Street Pairs

Roadmap — proposed format, not yet emitted

The street-pair schema below is the designed training-data unit. It is the proposed format and is not yet emitted by built code in the audited codebase. (See the overview for full status.)

What a street pair is

The unit of operator-language training data. Each pair is input → output:

Pair type	Input	Output
COMMUNICATOR	Raw street language	Structured intent · canonical terms · client explanation · backend directive
SWARMFIXER	Failed agent output OR risk-flagged segment	Repair directive · remediation hint
VOCABULARY	Unknown emerging phrase	Proposed canonical term · slug · category

Schema

{
  "pair_type": "COMMUNICATOR | SWARMFIXER | VOCABULARY",
  "pair_id": "SPAIR-{YYYYMMDD}-{ULID}",
  "source_event_id": "...",
  "input": { ... },
  "output": { ... },
  "tribunal_grade": { "classification": ..., "score": ..., "reason": ... },
  "training_value": { "value_score": ..., "use_for_training": true }
}

Output format

Pairs are designed to be written as .jsonl files (one JSON object per line) for direct ingestion by training pipelines. See Examples · Sample Street Pair for a complete example.

The tribunal_grade.classification field carries the same verdict tiers that ARE real in DefendableCloud (defendable-cloud-v2): honey (pass), jelly (risk), propolis (fail). The street-pair grade speaks the Cloud referee’s rulebook vocabulary.

---

🐝 Operator-grade · books and records · to the shed.