Feedback Attribution

Aionis does not treat every successful run as proof that every retrieved memory helped.

Feedback attribution links four facts:


guide exposure -> host use -> outcome -> memory lifecycle evidence

What The Host Must Track

Field	Why
`guide_trace_id`	Identifies the exact guide exposure.
exposed memory IDs	Defines what memory was available to the Agent.
actually used memory IDs	Prevents false credit for unused recall.
`run_id`	Links feedback to a real execution run.
outcome and verifier signal	Distinguishes successful reuse from harmful reuse.

Why Aionis Is Strict

Feedback is rejected when the host attributes outcome to memory that was not exposed by the referenced guide. This is intentional. Otherwise the dataset would mix real usage with guesses.


import { feedbackFromGuide } from "@aionis/sdk";
 
const usedMemoryIds = result.usedMemoryIds ?? [];
 
if (usedMemoryIds.length > 0) {
  await aionis.feedback(feedbackFromGuide({
    guide,
    run_id,
    outcome: "positive",
    reason: "The Agent used the active path and verifier passed.",
    used_memory_ids: usedMemoryIds,
    used_surface: "use_now",
  }));
}

Positive, Negative, Neutral

Outcome	Meaning
`positive`	The memory helped or was consistent with a successful run.
`negative`	The memory contributed to a bad or rejected outcome.
`neutral`	The memory was observed but effect is unclear.

Negative feedback does not automatically delete memory. It becomes lifecycle evidence that can demote, suppress, or route memory to inspect-first.

What This Enables

Feedback attribution powers:

memory decision traces
memory use receipts
admission dataset export
future learned admission policy
Agent Flight Recorder replay
controlled forgetting and reactivation