Editions

MaudeDash comes in two forms. Both run the same analyses and the same statistics; they differ in what data they can reach and what you have to install.

No installation

Browser Edition

The tool at maudedash.com. Open it and start querying. A database engine compiled to WebAssembly runs inside the page and reads a 1.4 GB Parquet extract of the corpus over the network, fetching only the parts each query needs.

What you get

All 20,746,963 reports, 1991–2024
Every analysis: harm rates with confidence intervals, subgroup forest plots, trend tests, disproportionality with both frequentist and Bayesian signal criteria
Event narratives to 4,000 characters per report
CSV export, shareable cohort URLs, generated STROBE methods

Limits

Narratives are capped at 4,000 characters. That covers 99.67% of source text segments, but not the complete multi-part text.
Analysis runs in a browser tab, so very large cohorts are capped to stay within browser memory.

Requirements

Any current browser. Roughly 18 MB on first visit, about half that on return visits.

Open the tool

Free download

Research Edition

A Streamlit application you run on your own machine against the complete corpus you build from the FDA's own files. Use this when you need the untruncated narratives, unlimited cohort sizes, or a fully auditable pipeline from FDA source file to published figure.

What it adds over the browser

Complete multi-part narratives, not truncated
No cohort size limits — export the full filtered population rather than a capped extract
The whole build pipeline: ingestion, narrative reassembly, outcome decoding, validation and a 25-test suite
Independent verification — rebuild the corpus yourself and confirm every number this site reports

Requirements

Python 3.10+, roughly 120 GB free disk for the raw FDA files and the built database, and 8 GB RAM or more. The full build takes a few hours, mostly unattended, and is resumable.

Get it on GitHub

Installing the Research Edition

Four steps. The long one is step 3, and it only happens once per FDA release.

1. Get the code

git clone https://github.com/mokshalstudios/MAUDE-DASH.git
cd MAUDE-DASH
pip install -r requirements.txt

2. Download the FDA source files

From the FDA's MDR data files page, take the MDR master, device, patient, foitext, foidevproblem and patientproblemcode archives, and unzip them all into one folder. That is roughly 40 GB of pipe-delimited text.

3. Build the database

python research-edition/maude_build.py --raw-dir /path/to/fda/files \
    --db maude_final.duckdb --skip-fts

A few hours on the full corpus. The build is resumable — if it is interrupted, re-run the same command and it continues from the last completed step rather than starting over. --skip-fts omits the full-text index, which the dashboard does not use and which roughly doubles build time.

4. Run it

streamlit run research-edition/maudedash_app.py -- --db maude_final.duckdb

Confirm the install with the test suite, which should report 25 passing:

python research-edition/test_maude.py

Publishing your own instance

The repository also contains everything needed to host a browser edition of your own — useful if you want a tool over a different cohort, a newer data vintage, or your own institution's branding. It is static files, so any web host will serve it.

python packaging/maude_export_web.py --db maude_final.duckdb --out web/data
python packaging/build_search_index.py --db maude_final.duckdb --out web/data
python packaging/vendor_assets.py
python packaging/serve_local.py        # preview at http://127.0.0.1:8777

deploy/DREAMHOST.md in the repository documents the full upload procedure, the Apache configuration required for HTTP range requests, and the measured bandwidth a deployment actually consumes.

Which one do I need?

If you want to…	Use
Look up a device and see harm rates, trends or signals	Browser Edition
Produce a figure or a methods paragraph for a manuscript	Browser Edition
Read complete, untruncated event narratives	Research Edition
Export an entire filtered population without a row cap	Research Edition
Verify a published number independently, from FDA sources	Research Edition
Work offline, or on data the FDA has since revised	Research Edition
Host your own instance for a lab or institution	Either — see above

ⓘ

Both editions compute identically. The statistical engine exists twice — in Python for the Research Edition and JavaScript for the browser — and the two are cross-validated against each other and against SciPy and statsmodels on every function, to 1e-9 or better. A number from one is a number from the other. See Methods → Validation.

MAUDE is passive surveillance with no denominator of exposed devices. Figures from either edition are proportions of reports, never incidence or device failure rates. See Methods → Coverage & caveats.