The recovered carbon black (rCB) industry has no public, structured, machine-readable data source that an AI agent can rely on. Trustbase Lab fills that gap.

• Patch (1.0 → 1.1): additive optional fields, documentation, no breaking changes.
• Minor (1.x → 2.0): new required fields, removed deprecated fields, breaking change.
• The current version is v1.3 (added the policies and techs top-level modules).
• Data files MUST include meta.schema_version. Consumers MUST pin to a major version.
• Breaking changes are announced 30 days before promotion to production data.

{
  "meta":         { /* dataset-level metadata */ },
  "domestic":     [ /* 31 Chinese companies / projects */ ],
  "international":[ /* 25 international companies */ ],
  "vendors":      [ /* 26 equipment vendors */ ],
  "policies":     [ /* 22 policy & regulation records  (v1.3) */ ],
  "techs":        [ /* 20 technical standards        (v1.3) */ ]
}

Total current count: 124 entities as of data_version: 0.8 — companies 82 (domestic 31 / international 25 / vendors 26), policies 22, techs 20.

{
  "id":          "rcb-cn-008",
  "public":      { /* 7 fields, safe to expose */ },
  "gated":       { /* 5-9 fields, detailed; phones desensitised */ },
  "description": { "zh": "...", "en": "..." },
  "as_of":       "2026-06"
}

These are the most important fields for AI agents. They are pre-shaped, structured summaries designed to be fed directly into an LLM prompt without further parsing. Both description.zh and description.en are required (P0-6).

Format: <prefix>-<3-digit-zero-padded-seq>, e.g. rcb-cn-008. The mapping name → id is persisted in id_map.json. Adding a new entity never changes an existing ID. Agents should hard-code IDs, not infer them from names.

This is the AI trust infrastructure of the schema. Agents auto-deciding procurement partners should filter on verified=true or call the MCP tool get_verified_suppliers.

{
  "public": {
    "verified": true,           // bool — AI agent decision signal
    ...
  },
  "gated": {
    "confidence": "A",          // required match for verified=true
    ...
  }
}

Rule: verified=true ⇔ confidence=A. The data steward enforces this in CI.

Breakdown: 16 Chinese + 6 international + 8 equipment vendors. The verified count is also surfaced in meta.verified_count (per module).

Full-corpus view (all 5 modules): 124 entities, 63 verified = 50.8% A-grade — policies 17/22 and techs 16/20 are additionally marked verified.

• Every record carries native (name, location, status) AND English (name_en) content.
• Two parallel description blocks are provided.
• Default output language is zh-CN (Chinese). English via language="en" on MCP tools.
• Status strings are translated via an internal dictionary (STATUS_EN); graceful degradation if missing.
• Locations are not yet translated (P1 limitation; use name_en instead).

The schema is the wire format. The MCP server at https://trustbaselab.com/mcp exposes 9 tools: query_companies, get_company, get_verified_suppliers, query_policies, get_policy, query_techs, get_tech, get_price_benchmark, get_database_overview.

curl -X POST https://trustbaselab.com/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc":"2.0","id":1,"method":"tools/call",
    "params":{
      "name":"get_verified_suppliers",
      "arguments":{"language":"en","response_format":"json"}
    }
  }'

Returns a structured object with verified_count, suppliers[] (id / name / name_en / region / location / status / verified / confidence / description / as_of).

curl -X POST https://trustbaselab.com/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc":"2.0","id":2,"method":"tools/call",
    "params":{
      "name":"query_companies",
      "arguments":{"region":"international","verified":true,"limit":10}
    }
  }'

curl -X POST https://trustbaselab.com/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc":"2.0","id":3,"method":"tools/call",
    "params":{
      "name":"get_company",
      "arguments":{"id":"rcb-cn-008","language":"zh","response_format":"json"}
    }
  }'

curl -X POST https://trustbaselab.com/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc":"2.0","id":4,"method":"tools/call",
    "params":{
      "name":"query_policies",
      "arguments":{"region":"欧盟","verified":true,"language":"en"}
    }
  }'

Returns policy records with doc_number, issuer, policy_type, scope, rcb_impact and compliance_deadline.

curl -X POST https://trustbaselab.com/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc":"2.0","id":5,"method":"tools/call",
    "params":{
      "name":"query_techs",
      "arguments":{"keyword":"ASTM","category":"测试标准","language":"en"}
    }
  }'

• Source-of-truth: an internal Excel workbook (信任_Lab_执行手册_v0.3.xlsx).
• Company data pipeline: xlsx → export_json.py → data_steward.py (10 P0 checks) → upload_data.py.
• Module data pipeline (v1.3): _src/policies.json + _src/techs.json + _src/vendors_extra.json → export_json.py (merge + assign stable IDs) → data_steward.py.
• Every record is assigned a stable entity ID via id_map.json — IDs are never reused or renumbered.
• Each export masks all 11-digit Chinese mobile numbers before publication.
• Re-export cadence: weekly inspection; urgent refresh on industry events.

When citing this schema in academic or commercial work: