OUTPUT_SCHEMAS.md

OPF Output Schemas

1. opf / opf redact JSON Output

Printed to stdout once per input example.

{
  "schema_version": 1,
  "summary": {
    "output_mode": "typed",
    "span_count": 3,
    "by_label": {
      "private_person": 1,
      "private_date": 2
    },
    "decoded_mismatch": false
  },
  "text": "Alice was born on 1990-01-02.",
  "detected_spans": [
    {
      "label": "private_person",
      "start": 0,
      "end": 5,
      "text": "Alice",
      "placeholder": "<PRIVATE_PERSON>"
    }
  ],
  "redacted_text": "<PRIVATE_PERSON> was born on <PRIVATE_DATE>."
}

Notes:

• In --output-mode redacted, every detected_spans[*].label becomes redacted.
• warning is present only when tokenizer decode does not exactly round-trip the input text.

2. opf eval Predictions Output (--predictions-out)

Written as JSONL when requested.

{
  "example_id": "stable-id",
  "text": "Alice was born on 1990-01-02.",
  "predicted_spans": {
    "private_person: Alice": [[0, 5]]
  }
}

Optional field:

• token_logprobs_topk: included only when --predictions-token-logprobs-topk > 0

Notes:

• This file is literal JSONL: one compact JSON object per line.

Stability Notes

• typed and untyped are the evaluation terms.
• typed and redacted are the prediction-output terms.
• Additive fields may appear over time, but existing keys should remain stable unless schema_version changes for API/CLI JSON payloads.