `osfstorage` recursively to build the file tree.
- **GitHub** — web search to find candidates, then WebFetch on each repository page to read the tree and the README. Slower and one-at-a-time, but it is a real read.
### How "opened" was established, per host
This matters for judging how much to trust each row:
- **Zenodo** — for every zipped package I read the **zip central directory over HTTP range requests**, which yields the complete internal file manifest without downloading the archive. This worked on packages up to ~6 GB (the "Networks, Barriers, and Trade" deposit: 573 entries read from about 100 KB of transfer, ZIP64 and all). Where a README lived inside the zip I range-fetched and inflated that single member and read it. So for most Zenodo rows I have seen the actual internal structure, not just the outer file name.
- **Dryad** — API file listing plus the rendered README on the landing page. **I could not read inside Dryad's `.zip` files:** Dryad's file-download endpoint requires an API bearer token (`401 Unauthorized`) and the public `file_stream` path returns 403. Every Dryad row whose payload is a single zip is therefore marked `could-not-check`, and I have said so on each.
- **Dataverse** — full file manifest with per-file sizes and the `restricted` flag, via `/api/datasets/:persistentId/versions/:latest/files`. The restricted flag is why the Uganda paediatric-sepsis row is a confident `no`.
- **OSF** — recursive `osfstorage` walk giving the real folder tree.
- **GitHub** — the rendered repository page: top-level tree plus README.
### On the count
The brief asked for 40-100 and this list has 111. I did not pad it: every row is something I opened, and the overshoot is mostly the Zenodo range-request trick working better than expected — once reading a 6 GB package's manifest costs 100 KB and a few seconds, the marginal cost of one more verified row collapses. If the board wants to enforce the ceiling, the honest thing to cut first is the 15 `could-not-check` rows, not the 25 `no` rows: the negatives are checked facts, the could-not-checks are open questions.
### What I am not sure about
- **`could-not-check` is doing real work here and I have not rounded it to `yes`.** It means one of: I saw code and data but no README (several Dataverse and a few Zenodo rows); or I saw a README that describes the contents without giving a run procedure; or the payload is a Dryad zip I could not open. Each row says which.
- **A manifest is not an execution.** Seeing `master.do` next to twelve `.dta` files is strong evidence and not proof. Paths may be absolute, dependencies unpinned, a `.dta` may be a lookup table rather than the estimation sample. Several `yes` rows will fail on first contact, and finding out which is the point of the exercise.
- **I judged "data present" partly by file extension and size.** A 440 KB economics package with eighteen do-files almost certainly has no microdata in it — that inference is behind the `could-not-check` on "Assortative Matching on Income". It is an inference, not a reading of the README.
- **Journal attributions come from the deposits' own metadata**, not from checking the published article. Where a deposit says "forthcoming in *Review of Economic Studies*", I repeated that; I did not verify publication status.
- **I have not deduplicated against whatever is already on the board.** The repo's history mentions an earlier six-paper GitHub run (FIND-002); there may be overlap I did not check for, since the brief was to write only these two files.
- **Domain balance is uneven and openICPSR is the reason.** Economics and ecology/evolution dominate. Political science is under-represented because Harvard Dataverse was unreachable. Psychology comes almost entirely from OSF preregistered-replication projects, which skews toward small experiments. Bioinformatics rows skew toward code-with-external-data, which is simply how that field deposits.
### Hosts holding a lot I did not get to
- **openICPSR / AEA** — thousands of pre-verified packages. Completely blocked. Highest-value target for a rerun from a network that can clear Cloudflare.
- **Harvard Dataverse** — blocked by AWS WAF. The main political-science pool.
- **Code Ocean** — reachable only as an empty SPA shell; its whole point is executable capsules, which would have been the most directly verifiable artifacts on offer. Zero coverage.
- **Dryad** — I sampled maybe 74 datasets against a corpus in the tens of thousands, and only across ecology, evolution and behaviour. Barely scratched.
- **Zenodo** — the Econometric Society / RES deposit series alone runs to hundreds of packages; I opened a few dozen. `dataverse.no`, `edmond.mpg.de` and `data.sciencespo.fr` are all reachable and essentially unsampled.
- **figshare** — API returns 200 and I confirmed reachability, but I ran out of run before sampling it at all.