@processengine/dataflows

Runtime for dataflow artifacts in ProcessEngine Flow 5.

@processengine/dataflows validates, prepares and executes linear dataflow artifacts composed of MAPPINGS, RULES and DECISIONS pipeline items. It is the runtime for PROCESS/DATA steps in Flow 5.

Role in ProcessEngine

PROCESS/DATA (flow graph step)
  → processor runtime calls executeDataflow(...)
  → @processengine/dataflows executes dataflow artifact
  → returns DataflowOutput.writes[]
  → @processengine/semantics.reduce applies writes to process state

Install

npm install @processengine/dataflows

Quick start

import { validateDataflow, prepareDataflow, executeDataflow } from '@processengine/dataflows';

// 1. Validate
const source = {
  id: 'dataflow.example',
  version: '1.0.0',
  schema: {
    '$.context.data.facts.result': {
      title: 'Result facts',
      description: 'Facts produced from the input application for the quick start example.',
      fields: {
        ok: {
          type: 'boolean',
          title: 'Input is acceptable',
          description: 'true when the application payload can continue through the example dataflow.'
        }
      }
    }
  },
  pipeline: [{
    id: 'map_input',
    type: 'MAPPINGS',
    kind: 'facts',
    artefactId: 'mappings.example',
    contract: {
      input: { ref: '$.context.input.application' },
      output: { ref: '$.context.data.facts.result' }
    }
  }]
};

const validation = validateDataflow(source);
if (!validation.ok) throw new Error('Invalid');

// 2. Prepare (once, cache the result)
const artifact = prepareDataflow(source);

// 3. Execute
const state = { context: { input: { application: { ok: true } }, data: {} } };
const registries = {
  mappings: {
    get: (id) => ({ id }),
    executeMappings: (_artifact, input) => ({ output: { ok: input.ok } })
  }
};
const result = executeDataflow(artifact, { state, registries });
// result.writes[] — transport-safe, pass directly to semantics.reduce

Design rule

The package is designed from the public contract inward:

• one normative source shape, not several accepted aliases;
• one runtime result contract from neighbour libraries: { output, trace? };
• child runtime trace, if returned, is accepted but not merged into dataflow trace in v1;
• no compatibility branches before the first release;
• no hidden fallback behavior for malformed artifacts;
• validation rejects unsupported fields instead of interpreting them leniently;
• runtime failure trace is observable in DataflowRuntimeError.details.trace when trace is enabled.

For v1 every pipeline item has exactly one input contract:

contract: {
  input: { ref: '$.context.data.payloads.clientComparison' },
  output: { ref: '$.context.data.facts.clientComparison' }
}

If a later version needs composite input assembly, it must be introduced as a new explicit contract, not as an implicit extension of this one.

Schema nodes and fields are part of the human-readable contract. Each schema node and each declared field must have both title and description. This keeps dataflow artifacts usable as code-as-docs and makes the data contract readable in Flow UI, reviews, and business-requirement traceability.

Lifecycle

validateDataflow(source, options?)  → ValidationResult
prepareDataflow(source, options?)   → DataflowArtifact
executeDataflow(artifact, input, options?) → DataflowOutput

Non-goals

@processengine/dataflows does not know the flow graph, does not move currentStepId, does not persist state, and does not own retry/fail policy.

Examples

The examples/ directory contains normative examples:

• happy-path.json — minimal valid dataflow artifact;
• failing-path-missing-input.json — valid artifact that demonstrates runtime missing-input behavior;
• interop-read-after-write.json — MAPPINGS → DECISIONS read-after-write interop.

Documentation

See SPEC_RU.md for full normative specification. See COMPATIBILITY.md for compatibility guarantees. See MIGRATION.md for migration from Flow 3.