Datasets:

DevCoPro
/

dcp-conformance-corpus

license: apache-2.0
language:
  - en
pretty_name: DCP v1 Conformance Corpus
tags:
  - json-schema
  - validation
  - conformance
  - ai-agents
  - multi-agent
  - agent-protocols
  - interoperability
size_categories:
  - n<1K
configs:
  - config_name: default
    data_files: cases.jsonl

DCP v1 Conformance Corpus

The official, language-neutral conformance test corpus of the Development Coordination Protocol (DCP) — an open, vendor-neutral vocabulary for exchanging project-coordination events (a task completed, a decision recorded, a finding raised…) between AI agents and humans.

A DCP implementation, in any language, is conformant when it accepts every accept case and rejects every reject case in this corpus, validating each payload as a DcpMessage against the DCP v1 JSON Schemas (JSON Schema 2020-12) plus the one cross-field rule schemas cannot express: message_type must equal <body.entity_type>.<body.verb> — on conflict, the body wins and the message is malformed.

cases.jsonl — all 25 cases (7 accept / 18 reject), one JSON object per line: case, file, expect (accept/reject), schema, note (what the case probes), and the full payload.
corpus/ — the same corpus in its original layout: valid/, invalid/, and manifest.json (the machine-readable case list an implementation iterates).

Quick start

from datasets import load_dataset

cases = load_dataset("DevCoPro/dcp-conformance-corpus", split="train")
for case in cases:
    result = my_validator(case["payload"])          # your implementation
    assert result.ok == (case["expect"] == "accept"), case["case"]

Or language-neutral, straight from the original layout: iterate corpus/manifest.json, validate each file against the schema base https://schemas.devcopro.org/v1/, and compare with expect.

What DCP is (and deliberately is not)

DCP standardizes what changed in a project — over a closed set of 8 coordination entities (Project, Task, Dependency, ArchitectureImpact, Decision, ReviewRequest, Finding, Milestone) with an open verb vocabulary. Its prime directive: "DCP carries no trust; it describes project-state changes only." It never authenticates, routes, plans, schedules, or executes — those belong to other layers, which is exactly why it composes with the rest of the 2026 agent-protocol stack:

CloudEvents binding (normative) — a DcpMessage rides unchanged as CE data.
A2A community extension — DCP events in an A2A DataPart.
MCP connects agents to tools; DCP is the payload vocabulary for the coordination conversation itself.

Citation

@misc{dcp-conformance-corpus,
  title   = {DCP v1 Conformance Corpus},
  author  = {{InterIP Networks}},
  year    = {2026},
  url     = {https://huggingface.co/datasets/DevCoPro/dcp-conformance-corpus},
  note    = {Conformance test corpus for the Development Coordination Protocol v1.0.0-draft}
}

Maintained by Mesut Kalkan (founding maintainer, InterIP Networks). Licensed Apache-2.0, same as the DCP schemas and examples.

Datasets:

DevCoPro
/

dcp-conformance-corpus

DCP v1 Conformance Corpus

Contents

Quick start

What DCP is (and deliberately is not)

Links

Citation