README.md

metadata

pretty_name: FalseMemBench
license: mit
task_categories:
  - text-retrieval
language:
  - en
tags:
  - retrieval
  - memory
  - llm-agents
  - adversarial
size_categories:
  - n<1K

FalseMemBench

FalseMemBench is a benchmark project for evaluating memory retrieval systems under adversarial distractors.

The goal is to measure whether a system can retrieve the right memory when many nearby but wrong memories are present.

Focus

The benchmark is designed for memory systems used by LLM agents.

It emphasizes:

• entity confusion
• environment confusion
• time/version confusion
• stale facts vs current facts
• speaker confusion
• near-duplicate paraphrases

Layout

• schema/case.schema.json: benchmark case schema
• data/cases.jsonl: current benchmark cases
• docs/: benchmark design notes
• scripts/validate.py: schema validator for the JSONL dataset
• scripts/run_benchmark.py: simple keyword baseline
• scripts/run_tagmem_benchmark.py: run the benchmark against a real tagmem binary

Case format

Each case contains:

• a query
• a set of entries
• one or more relevant_ids
• a single adversary_type
• optional metadata for analysis

Example

{
  "id": "env-001",
  "query": "What database does staging use?",
  "adversary_type": "environment_swap",
  "entries": [
    {
      "id": "e1",
      "text": "The staging environment uses db-staging.internal.",
      "tags": ["staging", "database", "infra"],
      "depth": 1
    },
    {
      "id": "e2",
      "text": "The production environment uses db-prod.internal.",
      "tags": ["production", "database", "infra"],
      "depth": 1
    }
  ],
  "relevant_ids": ["e1"]
}

Current adversary types

• entity_swap
• environment_swap
• time_swap
• state_update
• speaker_swap
• near_duplicate_paraphrase

Current dataset size:

• 573 cases

Intended use

The benchmark is intended to be:

• model-agnostic
• storage-agnostic
• metadata-friendly
• easy to publish to GitHub and Hugging Face