Skip to main content
---
title: "Authoring Evidence"
description: "Understanding the authoring evidence summary attached to improvement proposals and frontier candidates"
---

## Overview

When selftune cloud evaluates improvement candidates, it attaches an **authoring evidence summary** to each comparison result and frontier parent. This summary captures how many eval cases were authored, seeded, or linked to runs — and whether those cases improved or regressed under the candidate.

Authoring evidence helps you understand the quality of signal behind an improvement proposal, especially when reviewing why a particular candidate was selected or promoted.

## Where it appears

`authoring_evidence` is an optional field on:

- **Comparison summaries** — returned in improve run results when comparing a candidate against the baseline
- **Frontier parent records** — attached to the parent candidate that seeded a given improvement run

## Fields

| Field | Type | Description |
|-------|------|-------------|
| `authoring_case_count` | `integer` | Total number of authored eval cases considered |
| `seeded_case_count` | `integer` | Cases that were seeded from existing signals |
| `run_linked_case_count` | `integer` | Cases linked to a specific improve run |
| `saved_suite_smoke_pass_count` | `integer` | Saved suite smoke tests that passed for this candidate |
| `saved_suite_smoke_fail_count` | `integer` | Saved suite smoke tests that failed for this candidate |
| `improved_authoring_case_count` | `integer` | Authored cases where the candidate improved over baseline |
| `regressed_authoring_case_count` | `integer` | Authored cases where the candidate regressed against baseline |
| `improved_smoke_backed_case_count` | `integer` | Smoke-backed cases that improved |
| `regressed_smoke_backed_case_count` | `integer` | Smoke-backed cases that regressed |
| `suggestion_provenances` | `string[]` | Reasons cases were included in this evaluation (see below) |

## Suggestion provenances

The `suggestion_provenances` array describes why eval cases were surfaced for this candidate:

| Value | Meaning |
|-------|---------|
| `failed_evaluation` | Case was included because an evaluation previously failed |
| `missed_query` | Case was included because a query went unmatched |
| `run_failure` | Case was included due to a prior run failure |
| `run_regression` | Case was included because a regression was detected in a prior run |

A candidate with `run_regression` provenances and a high `regressed_authoring_case_count` is a signal that the candidate may be moving in the wrong direction.

## Example

A comparison summary with authoring evidence attached:

```json
{
  "candidate_id": "...",
  "validation_status": "valid",
  "authoring_evidence": {
    "authoring_case_count": 12,
    "seeded_case_count": 8,
    "run_linked_case_count": 4,
    "saved_suite_smoke_pass_count": 10,
    "saved_suite_smoke_fail_count": 2,
    "improved_authoring_case_count": 9,
    "regressed_authoring_case_count": 1,
    "improved_smoke_backed_case_count": 7,
    "regressed_smoke_backed_case_count": 1,
    "suggestion_provenances": ["missed_query", "run_regression"]
  }
}