Clinical Data Toolkit (CDT)

← All projects

One Streamlit workbench that hosts these tools behind a common entry point with structured logging, a plugin-style QC registry for figure checks, and a multi-file TFL version-comparison tool.

What it adds

Conformance tools check data, not figures. This checks a figure against the dataset that produced it — a class of QC that neither a standard checker nor a manual eyeball reliably catches.

How it works

InputsFigures (PDF / PNG)Source datasetsTFL output versions

Process1Ingestion to extractor2QC registry (plugin checks)3Dataset bridge (figure vs data)4Issue aggregator

OutputsFigure QC issuesTFL diff reportExcel / PDF export

Typical layout

By the numbers

v25

Version

209

Test functions

10+

Figure check groups

Export formats

Screenshots

Add image

The CDT landing page showing the available tools (Figure QC, TFL Compare, dataset utilities) and the theme picker

Drop clinical-data-toolkit-01-home.png into
/public/screenshots/clinical-data-toolkit/

Add image

The Figure QC tool after a figure and its dataset are loaded, showing detected issues grouped by check category

Drop clinical-data-toolkit-02-figureqc.png into
/public/screenshots/clinical-data-toolkit/

Add image

The TFL Compare tool showing a side-by-side diff between two versions of an output with changed cells highlighted

Drop clinical-data-toolkit-03-compare.png into
/public/screenshots/clinical-data-toolkit/

Data flow

Several recurring clinical-programming chores — checking figures against their data, comparing two versions of a TFL, inspecting datasets — were scattered across ad-hoc scripts with no shared logging or structure.

Entry point: app.py (one logging setup for all tools)
        |
        +----------------------------+----------------------------+
        v                            v                            v
  Figure QC                    TFL Version Compare          Dataset utilities
  (figureqc/)                                                
   ingestion -> extractor                                   
        |                                                    
        v                                                    
   QC Registry (qc/registry.py)   plugin checks:            
        |   axis / legend / unit / errorbar / structural /  
        |   metadata / cross-figure / advanced_*            
        v                                                    
   Issue Aggregator -> Export (Excel / PDF)

Engineering trade-offs

Plugin-style QC registry for figure checks

Each figure check (axis, legend, unit, errorbar, structural ...) registers itself, so the check set grows without editing a central dispatcher.

One logging setup at the app entry point

Every module uses logging.getLogger(__name__), so all tools write to one run log a user can download.

Pinned dependency ranges with upper bounds

Upper bounds prevent silent adoption of major versions with known breaks, e.g. pyreadstat pinned to 1.2.x for a stable metadata API.

Dataset bridge for figure-vs-data QC

Figures are checked against the dataset that produced them, catching mismatches a visual review would miss.

At a glance

A quick visual read of the countable facts; full detail in the table.

Test functions209

Figure check groups10

Export formats2

Relative scale · values labelled · unit: count

Processing characteristics

Metric	Value	Notes
Test functions	209	Counted across the test suite
Version	v25	From app.py header and changelog
Figure QC check groups	10+	axis, legend, unit, errorbar, structural, metadata, cross-figure, advanced_*
Tools hosted	Figure QC, TFL Compare, utilities	Multi-file TFL version compare added in this line
Exports	Excel, PDF	Issue export from Figure QC
Logging	Structured, downloadable	Per-session run log buffer

Functional wins

01Hosts multiple clinical-programming tools behind one entry point with a single shared logging setup.

02Checks figures against the dataset that produced them through a dataset bridge, catching data-vs-figure mismatches.

03Adds figure checks via a self-registering plugin registry, so the check set grows without touching a central dispatcher.

04Compares two versions of a TFL across multiple files and highlights what changed.

Module dependencies

core

Python
pydantic
pyyaml
Pillow

streamlit

data

pandas
numpy
openpyxl
XlsxWriter
pyreadstat

testing

pytest