Skip to content
Vibedata

Build something that is not there yet

Build clinical-trial enrollment and site-performance models

  • When enrollment against target lives in a slide deck that gets updated by hand from EDC exports.
  • When screen-failure rate and enrollment velocity are recalculated differently by every site coordinator.

Build screened, enrolled, and randomized counts per site per week against target, with screen-failure rate and enrollment velocity. Covers site and study-level counts, not subject-level clinical data.

Area
Transformation
Runs on
  • Microsoft Fabric Lakehouse
  • Microsoft Fabric Warehouse
  • MotherDuck
  • DuckDB
Built with
  • dbt
Industry
Life sciences
Readiness
SupportedEverything this Recipe composes runs today, without a case that proves this exact shape.
Before you start
needs agreed stage definitions and a target per site

Sample This Recipe has not been materialized in the Cookbook repository yet. Its trigger, description, prompt, agent guidance and acceptance conditions, and the explanation below, are prototype drafts. Its name, job, area and readiness come from the reconciled Cookbook seed snapshot. Readiness is a separate question from this one: it says whether the capability exists, not whether the writing has been reviewed.

Use this Recipe

Use the VibeData Recipe `life-sciences-enrollment` at https://getvibedata.ai/cookbook/life-sciences-enrollment Read the Recipe and execute it in the context of the current Intent.

Recipe id life-sciences-enrollment · Not yet materialized in the Cookbook repository, so the pointer addresses this page.

Verified by

What has to be observably true before this Recipe is finished.

  • randomized is never greater than enrolled, and enrolled is never greater than screened, at any site-week
  • a reproduced site's counts match the EDC export for that week
  • the stage definitions for screened, enrolled, and randomized are documented on the model and applied consistently across sites
  • enrollment velocity is computed on the same window for every site
Recipe promptThe task specification the agent reads. Reference only — it is not what you copy.
Deliver: Build clinical-trial enrollment and site-performance models.

Build screened, enrolled, and randomized counts per site per week against target, with screen-failure rate and enrollment velocity. Covers site and study-level counts, not subject-level clinical data.

Execute inside the current Intent. Its Domain, repository, platform, environment and attached sources are the context for this work — read them rather than asking for them.

The work is done when:
- randomized is never greater than enrolled, and enrolled is never greater than screened, at any site-week
- a reproduced site's counts match the EDC export for that week
- the stage definitions for screened, enrolled, and randomized are documented on the model and applied consistently across sites
- enrollment velocity is computed on the same window for every site

Report the evidence for each condition above with the result. A condition you cannot meet is something to say, not something to work around.
Agent guidanceHow the agent approaches the work, and what it will not do.

Profile the inputs the grain, joins and measures actually depend on before proposing a model. Put the design up for review — grain first — then build in an isolated copy with tests and documentation landing beside the model rather than after it.

Composes

  • dbt model authoring and layering
  • dbt in the project
  • isolated-copy execution and gate verification

Asks first

Semantic decisions the Intent cannot supply. Never context Studio already holds.

  • which Fabric target this work lands on, when the Domain carries both a Lakehouse and a Warehouse

Guardrails

  • Build in an isolated copy. Production is read, never written.

What you need

  • Microsoft Fabric Lakehouse, Microsoft Fabric Warehouse, MotherDuck, or DuckDB.
  • A dbt project you can build, and read access to the models it starts from.
  • dbt in the project, or the intent to add it.
  • Needs agreed stage definitions and a target per site.

How it goes

  1. State the outcome in one sentence, in the language the request arrived in.
  2. Let it profile the inputs the grain, the joins and the measures actually depend on.
  3. Review the design. Disagreeing about grain here costs a sentence; after the model exists it costs a rewrite.
  4. Let it build in an isolated copy, with the tests and the documentation landing beside the model rather than after it.
  5. Read the acceptance conditions against the run.

What you end up with

The deliverable, in your own repository, as transformation work a reviewer who knows the project reads as native to it. Alongside it, the evidence for every one of the acceptance conditions above — which is the part that is still there in three weeks when somebody asks.

Next to this one