Datasets:

MatMaster-DP
/

Materials-AutoLab-144-evaluator-access

language:
  - en
pretty_name: Materials AutoLab 144 Evaluator Access
size_categories:
  - n<1K
task_categories:
  - question-answering
tags:
  - materials-science
  - agent
  - laboratory-automation
  - private-evaluation
gated: manual
extra_gated_heading: Request access to Materials AutoLab 144 evaluator references
extra_gated_description: >-
  Access is intended for benchmark maintainers who will isolate Gold action
  references and target states from evaluated Agents.
extra_gated_prompt: Approved users may use the versioned Gold package for self-hosted evaluation.
extra_gated_fields:
  Full name: text
  Affiliation: text
  Intended evaluation use: text
  Agent or model to be evaluated: text
  Gold isolation and anti-cheating plan: text
  Planned publication of aggregate results: text
  I understand that the evaluated Agent must not access Gold actions or target states: checkbox
  I understand that access may be revoked but downloaded copies cannot be recalled: checkbox

Materials AutoLab 144 Evaluator Access

This manual-gated repository provides the versioned evaluator reference package for MatMaster-DP/Materials-AutoLab-144. Each of the 144 records contains the semantic action DAG, target terminal state, task identity, and source-binding hashes used by an evaluator operator.

Source release commitment:

c7274098a514db74398f2e27b42a48e31286f322ec97d64acb903a4fc0ec1095

Repository contents

Materials-AutoLab-144-evaluator-access/
├── README.md
├── ACCESS_POLICY.md
├── GOLD_MANIFEST.json
├── manifest.json
└── answers/
    └── hb-alchemybench-<task-id>/gold_design.json

Path	Description
`GOLD_MANIFEST.json`	Canonical index and SHA-256 for all 144 Gold records.
`answers/<task>/gold_design.json`	Semantic action DAG, target terminal state, scenario identity, and source-binding hashes.
`ACCESS_POLICY.md`	Access, isolation, and reporting requirements.
`manifest.json`	Machine-readable package identity and source commitment.

Evaluator operators must keep this repository and its access token outside the Agent environment. Use the public task package as the Agent-visible input and freeze each submission before comparison with these references.