---
schema_version: "newruntime-agent-readable-v0.1"
type: "raw_signal"
id: "tg-2625"
slug: "agent-memory-evaluated-as-system-behavior"
title: "Agent memory is being evaluated as system behavior"
description: "MemoryData evaluates what an agent stores, retrieves, updates, and forgets across a sequence instead of grading one final answer."
observed_at: "2026-07-06"
why_it_matters: "Useful memory is a lifecycle with write and invalidation decisions, so point-in-time retrieval metrics cannot establish whether an agent learns safely over time."
novelty: "structural"
verification_level: "source-inspected"
signal_type: "field-report"
evidence_kind: "mixed"
status: "published"
telegram_message_id: 2625
telegram_url: "https://t.me/qwgai/2625"
topics: ["agent-memory","evals","knowledge-systems"]
entities: ["MemoryData"]
related_patterns: []
source_urls: ["https://huggingface.co/papers/2606.24775","https://github.com/OpenDataBox/MemoryData"]
import_batch: "telegram-2026-07-17-v1"
routes: {"html":"https://newruntime.com/signals/agent-memory-evaluated-as-system-behavior/","markdown":"https://newruntime.com/signals/agent-memory-evaluated-as-system-behavior.md","json":"https://newruntime.com/signals/agent-memory-evaluated-as-system-behavior.json"}
source_format: "telegram-export-normalized-json"
---

# Agent memory is being evaluated as system behavior

## Observation

MemoryData evaluates what an agent stores, retrieves, updates, and forgets across a sequence instead of grading one final answer.

## Why it matters

Useful memory is a lifecycle with write and invalidation decisions, so point-in-time retrieval metrics cannot establish whether an agent learns safely over time.

## Provenance

Normalized from QWG AI Telegram message 2625. The original Russian-language record remains available at https://t.me/qwgai/2625.