jinflow a jazzisnow engine

An introduction · for scientists

jinflow

A declarative engine for diagnostic data pipelines — where a result carries the method that made it, and a number remembers it is a measurement.

The ache

Three quiet failures we've made our peace with.

01

Results you can't reproduce

The figure is in the paper. The spreadsheet that made it left with the postdoc.

02

Tools you can't leave

Your method lives inside an application's clicks — not written down, not portable, not yours.

03

Numbers that forgot

Units on a label if you're lucky, uncertainty nowhere, and a mean of things that should never be averaged.

The turn

jinflow is an engine for Studys.

A Study is your data and your declared method, kept together and kept honest. You don't rent a seat — you keep a Study: as long-lived as your programme, as private as your laptop, as reproducible as a commit.

Its framework is the AFS; its built store is the KLS; you read it in the Explorer. We'll meet each where it earns its keep.

One word, many sizes

A Study is not one size — it stretches across five axes.

Longevity
an afternoon's explorationa decade of monitoring
Deployment
your laptop — data never leavesan always-on public showcase
Complexity
one CSVa pipeline — bronze→silver→gold
Volume
a handful of rows30.6 M measurements
Sovereignty
fully open dataconfidential — your data · your machine · their browser

The method · the diamond

Ask a question. Let the data answer — and show its work.

Signal

A declared query

A diagnostic pattern over your gold data — a balance, an outlier, a stale reference — emitting standardized findings.

Thesis

A claim, weighed

A plain-language hypothesis, scored against the evidence of its signals: confirmed, plausible, or not observed.

Verdict

The reason why

When a thesis is confirmed, a rule-based root cause — with a recommendation, and a trail back to every row.

You write YAML; jinflow compiles the SQL, builds the store, and serves the finding. Prior expertise enters too — as attributed Subject Matter.

Honest by construction

Reproducible. And measured.

Reproducible

The store carries its own method

Every built store travels with the exact framework that produced it — the Imprint — stamped to a commit. Hand it over, and a colleague can see, and re-run, how every number was made. No black box, no lost macro.

Measured

A number that knows what it is

Every column declares its kind, its unit (UCUM), how it may roll up, and its uncertainty (GUM). So a value isn't 35.1

35.1 µg/m³ ± 2.3 (1σ) · mass concentration · averageable

The engine won't average what mustn't be averaged.

See it breathe · aeros

Air quality, honestly measured.

30.6 M
measurements
6
pollutants
hundreds
of stations

Click a station on the map → its co-located pollutant streams → the measured series, with its uncertainty and reference line. Magnitude, not a yes/no threshold, does the colouring.

within> 4× the guideline
NO₂ · monthly mean1.52× WHO
2022— dashed: WHO annual guideline —2024

What a Study gives you

Your data. Your method. Reproducible, measured, and yours.

Keep it as long as your science — a Study outlives any single question.
Run it where your data must stay — from your laptop to the open web, unchanged.
Hand it over, and it still tells the truth — units, uncertainty, and the whole recipe travel with it.

jinflow — a jazzisnow engine.