Datasets:

Multilingual-Multimodal-NLP
/

AutoMemoryBench

---
language:
- en
pretty_name: AutoMemoryBench
size_categories:
- 100K<n<1M
task_categories:
- question-answering
- text-retrieval
tags:
- benchmark
- agent-memory
- memory-systems
- llm-agents
- evaluation
- text
configs:
- config_name: default
  data_files:
  - split: audit_subset
    path: data/viewer/audit_subset_probes.jsonl.gz
  - split: public_dev
    path: data/viewer/public_dev_probes.jsonl.gz
  - split: public_test
    path: data/viewer/public_test_probes.jsonl.gz
---

<div align="center">

# AutoMemoryBench

### State-Contract Evaluation for Auditable Agent Memory

[![Paper](https://img.shields.io/badge/Paper-PDF-b31b1b.svg)](AutoMemoryBench-arXiv.pdf)
[![Source](https://img.shields.io/badge/arXiv-Source-b31b1b.svg)](AutoMemoryBench-arXiv-source.tar.gz)
[![GitHub](https://img.shields.io/badge/Code%20%26%20Data-GitHub-181717.svg)](https://github.com/wuyuVerse/AutoMemoryBench)
[![License](https://img.shields.io/badge/Data-Apache--2.0-2f7d32.svg)](https://github.com/wuyuVerse/AutoMemoryBench/blob/main/LICENSE)

</div>

AutoMemoryBench evaluates whether an agent uses the *right* memory, and only
the admissible memory, under a query-time state contract. Each executable
contract partitions memory into **required**, **admissible**, and
**prohibited** sets. Prohibited memories are typed as superseded, deleted,
restricted, cross-namespace, or stale-tool.

<p align="center">
  <img src="assets/query-time-admissibility.png" alt="Query-time memory admissibility" width="760">
</p>
<p align="center"><em>Relevance is not enough: remembered evidence must also be allowed now.</em></p>

## Benchmark at a Glance

| Property | Value |
| --- | ---: |
| Domains | 8 |
| Case variants | 7,200 |
| Total probes | 151,200 |
| Public probes | 120,960 |
| Withheld hidden-test probes | 30,240 |
| Probe types | 20 |
| Memory systems evaluated | 18 |
| Agent/CLI controls | 12 |
| Backbones | Up to 8 |

The eight domains cover personal assistance, education and tutoring, customer
support, DevOps workflows, research assistance, coding agents, multi-party
collaboration, and office collaboration. Minimal single-axis interventions
change exactly one admissibility dimension while preserving the surrounding
case.

## Benchmark Design

AutoMemoryBench compiles a chronological event graph into query-conditioned
memory contracts. The same benchmark case is exposed through an interaction
trace view for the system under test and a contract view used only by the
evaluator. This separates task success from prohibited-memory influence without
revealing the gold state to the tested system.

<p align="center">
  <img src="assets/benchmark-framework.png" alt="AutoMemoryBench evaluation pipeline" width="100%">
</p>
<p align="center"><em>From event graphs to executable state contracts, paired interventions, and hard-gated scoring.</em></p>

### One Case, Two Views

<p align="center">
  <img src="assets/two-view-evaluation.png" alt="Interaction trace and compiled state-contract views" width="900">
</p>
<p align="center"><em>One benchmark case is evaluated through an interaction view and a hidden contract view.</em></p>

The interaction view contains the user turns, memory operations, tool calls,
tool results, and observed answer or action. The compiled contract view records
what the current query must use, may use, and must not use under the active
time, role, namespace, lifecycle, and tool state.

## Public Data

The Dataset Viewer exposes all **120,960 public probes** through three splits:
`audit_subset`, `public_dev`, and `public_test`. Each row exposes the query,
probe type, task type, required and prohibited memory IDs, expected behavior,
scoring rule, and state-contract identifier. This flattened view is intended
for fast browsing and filtering; the canonical nested cases remain in the
downloadable domain shards.

The complete public release is hosted directly in this repository:

| Split | Cases | Probes | Download |
| --- | ---: | ---: | --- |
| `audit_subset` | 720 | 15,120 | [archive](data/full/audit_subset_shards.tar.gz) |
| `public_dev` | 1,440 | 30,240 | [archive](data/full/public_dev_shards.tar.gz) |
| `public_test` | 3,600 | 75,600 | [archive](data/full/public_test_shards.tar.gz) |
| **Public total** | **5,760** | **120,960** | [checksums](data/SHA256SUMS) |

Each archive contains one canonical JSON shard per domain. The original sample
manifest and domain shards are under [`data/sample`](data/sample). The same
release is versioned with the evaluator in the
[GitHub repository](https://github.com/wuyuVerse/AutoMemoryBench). The
30,240-probe hidden-test split is withheld to prevent overfitting.

```bash
# Download the three public split archives from Hugging Face.
hf download Multilingual-Multimodal-NLP/AutoMemoryBench \
  --repo-type dataset --include "data/full/*" --local-dir AutoMemoryBench-data
```

### Canonical Case Fields

| Field | Description |
| --- | --- |
| `case_id` | Stable identifier for a benchmark case variant |
| `domain` | Application domain of the interaction |
| `events` | Chronological lifecycle, authorization, and tool events |
| `sessions` | User and assistant interaction trace |
| `gold_memory_units` | Auditable memory objects compiled from events |
| `state_contracts` | Query-time admissibility partitions and transitions |
| `queries` | Memory-required and diagnostic probes with scoring contracts |

## Evaluation

- **StrictCore** is the primary hard gate. A memory-required query counts only
  when the task is solved, required evidence is used, and no prohibited memory
  influences the output.
- **PairAcc** measures whether both sides of a minimal intervention pair pass
  StrictCore.
- **LeakRate** measures how often prohibited memory appears in an answer or
  trace.
- **AMQ** is a diagnostic composite across writing, retrieval, update,
  compression, task utility, safety, and efficiency; it is not the headline
  metric.

The canonical evaluator is implemented in
[`amb/benchmark/evaluation`](https://github.com/wuyuVerse/AutoMemoryBench/tree/main/amb/benchmark/evaluation)
and
[`amb/benchmark/metrics`](https://github.com/wuyuVerse/AutoMemoryBench/tree/main/amb/benchmark/metrics).

## Paper

- [AutoMemoryBench paper PDF](AutoMemoryBench-arXiv.pdf)
- [Submission-ready arXiv source package](AutoMemoryBench-arXiv-source.tar.gz)
- [SHA256 checksums](SHA256SUMS)

The formal Hugging Face Paper page and arXiv tag will be added after the paper
receives an arXiv identifier.

## Authors

Jiajun Wu, Jian Yang, Chen Li, Linzheng Chai, Ge Gao, Ensheng Shi, and Yuchi Ma.

Jiajun Wu and Jian Yang are affiliated with Beihang University. Ensheng Shi and
Yuchi Ma are affiliated with Huawei Cloud Computing Technologies Co., Ltd.

## Citation

```bibtex
@misc{automemorybench,
  title  = {AutoMemoryBench: State-Contract Evaluation for Auditable Agent Memory},
  author = {Wu, Jiajun and Yang, Jian and Li, Chen and Chai, Linzheng and
            Gao, Ge and Shi, Ensheng and Ma, Yuchi},
  year   = {2027},
  url    = {https://github.com/wuyuVerse/AutoMemoryBench}
}
```

## License

The code and benchmark data are released under Apache-2.0. The paper license
will follow the license selected for the arXiv submission.