Which diagnostic check should run next?

An independently authored, isolated Yumi utility providing a read-only JSON calculation over finite diagnostic models.

Use the endpoint

This application is stateless. Its code does not persist or log submitted models, add tracking, make outbound calls, or inspect a live system. The hosting provider may process ordinary request metadata.

• GET /api/yumi/probe-plan returns the service descriptor.
• GET /api/yumi/probe-plan?model=... accepts one URL-encoded JSON model. GET is also subject to host URL limits; use POST for larger models within the documented body limit.
• POST /api/yumi/probe-plan accepts an application/json request body.

Model shape and limits

The top-level object has exactly two keys: hypotheses and probes. There must be one to twelve hypotheses and zero to twelve probes.

Each hypothesis is exactly { id, weight }. Each probe is exactly { id, cost, outcomes }. Hypothesis IDs are unique within the hypothesis list; probe IDs are unique within the probe list. IDs and outcome strings contain one to 64 UTF-16 code units.

Weights and costs are positive integers from 1 through 1,000,000. Weights are relative weights, not inferred probabilities. Costs use caller-defined units.

Every probe must provide a complete deterministic outcome mapping for every hypothesis. Each GET model and POST body is limited to 65,536 UTF-8 bytes; the POST limit is enforced while the stream is read. Submit only public, non-secret models.

Results

The integer numerator and denominator represent the exact expected-cost ratio for the supplied model. Decimal fields are approximate renderings. The response also reports equivalence groups for hypotheses that remain undistinguishable under the supplied outcomes.

When several choices have the same exact value, the stable tie rule follows input order. The planner does not infer real-world truth or real-world costs, and it makes no learning, external-model, AGI, personality, or adoption claim.

Compact model example

{
  "hypotheses": [
    { "id": "network", "weight": 2 },
    { "id": "storage", "weight": 1 }
  ],
  "probes": [
    {
      "id": "ping",
      "cost": 1,
      "outcomes": {
        "network": "timeout",
        "storage": "ok"
      }
    }
  ]
}