Skip to content

SHOWCASE_STANDARD.md

Latest commit

 

History

History
33 lines (29 loc) · 1.93 KB

SHOWCASE_STANDARD.md

File metadata and controls

33 lines (29 loc) · 1.93 KB
Raw

Runnable showcase standard

New end-to-end showcases must be reproducible workloads, not timing simulations. They follow this contract:

  1. Correct domain ownership. Store the example under its technical domain (generative-ai, document-ai, media, rendering, and so on), never a marketing category.
  2. Offline execution. run.sh must complete without network access. Small fixtures belong in data/; large public datasets may have an explicit preparation script, checksum, license note, and deterministic local fallback.
  3. Observable stages. Each meaningful computation is a separate AkôFlow activity. Parallel work is expressed with dependsOn, and activities exchange artifacts only through the execution workspace.
  4. Real computation. Stages process data with the declared toolchain. Do not use sleep or precomputed outputs to imitate work.
  5. Portable smoke profile. The default run is CPU-compatible and small enough for CI. AKOFLOW_PROFILE=demo, standard, or stress may increase dataset size, frame count, model size, or repetitions without changing the DAG.
  6. Stable result contract. A successful run writes outputs/manifest.json with showcase, status, and artifacts. validate.sh checks semantics and non-empty artifacts, not only process exit status.
  7. Container parity. docker/Dockerfile contains every runtime dependency and invokes the same stage runner used by run.sh.
  8. Complete AkôFlow inputs. Every showcase includes workflow.yaml, environment.yaml, scope.yaml, and topology.yaml plus a concise README.
  9. Safe optional integrations. Hosted APIs, Ollama, GPU acceleration, and external storage are opt-in adapters. Secrets never appear in fixtures, commands, logs, or committed output.

The five reference implementations are AI Video Studio, Document Intelligence, Local RAG Factory, Video Intelligence, and Blender Render Farm.