HumanSignal/label-studio · error · ValidationError

Error validating prediction: {validation_errors}

Error message

Error validating prediction: {validation_errors}

What it means

PredictionSerializer.validate() supports a pluggable validator via settings.CUSTOM_INTERFACE_PREDICTION_VALIDATOR (loaded with load_func). When that setting is configured and the custom validator returns non-empty errors for the prediction's result, ValidationError wraps and re-raises those errors verbatim.

Source

Thrown at label_studio/tasks/serializers.py:118

        project = None
        if 'task' in data:
            project = data['task'].project
        elif 'project' in data:
            project = data['project']
        ff_user = project.organization.created_by if project else 'auto'

        # Only validate if we're updating the result field
        if 'result' not in data:
            return data

        if not project:
            raise ValidationError('Project is required for prediction validation')

        custom_interface_validator = load_func(getattr(settings, 'CUSTOM_INTERFACE_PREDICTION_VALIDATOR', None))
        if custom_interface_validator:
            validation_errors = custom_interface_validator(project, data.get('result', []))
            if validation_errors:
                raise ValidationError(f'Error validating prediction: {validation_errors}')

        if not flag_set('fflag_feat_utc_210_prediction_validation_15082025', user=ff_user):
            # Skip validation if feature flag is not set
            logger.info(f'Skipping prediction validation in PredictionSerializer for user {ff_user}')
            return super().validate(data)

        # Custom Interface projects normally keep the default <View></View>
        # label_config and are validated above against output_schema instead.
        if not project.label_config_is_not_default:
            return super().validate(data)

        # Validate prediction using LabelInterface
        li = LabelInterface(project.label_config)
        validation_errors = li.validate_prediction(data, return_errors=True)

        if validation_errors:
            raise ValidationError(f'Error validating prediction: {validation_errors}')

View on GitHub (pinned to 0b49e9b539)

Solutions

1. Read the returned validation_errors detail — it comes from your custom validator, so fix the result payload to satisfy it
2. Check the CUSTOM_INTERFACE_PREDICTION_VALIDATOR implementation in your settings to understand its rules
3. Update the custom validator if it is stale relative to the current label config
4. Temporarily remove/adjust the setting to confirm it is the source of the rejection

Example fix

// before
CUSTOM_INTERFACE_PREDICTION_VALIDATOR = 'myapp.validators.old_pred_validator'
// after
CUSTOM_INTERFACE_PREDICTION_VALIDATOR = None  # or an updated validator matching current label config

from django.conf import settings

def precheck_with_custom_validator(project, result):
    loader = getattr(settings, 'CUSTOM_INTERFACE_PREDICTION_VALIDATOR', None)
    if not loader:
        return
    validator = __import__(loader.rsplit('.', 1)[0], fromlist=['x'])
    fn = getattr(validator, loader.rsplit('.', 1)[1])
    errors = fn(project, result)
    if errors:
        raise ValueError(f"custom validator rejects result: {errors}")

def custom_validator_configured(settings) -> bool:
    return bool(getattr(settings, 'CUSTOM_INTERFACE_PREDICTION_VALIDATOR', None))

from rest_framework.exceptions import ValidationError
try:
    ser = PredictionSerializer(data=payload)
    ser.is_valid(raise_exception=True)
except ValidationError as e:
    errors = e.detail
    # detail is the custom validator's message — fix result per its rules
    logger.error("custom prediction validation failed: %s", errors)

When it happens

Trigger: Creating/updating a prediction whose result violates rules enforced by the custom validator configured in settings (e.g. organization-specific schema constraints on allowed labels/types), with the prediction-validation feature in play.

Common situations: Deployments with a custom validator from an older project config that rejects results from newly changed label configs; mismatches after editing the labeling interface; validator expecting a different result shape than the client sends.

Related errors

• Each item in prediction result should be dict
• Incorrect format {type(result)} for prediction result {resul
• Project is required for prediction validation
• Validation failed on {}: {}
• Label config contains non-unique names:

AI-assisted analysis of HumanSignal/label-studio@0b49e9b539 (2026-08-29). Data as JSON: /api/errors/02321212768bf930. Report an issue: GitHub.