SamuelChien821's picture
Add files using upload-large-folder tool
11de5e5 verified
|
Raw
History Blame Contribute Delete
5.19 kB
metadata
license: other
language:
  - en
pretty_name: Blobfish DomainBench-24
task_categories:
  - question-answering
  - reinforcement-learning
tags:
  - agent-evaluation
  - mcp
  - tool-use
  - trajectories
  - sqlite
size_categories:
  - n<1K
configs:
  - config_name: tasks
    data_files:
      - split: train
        path: data/tasks.jsonl
  - config_name: trajectories
    data_files:
      - split: train
        path: data/trajectories.jsonl
  - config_name: assets
    data_files:
      - split: train
        path: data/assets.jsonl

Blobfish DomainBench-24

DomainBench-24 is the public release record for 24 stateful agent tasks across six professional domains. Every task runs in a checked-in SQLite-backed MCP world with typed read/write tools and deterministic verification.

The release includes:

  • 24 task records across six independent companion worlds;
  • a 24/24 deterministic reference ceiling and a 0/24 seeded-random floor;
  • 12 full sample trajectories: one successful reference replay and one failing floor episode per world;
  • 42 hashed world assets, including the runnable database, schema, seed, tools, task manifest, MCP manifest, and world manifest.
  • a digest-pinned Harbor release whose 24-task Oracle gate passed at reward 1.0 with zero exceptions.

No reference or random control is a model result. There are no frontier-model rankings in this release.

Domains and provenance

The Harbor datasets below are upstream professional-domain evaluation suites. The Blobfish worlds are independent diagnostic environments in the same domains; their 24 tasks are not converted or claimed as subsets of the upstream suites.

Domain Blobfish dataset page Upstream Harbor source Companion world
Financial research FilingBench-50 vals/financeagent world
Customer operations ServiceBench-375 sierra-research/tau3-bench world
Scientific discovery DiscoveryBench-102 scienceagentbench/scienceagentbench world
Defensive security DefenseBench-100 polyvorlabs/cyberdefense-bench world
Enterprise operations OpsBench-14 Enterprise-Bench/l1-l2-bench world
Software engineering RepoBench-500 swe-bench/swe-bench-verified world

Files

  • data/tasks.jsonl — prompts, split labels, required tools, verifier basis, and control outcomes.
  • data/trajectories.jsonl — stable trajectory metadata plus serialized tool steps and verifier results.
  • data/assets.jsonl — byte size, SHA-256, and repository URL for each released asset.
  • evidence/*.json — complete public evidence bundle used by blobfish.ai.
  • worlds/<benchmark>/ — the actual released world files.
  • harbor-release.json — Harbor revision, task-package digests, runtime contract, and Oracle-gate receipt.
  • MANIFEST.json — SHA-256 and byte length for every uploaded file.

Evaluation contract

Each episode starts from a fresh copy of the released SQLite state. The harness records tool calls, arguments, outputs, and errors, then grades the final state with deterministic VCode or an exact-answer check. Reference replay proves solvability; the seeded-random failure floor proves basic discrimination. Neither establishes empirical model difficulty.

License and upstream terms

Blobfish-authored release metadata and scripts follow the terms of the linked Blobfish repository. Upstream benchmark names, task sets, and source projects retain their own licenses. This release does not redistribute upstream Harbor tasks; consult each linked source page before using those separate datasets.

Reproduce

git clone https://github.com/blobfishai/blobfishai.git
cd blobfishai
node benchmark/validate.mjs

harbor run -d blobfishai/domainbench-24 -a <agent> -m <provider/model>

See the portfolio page for interactive task, asset, environment, trajectory, and methodology views.