Data sourceData fabric · synthetic v1.0 · 2026-09-02
Current genomedraft · uncommitted
Target
Change history
Phase status
Start with your marketing data
Drop product, audience, channel and keyword exports plus a brand brief; the four-phase loop runs on them.
Explore the MORIÉ sample
A fully synthetic K-beauty dataset — not real company or market data. Good for learning the method first.
Workspaces in this browser
1 · Upload marketing data
CSV · JSON · MDDrop files here or click to choose
Products · ingredients/proof points · segments · needs · channels · keywords · funnel · metrics · social themes/terms · brand brief (.md) · workspace/grammar (.json). Google Ads, Naver and GA4 export headers are recognised.
Uploaded files & column mapping
2 · Brand profile
Invariants · claim policy3 · Readiness check
Signal coverageSet the target
Audience-Intent StateExplore the brand map — click a node to set the target
Provenanced Brand Knowledge GraphData sources feeding this workspace
Governed Signal SetContent genome — the 12 dimensions of this version
Versioned Targeted Content GenomeSet the taste range
Taste Envelope EngineBrand non-negotiables (from your data)
Sliders are bounded to the envelope; the gate enforces them.
Rules for the creative
Taste, Claims & Exclusion RegistryPrompt package
Grok Imagine Prompt CompilerPrompt checks
Open Grok Imagine ↗Control and variant share one seed; only the selected dimension differs.
Grok Imagine
Image creationGenerated images
Review generated images before publishing. Copy scores do not assess the image.
Variants
Generated Variant StreamHow the gate scores
Multimodal Taste & Evidence GateScore weights (learned from outcomes)
Pass rules
Experiments
Outcome Event StreamResults & learning
Observed Outcomes
Causal Attribution
Bounded Weight Update
Learning rate 0.15 · max ±5 points per update · priors feed the Phase 2 slider markers and the Phase 1 channel recommendation.