Forge Pool

Appearance

Compute Adapter Example ​

Overview ​

This example shows a compute-participating adapter.

It accepts domain input, maps it into a primitive request, delegates execution, and shapes the result.

This is the standard pattern for a single-stage execution adapter.

When To Use This Pattern ​

Use this pattern when the adapter’s main job is:

  • mapping domain intent to one primitive
  • building one canonical execution request
  • returning one primary result surface

Typical examples:

  • ETA execution
  • simple risk simulation
  • climate ensemble call
  • one-stage search flow

Lifecycle ​

text
Input → Validate → Transform → Plan → Execute → Finalize → Output

Example Use Case ​

A logistics client wants a probabilistic ETA instead of a single deterministic arrival estimate.

The adapter:

  • receives transport context
  • validates the request
  • builds an mc@1 execution request
  • delegates compute
  • returns ETA distribution and deadline probability

Example Flow ​

text
Transport input
    ↓
ETA adapter
    ↓
mc@1 / eta.v1
    ↓
distributed execution
    ↓
ETA adapter
    ↓
ETA distribution output

Example Request ​

json
{
  "baseline_seconds": 14400,
  "traffic": {
    "mean_delay_seconds": 900,
    "stdev_seconds": 420
  },
  "weather": {
    "enabled": true,
    "slowdown_probability": 0.22,
    "penalty_seconds_mean": 540
  },
  "deadline_at": "2026-04-05T18:00:00Z",
  "iterations": 5000000,
  "quantiles": [0.5, 0.9, 0.95],
  "seed": 19754221,
  "verify": true
}

Example Primitive Request Built by the Adapter ​

json
{
  "ctx": {
    "job_id": "eta-job-001",
    "run_id": "run-1"
  },
  "op": {
    "name": "mc",
    "version": 1,
    "profile": "eta.v1"
  },
  "args": {
    "iterations": 5000000,
    "baseline_seconds": 14400,
    "traffic_mu": 900,
    "traffic_sigma": 420,
    "weather_enabled": true,
    "weather_slowdown_probability": 0.22,
    "weather_penalty_mean": 540,
    "deadline_at": "2026-04-05T18:00:00Z",
    "quantiles": [0.5, 0.9, 0.95]
  },
  "seed": {
    "mode": "explicit",
    "value": "19754221"
  },
  "policy": {
    "verify": "spotcheck"
  }
}

Example Output ​

json
{
  "adapter_name": "eta-adapter",
  "adapter_version": "1.0.0",
  "eta_distribution": {
    "p50_seconds": 15120,
    "p90_seconds": 16184,
    "p95_seconds": 16642
  },
  "deadline_hit_probability": 0.73,
  "risk": "YELLOW",
  "replay_refs": {
    "primitive": "mc@1",
    "profile": "eta.v1",
    "seed": "19754221"
  }
}

What the Adapter Actually Does ​

The adapter is responsible for:

  • validating transport input
  • normalizing the request
  • building the canonical primitive request
  • delegating execution
  • shaping the returned result surface

The adapter is not responsible for:

  • Monte Carlo execution
  • result truth
  • stochastic engine behavior
  • primitive semantics

Pseudocode ​

text
receive transport request
validate request
map transport request to eta.v1 primitive args
send canonical request to Forge execution surface
receive primitive result
shape output for logistics consumer
return output

Why This Pattern Matters ​

This is the cleanest example of:

domain-specific intent mapped into canonical compute truth

It demonstrates the exact relationship between adapters and primitives.

Invariants Satisfied ​

This pattern preserves:

  • explicit primitive selection
  • visible execution boundary
  • no shadow compute
  • preserved traceability
  • no semantic override

Mental Model ​

The adapter asks the right question.

Forge computes the answer.

Next ​