Data Integrity · WALKRI
The data is only as good as the field that captured it.
A form that collects answers to imprecisely defined questions produces data that cannot be trusted, compared, or used as evidence, and no pipeline, dashboard, or model can fix that after the fact. WALKRI, Working Architecture for Legible, Knowable, Reliable Instrumentation, is the standard for epistemic quality at the point of capture. It specifies how each instrument must be defined so the data it produces is valid, consistent, provenance-aware, and reusable, by people and machines alike.
The unit is the instrument
An instrument is anywhere one piece of data gets captured.
A form field is the most familiar instrument, but it is only one instantiation. A prompt constraint given to a language model, an element of a structured-output schema, a rubric item a reviewer applies, and an extraction specification run over a document are the same underlying thing in different modalities. The precision an instrument requires does not change when the modality does.
WALKRI carries a bidirectional precision principle: the precision demanded of what an applicant must demonstrate is demanded equally of the fields used to collect that demonstration. It operates on JSON Schema above any one form tool, and its output aligns with Croissant, FAIR, and W3C PROV, so data collected through a conformant instrument can enter any pipeline that consumes it.
What every instrument has to carry
Five requirements, and one that applies when it should.
Each requirement closes a way an instrument can produce data that looks consistent and is not. An instrument that satisfies all five is conformant; one that does not is where the data quietly goes wrong.
Criterion Intent
A field states its measurement claim, not just a label. “Community Engagement” can mean members reached, the depth of co-design, or the frequency of contact, and each is a different instrument. Without a written intent, no one can tell whether the response form captures what was meant.
Operational Definition
Every option or category is defined completely, with qualifying and non-qualifying examples. For numbers: the unit, the counting rule, the boundaries. An undefined option hands interpretation to whoever answers, and nominal agreement then hides real definitional variance in the data.
Response Form
The response type is a measurement decision, not a formatting one. A single-select assumes mutually exclusive categories; a numeric field assumes one common scale. The wrong form for a construct produces systematic error that no later analysis can remove.
Evidence Form
Where a field requires evidence, that evidence must resolve without a login or an access request. Material behind an authentication wall is unreadable to an independent reviewer and does not satisfy the requirement, however strong the underlying document is.
Conformance Threshold
When a field references an external standard, it states the passing threshold before any response is reviewed. “Qualifies as a Digital Public Good” has nine indicators; without a stated threshold, one reviewer requires all nine and another accepts three.
Participatory Specification
OptionalWhere a field measures a condition a specific population lives, WALKRI records whether the definition was validated with that population, so a funder's model of a term like food security can be checked against how the people it measures actually experience it. It applies only where such a population exists.
Where it sits
At the point of capture, where the data is made.
The boundary standards bracket a process; WALKRI works inside it, at the instrument. ORE grades the source that feeds a field; STRUCK specifies what the output owes on the way out; CRAFT makes the chain between them legible. WALKRI grades the field itself: would two independent readers collect the same thing from its definition. A record can fail at any of these, and none of them substitutes for another.
WALKRI is a general instrument standard, first put to work in grants. There, CROSS configures what a round must obligate and WALKRI specifies the data quality of the forms that collect the evidence for it. The instrument discipline is the same wherever structured data is captured; grants is where it is applied first.
WALKRI in grants
crosswalkri.com →How WALKRI configures the fields of a grant round so the evidence a program collects is comparable across applicants and reusable afterward, shown as a round actually runs.