4.7 Multiple candidates and selection

Generate multiple candidates → evaluate individually → select or combine → verify and deliver. The example creates distinct product-copy options from a real catalog and preserves candidates, grades, exclusion reasons and fusion lineage.

flowchart TD
  L[Main Loop] --> S[Load catalog and Memory]
  S --> G[Generate candidate batch]
  G --> D[Hard checks and duplicate filtering]
  D --> E[Evaluate each candidate]
  E --> R[Rank qualified candidates]
  R -->|select| P[Publish unchanged winner]
  R -->|fuse| F[Compose fields from top two]
  F --> V[Fresh checks and evaluation]
  V -->|pass| P
  V -->|reject| B[Allowed fallback or human handoff]
  R -->|none qualified| H[Human handoff]

One main Loop schedules flat stage Graphs through graphStep. Graphs contain public Worker nodes; arrows show Loop decisions. Context uses Redis, Memory uses file-backed SQLite and catalog/artifact tools remain application-owned.

Run

Use Node 24, real Redis, configured ditto.yaml and model credentials in the environment. Artifacts are stored under the ignored .examples-candidates-tasks/ directory.

export DITTO_WORKER_CONTEXT_REDIS_URL=redis://127.0.0.1:6379
npm run example:candidates -- --provider deepseek
npm run example:candidates -- --provider deepseek --mode fuse
npm run example:candidates -- --provider deepseek --mode fuse --no-fallback
npm run example:candidates -- --provider deepseek --stop-after assessment
npm run example:candidates -- --provider deepseek --directory .examples-candidates-tasks/cli-XXXXXX
npm run check:examples:candidates:package -- --provider deepseek

--count requests 2–4 candidates, --score sets the minimum (0–25), --calls limits model attempts and --goal supplies the objective. Fusion is re-evaluated; fallback uses an original candidate only when allowed. output/copy.md is the delivered copy; output/report.json and output/report.md retain comparison evidence and provenance.

API, runnable consumer and recovery contracts · Application tools.