Remove sentiment, emotion, and the TopicsUnnested join from the Cube
schema, web registry, AI prompt, audit-allowlist, and i18n keys — these
required metadata enrichment that is no longer planned. Tests swap the
removed string dimensions for sourceType to keep coverage intact.
Add csatDissatisfiedCount and csatNeutralCount so dashboards can render
the standard satisfied/neutral/dissatisfied distribution alongside the
existing top-2-box metric.
Expose field_label, field_group_label, language, value_boolean,
value_date, created_at, and updated_at as Cube dimensions. fieldLabel
and fieldGroupLabel unlock "group by question" and the matrix/ranking
aggregations enabled by the recent composite-question PR; the others
round out coverage of the underlying feedback_records columns. Extend
FieldDefinition with a boolean type and matching filter operators.
Adds CSAT and CES measures, plus a couple of universal cross-type measures,
to the FeedbackRecords Cube. Also fixes correctness bugs in the existing NPS
measures and constrains the AI chart-query output to known measure / dimension
ids.
Measures added
- csatScore: % of answered CSAT responses rated 4 or 5 (top-2-box on 1-5).
- csatAverage: AVG of answered CSAT ratings.
- csatSatisfiedCount: count where field_type='csat' AND value_number >= 4.
- csatCount: count of answered CSAT responses.
- cesAverage: AVG of answered CES ratings (1-5 or 1-7 depending on the question).
- cesCount: count of answered CES responses.
- uniqueRespondents: countDistinct(user_id).
- uniqueResponses: countDistinct(submission_id).
- npsAverage: AVG of answered NPS ratings (replaces the old `averageScore`
measure, which silently averaged every numeric value across all field types).
Bug fixes (existing NPS measures)
- promoterCount / passiveCount / detractorCount / npsScore now filter
field_type = 'nps'. Before, any numeric value in the band was counted
(so a CSAT 5/5 was lumped in as a detractor).
- npsScore now divides by *answered* NPS responses only (value_number IS NOT
NULL); dismissed/abandoned NPS records no longer drag the score toward 0.
- npsScore returns NULL when no answered NPS responses exist instead of the
literal 0, so empty-data days render as gaps rather than a misleading
flat-zero line.
- Same NULL-safe denominator fix applied to csatScore.
- csatCount / cesCount now exclude dismissed (value_number IS NULL) responses
so they match the score denominators.
Dimensions
- Renamed `npsValue` -> `valueNumber` (the underlying value_number column
stores values for NPS, CSAT, CES, rating, and number field types).
- New `valueText` dimension for grouping by categorical / open-text answers.
AI chart generation
- ZGenerateAIQueryResponse now enum-constrains `measures`, `dimensions`,
`timeDimensions.dimension`, and `filters.member` against the known ids
derived from FEEDBACK_FIELDS. Vercel AI SDK + Gemini structured outputs
enforce these at decoding time, so the model can no longer return invalid
or hallucinated measure names.
- generateText now pins temperature: 0. Same prompt produces the same query.
Breaking changes
- Charts that referenced `FeedbackRecords.averageScore` need to switch to
`FeedbackRecords.npsAverage`.
- Charts that referenced `FeedbackRecords.npsValue` need to switch to
`FeedbackRecords.valueNumber`.
Pre-GA on epic/v5; no production charts are expected to be impacted yet.
Hub 0.3.0 renames the `user_identifier` API field to `user_id` (breaking
change). This commit bumps the Hub Docker image, upgrades the
@formbricks/hub TypeScript SDK from 0.4.3 to 0.5.0, and renames every
`user_identifier` reference in Zod schemas, server actions, transform
pipeline, form components, CubeJS schema, connector types, and seed data
to match the new API contract.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Dhruwang