M5: Varitea closed-loop acceptance test — E2E proof the full architecture works

[ninan-versioning]

Milestone 5 — Dogfood Closed-Loop Proof

Vertical: enterprise
Revenue-Signal: protocol-adoption
Customer-Type: all

What

Define and pass a single end-to-end acceptance test that proves the architecture loop closes for Drink Varitea: catalog → twin → render → publish → ad → signal → brand_rules update → next-gen render. This is the gate every other milestone is in service of.

Scope

Acceptance sequence:

1. UCP catalog_lookup on Varitea storefront returns ≥6 SKUs
2. Each SKU has a populated twin (2D wedge path acceptable in M9)
3. brand_rules.json populated with all 6 spatial primitives + soul.md compiled
4. 6-variant creative pack rendered, passes Kinfolk regression + policy gate
5. Pack pushed to Shopify PDP + 1 Meta ad set live
6. 72-hour observation window collects CVR, CTR, ROAS per variant
7. Thompson Sampling produces a winning variant, commits to brand_rules
8. Next render cycle uses the updated brand_rules — variant drift verified

Why

Every existing ticket says "build X." This one says "prove X works together on the actual dogfood." Without this, milestones complete in isolation and the loop never demonstrably closes. This is also the only artifact that converts an investor demo from claim to proof.

Acceptance criteria

• One reproducible script: npm run varitea:e2e runs the full loop in CI
• Test passes with a real Meta spend < $50
• Output report committed to docs/varitea-closed-loop-{date}.md
• Recorded screen capture used in App Store demo + investor deck

Affected repos

ALL FOUR — shoploop (orchestrator), shoploop-core (twin + rules), shoploop-shopify (ingest + publish), shoploop-growth (ad + signal)

Cost

< $50 Meta ad spend per test run. AWS infra reuses existing.