@sanity/groq-condition-describe

Describe GROQ boolean expressions — filters, predicates, gates — without an LLM:

• analyzeCondition — static analysis from the groq-js AST: which $params the expression reads, and its boolean structure as an and/or clause tree (negation pushed to the leaves).
• explainCondition — evaluate against an in-memory dataset + params: three-valued per-atom verdicts (GROQ's Kleene logic), the unmet frontier (blockedBy), single-fix pivotality, and solvable requirements ("$fields.approved must be true").
• whatIfCondition — counterfactual re-evaluation with one param value replaced.
• describeCondition / checklistLines — deterministic English phrasing as a checklist; every phrase is an InsightPhrase ({id, params, text}) so i18n consumers translate by message id. Hosts wire their own vocabulary through DescribeContext seams (renderRead, renderRequirement, atomPatterns, nowParam); anything unphrasable falls back to its GROQ and marks fullyDescribed: false.

import {explainCondition, describeCondition, checklistLines} from '@sanity/groq-condition-describe'

const insight = await explainCondition({
  condition: '$fields.legalApproved == true && $allActivitiesDone',
  dataset: [],
  params: {fields: {}, allActivitiesDone: true},
})
const description = describeCondition(insight, {})
description.summary // 'Needs: $fields (legalApproved) must be true'
checklistLines(description) // ['All of:', '  ✗ …', '  ✓ …']
Copy

Extracted from @sanity/workflow-engine, which layers its workflow vocabulary (field titles, role/claim desugar shapes, $now) onto these seams — see its insight surface for the fully-worded result.