CDK Outputs

cdk.github.io/cdk-outputs

The software, data resources and services from the Chemistry Development Kit ecosystem.

What this is not

It is not a bibliometric dashboard. The department's publication record lives in Pure and is reported from there; duplicating it here would only produce a second set of books that disagrees with the first. What this tracks is software: repositories, releases, packages, containers, running services, and the citations of the papers that describe each tool.

It is not a measure of individuals. There are no per-person pages, counts or rankings, and there never will be — no person is queried and no ORCID is stored. Nor are there stars, forks, h-indices or journal rankings: every figure names what it counts, and the Methods page says what each one does not mean.

It is also not service monitoring. Whether the department's endpoints are up right now is a different job with a different cadence, and the cluster already has monitoring for it.

How it works

flowchart LR
    subgraph edit["config/ (CSV tables, the only files a person edits)"]
        direction TB
        proj["<b>projects.csv</b> + <b>identifiers.csv</b><br/>services.csv · links.csv<br/>the tools and what they publish"]
        sem["<b>metrics.csv</b><br/>what each number counts"]
        exc["<b>exclusions.csv</b><br/>what is left out, and why"]
    end

    subgraph coll["collectors (read public APIs, no credentials)"]
        direction TB
        tools["<b>per project</b><br/>github · ecosystems · bioconductor<br/>dockerhub · rsd · citations"]
    end

    proj --> tools
    tools --> guards{{guards}}
    sem -. "no definition,<br/>no figure" .-> guards
    exc -. "and the reason<br/>gets published" .-> guards

    guards -- "fails a rule" --> quar["quarantined in the run<br/>manifest, never published"]
    guards -- passes --> snap[("data branch<br/>one whole-state<br/>snapshot per run")]
    snap --> csv["docs/data/*.csv<br/>one file per metric"]
    csv --> build["build"]
    sem -. "labels and definitions" .-> build
    build --> site["GitHub Pages"]

    classDef cfg fill:#3d6fb422,stroke:#3d6fb4,color:#000
    classDef gate fill:#c26a3a22,stroke:#c26a3a,color:#000
    classDef out fill:#4a8a7222,stroke:#4a8a72,color:#000
    class proj,sem,exc cfg
    class guards,quar gate
    class snap,csv,site out

Loading

One GitHub Actions job, weekly, does the whole run: collect, check, build, deploy, then commit the snapshot. Deploying in the same job is deliberate — a push made with GITHUB_TOKEN does not trigger another workflow, which is how an earlier reporting dashboard served a two-month-old page while every run showed green.

Three properties hold by construction, and each has a test:

• Nothing is published without a definition. A metric absent from metrics.csv does not render, and the Methods page is generated from that same file, so a definition cannot drift from the figure it describes.
• A failed source reads as "not collected", never as 0. Summing the records of a source that returned nothing gives zero, and a tile then states it as fact.
• No stored credentials. Every source is public. tests/test_no_secrets_required.py fails if that stops being true.

Working on it

make install     # into a virtualenv
make check       # config validation, lint, offline tests
make offline     # build the entire site from fixtures, no network, no credentials
make serve       # http://localhost:8000

make offline is the one that matters. If it passes, the project can be forked, handed over, and rebuilt in five years.

Changing what is tracked

Everything a human edits lives in config/ as a CSV table, one row per thing, so it opens in a spreadsheet and a change reads as a one-line diff. tgx doctor validates every table in CI — columns, foreign keys, enumerations — so a malformed change fails the pull request rather than the next refresh. The columns are documented in config/README.md.

To do this	Edit
Track another project	config/projects.csv and config/identifiers.csv, then tgx doctor --projects
Leave something out	config/exclusions.csv; a reason is required and is published
Add a number to the page	config/metrics.csv first, or it will not render

To have a project, repository, package or DOI excluded from the queries, open an issue or email the address in config/settings.csv. No reason is needed and none will be asked for.

Numbers that look wrong

They may well be. Every figure links to the CSV behind it and shows the exact source and collection date, and the run manifests on the data branch record what each source returned, what failed and what was quarantined. Open an issue — see RUNBOOK.md for what to do when a collector breaks.

Prior art

Modelled on RECETOX/specdatri_reporting (MIT), the reporting tool RECETOX built for the same problem. The per-source card grid over a per-project table is theirs, and several of the guards here exist because that project met the failure first and it was worth turning into a test. No code was copied — this was written from scratch — so the debt is one of design, and it is recorded in CITATION.cff and .zenodo.json as well as here.

The Research Software Directory already tracks much of this department's software and computes literature mentions for it; this project consumes that rather than rebuilding it.

Licence

Code MIT. Figures are derived aggregates of public data; every source and its terms are listed on the site's Methods page, which is generated from the same config the collectors read.