Give Every Number One Definition

Why do your programs report the same thing three different ways?

Because a word like “employed” or “completed” was never written down once and agreed on. A data dictionary fixes that: one plain definition per field, so every program, every year, and every funder counts the same way. It’s the least glamorous step in the journey and the one that saves you the most pain.

You already have a short metric list from the last step. A data dictionary is that list with a definition attached to each item — what the field means, what unit it’s in, and which answers are allowed.

One definition, written down, applied everywhere

The problem it solves is quiet but expensive. One program marks someone “employed” if they have any paid work; another only counts full-time; a third includes a paid internship. Each is reasonable. Together they make your rolled-up number meaningless. Writing the single definition down — “employed = paid work of 20+ hours/week, held for at least 30 days” — is what makes two programs finally comparable.

The same goes for the messy details that eat your time later: how names are spelled, how dates are formatted, what counts as “other.” Decide once, up front.

Try it: turn your metrics into a starter dictionary

Paste your metric list. This prompt drafts a first data dictionary you can refine with your team.

Turn the metrics below into a data dictionary. For each field give: the field name, a one-sentence plain-English definition, the unit or answer type (number, date, single-choice, short text), and the allowed values or categories. Where a definition could be read two ways, flag it and propose the stricter reading. Return a table: Field · Definition · Type · Allowed values. Metrics: [paste from step 2]

That draft is genuinely useful — print it, argue about it, tighten it.

Where a chat tool stops

A general model writes a tidy dictionary. What it can’t do is make anyone use it. The definition lives in a document; the next survey still lets someone type “F/T” instead of the agreed value; last year’s data still counts things the old way. And when the same person shows up under two spellings, the tool has no way to know they’re the same. So the dictionary exists, but your data still drifts.

How the definition actually holds

In Sopact Sense the dictionary isn’t a document beside the data — it is how the data is collected and read. A field means the same thing at entry and in analysis, across every program and year. Change a definition and past responses re-score to match, instead of leaving you with two incompatible histories. And when a name or date is entered slightly differently, the system flags and reconciles it instead of silently double-counting — the kind of quiet correction one nonprofit leader noticed:

“It picks up when the same person is entered with different spellings, or a date is clearly a typo, and reconciles it — with a note so I know. That used to be manual.” — Marco, Open Play Foundation

When this helps — and when it doesn’t

Build a dictionary the moment a field will be used by more than one person, program, or year. If you truly collect one thing, once, and never compare it to anything, you can skip it.

Next: Collect Clean Data at the Source → · or see the data dictionary in Sopact Sense →