Self-Service Content Evaluation

Test any text, or an existing comment or chat message, against your application's comment quality and AI moderation pipelines and see how each rule scores it.

Evaluations use your saved settings, and you can override them for a single request to try changes before you save them. Evaluations are read-only, they never save or change anything.

⚠️
Producer-only debugging endpoint.
This is a self-service endpoint for application producers to test and debug their comment quality and moderation settings. It needs a producer access token and is not meant to be integrated into end-user apps. Its request, payload and response may change without prior notice, so always refer to this page for the latest details before use.

Endpoints

Method	Path	Purpose
`POST`	`/api/v1/applications/{client_id}/evaluations/`	Start an evaluation
`GET`	`/api/v1/applications/{client_id}/evaluations/{evaluation_id}/`	Poll the result

Authentication:

A producer access token for the application (Authorization: Bearer <token>)

Pipelines

#	`pipeline`	What it checks
1	`comment_quality`	Whether a comment qualifies as a quality comment: length, characters, emoji, GIFs, sentiment and toxicity
2	`comment_moderation`	Whether AI moderation would flag a text or comment
3	`chat_moderation`	Whether AI moderation would flag a text or chat message

Payload

Each pipeline(POST API) needs exactly one input:

Either text, or the ID of an existing item (e.g Comment ID or Chat message ID).
An optional override_config . This changes your saved settings for this request only, fields you leave out keep their saved values. This allows you to evaluate the tuned configuration before saving.

Response

Every response, from every pipeline, has the same five keys:

url : Where to poll this evaluation for result.
pipeline : The pipeline that was run.
status : pending, completed or failed.
effective_config : The final settings used for this evaluation: your saved settings with override_config applied. It has the same shape as override_config .
result : null until status is completed, then the outcome.
- decision : pass or fail
- rules : A list of rules. One entry per rule. Each rule in the list has 4 keys -
  - name : The rule field (quality) or category (moderation)
  - outcome : pass or fail .error when a rule couldn't run.
  - value : What was measured, e.g. character count, emoji ratio or score
  - threshold : The threshold it was compared with.

Response examples

Pending (async POST):

{
  "url": "https://<host>/api/v1/applications/<client_id>/evaluations/<evaluation-id>/",
  "pipeline": "comment_quality",
  "status": "pending",
  "effective_config": {
    "min_char_count": 0,
    "disallowed_chars": "*",
    "emoji_ratio_threshold": 0.7,
    "exclude_gif": false,
    "min_sentiment_score": 0.2,
    "max_toxicity_score": 0.1
  },
  "result": null
}

Comment quality(Completed):

{
  "url": "https://<host>/api/v1/applications/<client_id>/evaluations/<evaluation-id>/",
  "pipeline": "comment_quality",
  "status": "completed",
  "effective_config": {
    "min_char_count": 0,
    "disallowed_chars": "*",
    "emoji_ratio_threshold": 0.7,
    "exclude_gif": false,
    "min_sentiment_score": 0.2,
    "max_toxicity_score": 0.1
  },
  "result": {
    "decision": "pass",
    "rules": [
      {"name": "min_char_count", "outcome": "pass", "value": null, "threshold": 0},
      {"name": "disallowed_chars", "outcome": "pass", "value": "", "threshold": "*"},
      {"name": "emoji_ratio_threshold", "outcome": "pass", "value": 0.0345, "threshold": 0.7},
      {"name": "exclude_gif", "outcome": "pass", "value": null, "threshold": false},
      {"name": "min_sentiment_score", "outcome": "pass", "value": 0.91, "threshold": 0.2},
      {"name": "max_toxicity_score", "outcome": "pass", "value": 0.03, "threshold": 0.1}
    ]
  }
}

Moderation (Completed):

{
  "url": "https://<host>/api/v1/applications/<client_id>/evaluations/<evaluation-id>/",
  "pipeline": "comment_moderation",
  "status": "completed",
  "effective_config": {
    "model_id": "unitary/toxic-bert",
    "toxic": 0.8,
    "severe_toxic": 0.8,
    "obscene": 0.8,
    "threat": 0.8,
    "insult": 0.5,
    "identity_hate": 0.8
  },
  "result": {
    "decision": "fail",
    "rules": [
      {"name": "toxic", "outcome": "pass", "value": 0.42, "threshold": 0.8},
      {"name": "severe_toxic", "outcome": "pass", "value": 0.01, "threshold": 0.8},
      {"name": "obscene", "outcome": "pass", "value": 0.05, "threshold": 0.8},
      {"name": "threat", "outcome": "pass", "value": 0.02, "threshold": 0.8},
      {"name": "insult", "outcome": "fail", "value": 0.67, "threshold": 0.5},
      {"name": "identity_hate", "outcome": "pass", "value": 0.01, "threshold": 0.8}
    ]
  }
}

Self-Service Content Evaluation

Producer-only debugging endpoint.

Endpoints

Pipelines

Payload

Response

Response examples

Note: Results expire 10 minutes after they're created. After that, the `GET` returns `404`.

Producer-only debugging endpoint.

Endpoints

Pipelines

Payload

Response

Response examples

Note: Results expire 10 minutes after they're created. After that, the GET returns 404.

Note: Results expire 10 minutes after they're created. After that, the `GET` returns `404`.