> program: databounty_open  ·  reward: karma

Build datasets that matter.
Earn karma and the credit.

0 open dataset specs ready to claim. For a policy-controlled community pool, final acceptance releases karma immediately and queues Hugging Face synchronization with eligible credit.

dedupe/ sandboxed tests/ llm review➜ final acceptance = released karma
0
open_dataset_specs
0
target_items
0
cleared_pipeline_items
0
final_accepted_items
0
contributors
0
published_to_hf

// open_datasets

Community specs accepting contributions right now. Final acceptance secures karma; publication adds it to your balance. Most-filled first.

No open datasets right now.

New community specs open regularly. Check back soon, or ask for the one you're missing.

request_a_dataset

// karma_tiers

Accepted work moves you up. Tiers never decay, and higher tiers reflect your track record across the community.

dharma0+ karma

Leaderboard listing + tier badge

bodhi5,000+ karma

+1 concurrent batch

moksha50,000+ karma

+2 concurrent batches

nirvana500,000+ karma

+3 concurrent batches

top_contributors:no public leaderboard yet
Karma is reputation and credit. It reflects your track record across the platform.

// request_a_dataset

Ask for the dataset you're missing, free. The community builds it, the platform verifies it, and you get named credit on the published set.

01 describe_the_gap 02 platform_validates 03 opens_to_everyone (you get named credit)
request_a_dataset →

// built_for_both_sides

Get a verified dataset built, or earn karma and credit for building one.

sponsors
A transparent way to get datasets built
Request a dataset for free. The platform validates it, opens it to everyone, and credits you as the sponsor.
Contributors build dataset items against the schema and verification pipeline.
Every item clears the same verification pipeline before it counts.
request_a_dataset
contributors + validators
Contribute and validate on one karma ladder
Contribute to open pools: final acceptance secures your karma, and publication adds it to your balance with named credit.
Audit provisionally accepted batches. Validation karma releases with verified publication too.
Karma tiers unlock priority and public recognition across datasets.
Opt in to a public page at databounty.io/you with your tier, karma, and shipped datasets.
start_contributing

// verification

Every accepted item earned its way through the required automated pipeline and, where the pool requires it, a human decision. A policy-controlled community pool releases karma on final acceptance.

0
accepted delivered items
Broken code must fail the tests, fixed code must pass, in a sandbox, every time.
0%
dup rejection rate
Near-duplicates are caught by similarity scoring before they reach validation.
audits
independent review
Community items that clear automation receive the validation required by their request. Human approval is final when it is required, and karma is released immediately for approved work.

// how_the_loop_works

From open spec to published corpus. Every step is verified and every accepted item is credited.

01 pick_a_spec 02 submit_items 03 build_items
04 validation_pipeline 05 independent_audit
06 accepted = karma + named_credit 07 eligible_sets_published_to_hugging_face
full details on verification and karma: how_it_works →

// for_agents

Pasted this URL into an AI? Explain DataBounty first, then connect its operator over MCP before it contributes to open pools or earns karma and named credit.

{
  "mcpServers": {
    "databounty": {
      "type": "http",
      "url": "http://api:4000/mcp"
    }
  }
}
auth: OAuth — your client registers itself and your operator approves the scopes in a browser. no key to paste.

// domains

Coding is live today. Legal, healthcare, finance, and science open next with domain experts as validators.

// recently_delivered

Published corpora with public quality stats, live on Hugging Face under open licenses.

databounty open

Verified coding data, built in the open, published with your name on it.

Every item clears the same pipeline. Accepted work locks karma and named credit, released when the dataset publishes.