Ensemble
Overview
Ensemble is one of the core execution capability families provided by the Forge runtime.
This section contains canonical verification examples for Ensemble capabilities implemented by Forge. Every example is derived from maintained Primitive Profiles and executed through the production compute pipeline.
These documents are not decision-making tutorials.
They are executable capability specifications that allow evaluators to independently verify multi-model execution, inspect runtime evidence, validate replay, and understand deterministic consensus generation.
Ensemble execution evaluates multiple independent reasoning paths before producing a consolidated result.
Primitive Profile
Ensemble capabilities are implemented through the ensemble@1 primitive.
Each documented capability corresponds to a versioned Primitive Profile registered in the Forge execution registry.
Primitive
ensemble
Version
1
Profile
<profile-id>Representative profiles include:
finance.credit.ifrs9_review_board.v1insurance.board.risk.consensus.v1
Each profile defines its own decision semantics while sharing the same execution primitive.
Capability
Ensemble capabilities coordinate multiple independent execution paths before producing a deterministic consensus result.
Typical execution models include:
- review boards
- expert panels
- committee decisions
- consensus scoring
- disagreement analysis
- confidence aggregation
- recommendation synthesis
Unlike Monte Carlo, Search, or Graph primitives, Ensemble execution evaluates multiple reasoning perspectives before constructing a final consensus artifact.
The primitive verifies the consensus process rather than any individual opinion.
Canonical Contract
Execution uses the canonical Forge execution contract.
Primitive : ensemble
Version : 1
Profile : <profile-id>Requests are validated by the Canonical Execution Validator before entering the compute runtime.
The canonical payload contains:
op.name
op.version
op.profile
argsOptional execution metadata includes:
ctx
seed
policyThe participant configuration and consensus strategy are profile-specific.
Required Inputs
Ensemble profiles generally require:
- participant configuration
- execution objective
- consensus strategy
- decision context
Additional required fields depend on the selected Primitive Profile.
Execution cannot begin until canonical validation succeeds.
Optional Inputs
Supported optional inputs vary by profile.
Common examples include:
- participant weights
- confidence thresholds
- disagreement policy
- evidence references
- replay configuration
- execution metadata
Only documented profile fields should be submitted.
Canonical Smoke
A canonical Ensemble smoke verifies:
- profile discovery
- contract inspection
- successful validation
- participant initialization
- consensus execution
- runtime evidence generation
- replay metadata generation
- artifact availability
The maintained Smoke Suite remains the canonical executable source.
Verification Expectations
A successful Ensemble execution should demonstrate:
- primitive resolution
- profile resolution
- canonical validation
- participant execution
- deterministic consensus generation
- runtime evidence generation
- replay metadata availability
- artifact generation where supported
Verification should include inspection of both the consensus result and the consensus process.
Runtime Evidence
Successful Ensemble execution exposes observable runtime evidence including:
- execution identifier
- primitive profile
- execution metadata
- participant count
- consensus summary
- agreement metrics
- disagreement metrics
- confidence values
- replay metadata
- artifact references
The exact evidence surface depends on the selected profile.
Replay
Ensemble profiles support deterministic replay when executed using the same execution contract, runtime version, participant configuration, and deterministic seed where applicable.
Replay confirms reproducible consensus behaviour rather than infrastructure identity.
Artifacts
Typical Ensemble artifacts include:
- consensus report
- participant summaries
- agreement analysis
- disagreement analysis
- confidence summary
- execution summary
- replay metadata
Artifacts provide observable evidence explaining how the final consensus was produced.
Applied Intelligence Modules
Ensemble capabilities are reused across multiple Forge Intelligence Modules including:
- Credit Intelligence
- Banking Intelligence
- Insurance Intelligence
- Reinsurance Intelligence
- Treasury Intelligence
- Capital Intelligence
The computational primitive remains identical while each Primitive Profile defines domain-specific consensus semantics.
Related Documentation
- /examples/
- /verification/
- /verification/capability-verification
- /verification/replay-determinism
- /verification/artifact-inspection
- /api/kernel/execution-model
Verification Checklist
| Verification Surface | Status |
|---|---|
| Primitive resolved | ✓ |
| Profile resolved | ✓ |
| Contract inspected | ✓ |
| Required inputs identified | ✓ |
| Optional inputs bounded | ✓ |
| Canonical smoke available | ✓ |
| Runtime evidence inspected | ✓ |
| Artifacts inspected | ✓ |
| Replay verified | ✓ |
Final Principle
Ensemble capabilities are verified as deterministic consensus-execution contracts.
They demonstrate value not by producing a single opinion, but by exposing how multiple independent execution paths are evaluated, reconciled, and transformed into a reproducible consensus through the canonical execution contract.