> ## Documentation Index
> Fetch the complete documentation index at: https://docs.sixtyfour.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Confidence Scoring

> Per-field confidence percentages for enrichment results. Understand exactly how reliable each piece of data is.

Field-level confidence provides a 0–100% confidence score for each enrichment field individually, replacing the single lead-level score. The response indicates which fields are reliable and which require verification.

Each score reflects how much assumption is required to connect the field value back to the target person. A high score means the evidence chain is strong; a low score means there's meaningful uncertainty about attribution.

<Warning>
  This feature is in early access. Contact [Sixtyfour Support](mailto:support@sixtyfour.ai) to gain access.
</Warning>

## How to enable

Pass `field_confidence: true` in your `/people-intelligence` request body:

```json theme={null}
{
  "lead_info": {"full_name": "Jane Smith", "company": "Acme Corp"},
  "struct": {
    "email": "Email address",
    "title": "Current job title",
    "linkedin": "LinkedIn profile URL"
  },
  "tier": "high",
  "field_confidence": true
}
```

## Response format

When enabled, the response includes a `field_confidence` object that mirrors the shape of your `struct`. Each leaf field gets a confidence score (0–100) or `null` if no evidence path exists for that field.

```json theme={null}
{
  "notes": "Research complete. Found Jane Smith on LinkedIn and Acme Corp website.",
  "structured_data": {
    "email": "jsmith@acmecorp.com",
    "title": "VP of Engineering",
    "linkedin": "https://www.linkedin.com/in/janesmith-acme/"
  },
  "findings": [],
  "references": {
    "https://linkedin.com/in/janesmith-acme/": "LinkedIn profile"
  },
  "field_confidence": {
    "email": 88,
    "title": 96,
    "linkedin": 98
  }
}
```

### Response fields

| Field                             | Type                    | Description                                                                                                                |
| --------------------------------- | ----------------------- | -------------------------------------------------------------------------------------------------------------------------- |
| `field_confidence`                | object                  | Confidence scores mirroring the struct shape. Only present when `field_confidence: true` is passed and the org has access. |
| `field_confidence.{scalar_field}` | integer (0–100) or null | Confidence for a scalar leaf field. `null` means no evidence path exists.                                                  |
| `field_confidence.{object_field}` | object                  | Same keys as the struct object; each subfield scored individually.                                                         |
| `field_confidence.{list_field}`   | array                   | One object per item in `structured_data`; each subfield scored individually.                                               |

<Warning>
  When `field_confidence` is present in the response, `confidence_score` is not included. The two fields are mutually exclusive and use different scales: `field_confidence` scores are 0–100 per field, while `confidence_score` is 0–10 at the lead level.
</Warning>

### When field\_confidence is off

The response includes a single `confidence_score` (0–10) at the lead level:

```json theme={null}
{
  "structured_data": {...},
  "confidence_score": 8
}
```

## Nested fields (objects and object\_lists)

For nested structs, `field_confidence` mirrors the exact shape of `structured_data`:

```json theme={null}
{
  "struct": {
    "email": "Email address",
    "company": {
      "name": "Company name",
      "size": "Company size"
    },
    "jobs": [
      {"title": "Job title", "company": "Company name", "start_date": "Start date"}
    ]
  }
}
```

Response:

```json theme={null}
{
  "structured_data": {
    "email": "jsmith@acmecorp.com",
    "company": {"name": "Acme Corp", "size": "500"},
    "jobs": [
      {"title": "VP of Engineering", "company": "Acme Corp", "start_date": "2020"},
      {"title": "Senior Engineer", "company": "Google", "start_date": "2015"}
    ]
  },
  "field_confidence": {
    "email": 88,
    "company": {
      "name": 96,
      "size": 45
    },
    "jobs": [
      {"title": 95, "company": 95, "start_date": 60},
      {"title": 72, "company": 72, "start_date": null}
    ]
  }
}
```

The `field_confidence` structure always matches `structured_data` recursively. See [Response fields](#response-fields) for the per-type contract.

## Workflow mode

In workflows, field confidence is returned as a single `enrichment_field_confidence` column containing a JSON string. The column parses as standard JSON while keeping the CSV flat:

| name       | email                                     | enrichment\_field\_confidence                |
| ---------- | ----------------------------------------- | -------------------------------------------- |
| Jane Smith | [jsmith@acme.com](mailto:jsmith@acme.com) | `{"email": 88, "title": 96, "linkedin": 98}` |

For nested structs, the JSON column contains the full nested structure:

```json theme={null}
{"email": 88, "company": {"name": 96, "size": 45}, "jobs": [{"title": 95, "company": 95, "start_date": 60}, {"title": 72, "company": 72, "start_date": null}]}
```

To use confidence in downstream workflow blocks, parse the JSON column and filter on the fields you care about.

## Confidence tiers

The 0–100 score maps to named confidence levels:

| Range   | Level       | What it means                                                                             |
| ------- | ----------- | ----------------------------------------------------------------------------------------- |
| 85–100% | Confirmed   | Near-zero assumption. Unique identifiers, direct assertions from authoritative sources.   |
| 65–84%  | Established | Minimal assumption. A careful person would rely on this. One small residual uncertainty.  |
| 40–64%  | Supported   | Moderate assumption. The identity or value link is reasonable but not tight.              |
| 15–39%  | Speculative | Significant assumption. Weak identifiers, multiple possible matches. Treat as hypothesis. |
| 0–14%   | Unreliable  | Major assumption or conflicting evidence. Do not rely on.                                 |

## Usage tips

* Parse in code: the `enrichment_field_confidence` column (workflow) and `field_confidence` field (API) are both standard JSON. Use `JSON.parse()` or `json.loads()` to extract individual scores.
* Threshold by use case: cold email outreach typically accepts 65%+. Legal and compliance work should require 85%+.
* The weakest link dominates: if the target has a very common name, even well-sourced fields score lower because the identity link is uncertain.
* Null means no evidence: a `null` value means no evidence chain exists in the attribution graph — either the value wasn't found, or it was provided without traceable sourcing. The key is always present in `field_confidence` (the shape mirrors `structured_data` exactly).