Measure / Snapshot

/v1/measure and /v1/operator/snapshot turn a guide loop into product evidence and operator replay.


guide + feedback + task evidence -> measure -> snapshot

`POST /v1/measure`

Reports whether memory changed future context or behavior, without claiming more than the provided evidence supports.

SDK


const measure = await aionis.execution.measureRun({
  run_id,
  task_signature,
  after_guide: guide,
  feedback_result: feedback,
  sufficient_evidence: true,
});

HTTP


curl -s http://127.0.0.1:3001/v1/measure \
  -H 'content-type: application/json' \
  -d '{
    "tenant_id": "default",
    "scope": "checkout-migration",
    "task": {
      "run_id": "run-001",
      "task_signature": "checkout-migration"
    },
    "product_trace": {
      "after_guide": {
        "guide_trace_id": "guide_trace:..."
      },
      "feedback_result": {
        "outcome": "positive"
      },
      "sufficient_evidence": true
    }
  }'

What Measure Can Say

Signal	Meaning
`effect_report`	Whether the evidence supports a product effect claim.
`memory_decision_trace`	How memory was admitted, blocked, or rehydrated.
`memory_decision_audit`	Compact operator audit of decisions.
insufficient evidence	Aionis refuses to overclaim when the loop is too thin.

Trace-Derived Skill Candidates

When measure sees enough positive continuity or workflow-reuse evidence, effect_report.training_candidates may include:


{
  candidate_type: "trace_derived_skill",
  export_ready: true,
  trace_derived_skill: {
    contract_version: "aionis_trace_derived_skill_candidate_v1",
    skill_name: "Continue verified execution state across sessions",
    applies_when: ["task_signature:checkout-migration"],
    does_not_apply_when: ["A newer memory contests the source trace."],
    procedure_steps: ["Recover the current Aionis guide before continuing."],
    acceptance_checks: ["effect_kernel_passed"],
    authority_state: "candidate",
    export_policy: {
      agent_prompt_included: false,
      runtime_mutation: false,
      required_gate: "admission_and_promotion_gate"
    }
  }
}

This field is for controlled learning, review, and promotion workflows. It is not added to agent_context, and it does not mutate memory rows by itself.

SDK helpers:

Helper	Use
`traceDerivedSkillCandidatesFromMeasure(measure)`	Return the raw Runtime candidate contracts.
`traceDerivedSkillReviewItemsFromMeasure(measure)`	Return compact review items for dashboards, approval queues, or host-side promotion workflows.

The review helper is still read-only. It preserves the same safety contract: authority_state: "candidate", agent_prompt_included: false, and runtime_mutation: false.

`POST /v1/operator/snapshot`

Builds a read-only operator view over guide, feedback, effect, execution state, and trace-to-procedure projections.

SDK


const snapshot = await aionis.execution.snapshotRun({
  run_id,
  task_signature,
  guide,
  measure_result: measure,
  include_markdown: true,
});

HTTP


curl -s http://127.0.0.1:3001/v1/operator/snapshot \
  -H 'content-type: application/json' \
  -d '{
    "tenant_id": "default",
    "scope": "checkout-migration",
    "run_id": "run-001",
    "task_signature": "checkout-migration",
    "guide_packet": {},
    "memory_decision_trace": {},
    "effect_report": {},
    "include_markdown": true
  }'

Boundary

Measure and snapshot are read-side product surfaces. They do not run tools, execute verifiers, mutate external memory, or prove benchmark-level success by themselves.

Use them to answer:

Did Aionis expose useful memory?
Did Aionis block unsafe memory?
Was feedback attributed to the memory actually used?
What would an operator need to debug the decision?