Recipe — Judgment Augment

The drop-in 'add intelligence-grade judgment to my workflow' recipe. One HTTP call adds the right substrate primitive bundle automatically.

What you'll build

A workflow that takes any data — text, JSON record, freeform input — and adds calibrated, audit-trail-ready judgment to it before passing the data to a downstream side-effect (Slack, email, CRM update, alert).

Three minutes from "I have a workflow" to "my workflow now ships with intelligence-grade judgment."

Step 1 — Get an API key

curl -sX POST https://api.leadershipunderuncertainty.org/api/mcp/generate-key

Returns a free-tier key (nb_* prefix). 25 calls/day, 500 per month, no credit card required.

Step 2 — Pick a workflow shape

Look at what your workflow's terminal output is, and pick the closest shape:

  • summarization — text in, summary out (digests, news recaps, brief generators)
  • enrichment — record in, record + context out (CRM enrichment, lead scoring)
  • monitoring — current+prior state in, alert out (drift detection, threshold monitors)
  • decision — question + options in, recommendation out (vendor pick, approval gate)
  • generation — LLM-generated content (memos, drafts, forecasts)
  • routing — classification (ticket triage, queue assignment)
  • evaluation — predictions vs outcomes (forecast scoring)
  • generic — fallback when no shape clearly fits

Step 3 — Call the augment endpoint

curl -X POST https://api.leadershipunderuncertainty.org/api/v1/augment \
  -H "Authorization: Bearer nb_..." \
  -H "Content-Type: application/json" \
  -d '{
    "value": "Quarterly market summary text...",
    "intent": "Summarize the market update for the executive team morning brief.",
    "workflowShape": "summarization",
    "options": {
      "confidenceGate": 0.6,
      "attestation": "hash",
      "maxCredits": 15
    }
  }'

Step 4 — Read the response

The augment endpoint returns a JudgmentEnvelope-shaped response:

{
  "value": "<original input, unchanged>",
  "augmented": {
    "claims": [{"text": "...", "confidence": 0.7, "type": "fact"}],
    "sources": [{"identifier": "...", "type": "news_article"}],
    "probabilityLanguage": {"score": 70, "issues": [], "suggestions": []}
  },
  "report": {
    "workflowShape": "summarization",
    "primitivesExecuted": [
      {"primitive": "extract_claims", "status": "ok", "creditCost": 5},
      {"primitive": "extract_sources", "status": "ok", "creditCost": 4},
      {"primitive": "score_probability_language", "status": "ok", "creditCost": 2}
    ],
    "substrateCoverage": {
      "populated": ["Claim", "Source", "UncertaintyEstimate"],
      "missing": ["Evidence", "Alternative", ...],
      "score": 3
    },
    "totalDurationMs": 4260,
    "totalCreditCost": 11,
    "confidence": {"mean": 0.7, "lowConfidenceFields": []},
    "downstreamGuidance": "Surface the score in your downstream system; flag any claim with confidence < 0.5 for human review.",
    "issues": []
  },
  "attestation": {"mode": "hash", "hash": "8b9c5c7a..."},
  "traceId": "trace-XYZ"
}

Step 5 — Wire it into your workflow

Insert the augment call right before your workflow's terminal output step. The original value flows through unchanged; you now also have augmented and report to use in the downstream step.

Per shape, surface specific fields:

  • summarization — append claims[].text and sources[].identifier to the digest
  • enrichment — write augmented.sourceCredibility[0].composite as a quality score
  • monitoring — fire alert only if analyticDelta.significance is major or critical
  • decision — block auto-action when confidenceGate.passes === false
  • generation — flag any assumptions[i].criticality === 'linchpin' for review

Distribution paths

You can do step 3 above from any of these surfaces:

  • REST — direct curl as shown
  • SDK — new Augment({ apiKey }).run({...})
  • MCP — luu_augment tool from Claude Desktop, Cursor, ChatGPT, etc.
  • n8n — import packages/n8n-templates/judgment-augment/workflow.json
  • Zapier — import packages/zapier-templates/judgment-augment/recipe.json
  • Google Agent Designer — import docs/distribution/flow-templates/judgment-augment.json
  • Lindy — follow docs/distribution/install-judgment-augment-lindy.md
  • Claude Code template — Judgment Co-pilot reads any workflow.json and rewrites it with augmentation injected

See also