Forge Pool

Appearance

Ensemble ​

ensemble@1 is the Forge Pool primitive family for reducer-centric multi-run execution.

It is designed for workloads where the output is derived from the controlled combination of multiple runs, models, strategies, or candidate result surfaces.

What ensemble@1 is For ​

Typical uses include:

  • confidence fusion
  • consensus generation
  • disagreement measurement
  • multi-model comparison
  • scenario stacking
  • stability and instability scoring

Kernel Role ​

Ensemble workloads are executed through the same canonical contract as other families.

http
POST /api/v0/ops/execute

The difference is not the route. The difference is the family semantics:

  • multiple component runs
  • reducer-driven aggregation
  • confidence-aware outputs
  • traceable ensemble composition

Mental Model ​

Where mc@1 asks, "what does this distribution look like under repeated sampling?"

ensemble@1 asks, "what happens when many result surfaces are fused under a deterministic reduction rule?"

Example Identity ​

json
{
  "op": {
    "name": "ensemble",
    "version": 1,
    "profile": "consensus.v1"
  }
}

Expected Output Shapes ​

Ensemble workloads commonly produce:

  • aggregated result surfaces
  • confidence bands
  • disagreement metrics
  • component contribution summaries
  • reducer metadata
  • replay references

Design Position ​

This family is central to Forge Pool's broader execution model because high-stakes systems often require more than one run or one model.

They require deterministic combination.

That is what ensemble@1 is designed to provide.