FIP Matrix App

A Streamlit-based interactive application for visualizing FAIR convergence data from nanopublications.

Three views over the same selection:

• Overview — headline statistics for the whole dataset.
• Matrix — the convergence matrix: FAIR Enabling Resources × communities.
• Analytics — interactive Plotly charts of the current selection.

🔁 Data Synchronization

This app automatically fetches the latest new_matrix.csv from the peta-pico/dsw-nanopub-api repository once a day (cron: '0 0 * * *') using GitHub Actions. The CSV is saved to data/new_matrix.csv and used live in the app.

data/new_matrix.csv is a contract with .github/workflows/sync_fip_matrix.yml — the workflow's curl -o writes to exactly that path and will not create the directory. Don't move or rename either.

📦 Structure

.
├── main.py                     # Streamlit entry point (must stay at the root)
├── config.py                   # Paths, FIP vocabulary, status scale, palette
├── utils.py                    # Loading, filtering, pivoting, aggregation
├── matrix.py                   # The convergence matrix grid
├── charts.py                   # Plotly figure builders
├── ui.py                       # Header, statistics band, legend, logos
├── .streamlit/config.toml      # Theme (light + dark)
├── tools/
│   └── check_palette.py        # Verifies every colour figure quoted below
├── data/
│   └── new_matrix.csv          # Auto-updated data file
└── .github/workflows/
    └── sync_fip_matrix.yml     # Data sync workflow

🎨 Status scale

Status	Meaning	Colour	Compact
0	No data	unfilled
1	Resource in development / future use	#f0a182 apricot	○
2	Available resource / future use	#2f6699 blue	◐
3	Available resource / current use	#b0f05c lime	●

Cells are 44px and isolated in a 96%-empty grid, where hue separation reads far more easily than lightness, so the scale is treated as categorical rather than as a single-hue ramp.

Every figure below is produced by the checker, so none of it is written by hand:

python tools/check_palette.py

It reads the palette from config.py and the theme from .streamlit/config.toml, so it cannot fall out of step with them. It exits non-zero when a rule is broken. PASS/FAIL lines are the rules; INFO lines are documented trade-offs.

Worst-pair separation, stated per metric so the figures cannot drift apart again:

	normal vision	worst dichromat
ΔE76	63.6	24.8 (protanopia)
ΔE2000	30.8	12.6 (deuteranopia)

Dichromat figures come from a Viénot–Brettel–Mollon (1999) simulation over protanopia, deuteranopia and tritanopia. On ΔE76 the worst dichromat pair clears the project's ΔE 15 floor; on the stricter ΔE2000 it falls short at 12.6. So the scale does not rest on colour alone: Compact mode's ○ ◐ ● and the labelled legend carry the same information as a secondary encoding.

These are pastels, so the trade is contrast. On the light surface the three measure 1.3–2.0:1 and read soft; on the dark surface they are 8.4–12.8:1 and pop. Level 3 is the faintest in light mode at 1.32:1 — darken each hue a step if the cells ever feel too weak.

Blue is deliberately absent here. Blue means "chrome or chart" throughout the app — the theme primaryColor, the SEQUENTIAL ramp and the SERIES mark colour all sit in one blue family — so keeping it out of the status scale is what stops a button or a chart bar from being read as a status.

The grid has two render modes. Colour paints cell backgrounds; Compact draws glyphs and involves no CSS at all, which keeps wide selections fast. It is selected automatically for large selections and can be overridden.

📈 Chart palette

Token	Colour	Used for
SERIES	#2f85c2 blue	single-series magnitude
SERIES_ALT	#c2553f brick	only where a chart has its own legend (Metadata vs Data)
SEQUENTIAL	#e3eefa → #2f6699	continuous magnitude, one hue, monotone in OKLCH lightness

Chart marks are mid-tones, not pastels, because they must clear 3:1 against both a near-white and a near-black background — no pastel does both. The pair separates at ΔE2000 45.2, and still at 44.5 under the worst simulated dichromat, which matters because colour is the only thing distinguishing the two series.

One chart reuses the status colours directly for its stacked segments. That is intentional: those segments are the status scale.

🚀 Run locally

pip install -r requirements.txt
streamlit run main.py

Run from the repository root. The CSV is resolved relative to config.py, so the app is not sensitive to the working directory.