Step 18 · Hands-on lab · Hands-on lab · Lab: From signals to campaign assets ENPT
Alembic Complete Visual Course

Lab: From signals to campaign assets

Ingest ideas, distill signals, and generate marketing assets with the Marketing Factory.

Read the plain version, or open the technical layer on any section.
1

Data-driven go-to-market

This lab connects ingestion, ETL, council, and marketing. You will:

  1. Create three Markdown idea files in a folder (e.g., features, pain points, testimonials).
  2. Run alembic ingest ./ideas to produce wiki packages.
  3. Run alembic distill ./ideas --from discover --to review --offline to score and verify business signals.
  4. Pick the highest-scoring signal from the output.
  5. Run alembic marketing <signal>.json --offline to generate positioning, copy, and creative manifests.
  6. Open the generated manifest and inspect the rejected/accepted creatives.

Think of it like… a product team that reads customer feedback, ranks opportunities, and hands a brief to the marketing agency.

Under the hood

runFunnel in @alembic/harness chains T0 deterministic scoring, T1 extraction, T2 shortlisting, and T3 verification. The resulting BusinessSignal is fed into runMarketingFactory. Offline mode uses deterministic copy templates and a fake Higgsfield client; online mode shells out to the real Higgsfield CLI and can use --soul-id.

2

In one picture

ingest distill signal marketing manifest
Ideas → wiki packages → verified signals → marketing assets.
3

In the code

# 1. create ./ideas/*.md
mkdir -p ./ideas
cat > ./ideas/pain.md <<'EOF'
# Pain: onboarding drop-off
Users leave during signup because the form is too long.
EOF

# 2. ingest and distill
alembic ingest ./ideas
alembic distill ./ideas --from discover --to review --offline

# 3. market the top signal
alembic marketing ./ideas/distill/signal-*.json --offline

Implementation notes

distill writes scored packages and verified signals under the corpus directory. The marketing command picks the signal, loads marketing skill contexts from <dataDir>/skills/marketing, builds a Higgsfield generation request, and writes a content-addressed manifest. The manifest records rejected creatives with reason codes so you can audit the pipeline.

4

Try it

Create three idea files and run the full lab. Compare the T0 score with the final T3 council score.

Audit: Why does the manifest keep rejected creatives?
5

Quick check

Which component verifies the business signal before marketing runs?