sample_predictions.json

{
  "description": "Functional smoke examples, not a benchmark",
  "examples": [
    {
      "text": "Design a controlled sequence that assigns a new support request, records each hand-off and pauses for a supervisor before any consequential change.",
      "expected_intent": "workflow_automation",
      "predicted_intent": "workflow_automation",
      "matches_expected": true,
      "top_probabilities": [
        {
          "intent": "workflow_automation",
          "probability": 0.25173970286178327
        },
        {
          "intent": "analytics",
          "probability": 0.14608161352719087
        },
        {
          "intent": "document_processing",
          "probability": 0.1172506057834092
        }
      ]
    },
    {
      "text": "Extract the reference, date and category from a batch of synthetic forms, preserve page citations and queue uncertain fields for review.",
      "expected_intent": "document_processing",
      "predicted_intent": "document_processing",
      "matches_expected": true,
      "top_probabilities": [
        {
          "intent": "document_processing",
          "probability": 0.26252662031364493
        },
        {
          "intent": "analytics",
          "probability": 0.11820465287105189
        },
        {
          "intent": "knowledge_retrieval",
          "probability": 0.11471476985565923
        }
      ]
    },
    {
      "text": "Analyse the demonstration queue by service type and ageing band, showing median turnaround, missing values and reproducible aggregates.",
      "expected_intent": "analytics",
      "predicted_intent": "analytics",
      "matches_expected": true,
      "top_probabilities": [
        {
          "intent": "analytics",
          "probability": 0.3112028647640029
        },
        {
          "intent": "reporting",
          "probability": 0.11827456054337938
        },
        {
          "intent": "workflow_automation",
          "probability": 0.113612834506195
        }
      ]
    },
    {
      "text": "Find the approved guidance for an internal procedure, cite the exact section and say clearly when the knowledge base does not contain an answer.",
      "expected_intent": "knowledge_retrieval",
      "predicted_intent": "knowledge_retrieval",
      "matches_expected": true,
      "top_probabilities": [
        {
          "intent": "knowledge_retrieval",
          "probability": 0.3725641890421974
        },
        {
          "intent": "exception_handling",
          "probability": 0.11718831857416644
        },
        {
          "intent": "human_escalation",
          "probability": 0.10669078884517795
        }
      ]
    },
    {
      "text": "Quarantine the affected record when validation fails, preserve the evidence, alert its owner and define safe retry conditions without stopping other work.",
      "expected_intent": "exception_handling",
      "predicted_intent": "exception_handling",
      "matches_expected": true,
      "top_probabilities": [
        {
          "intent": "exception_handling",
          "probability": 0.39957975522975697
        },
        {
          "intent": "workflow_automation",
          "probability": 0.14827720513606246
        },
        {
          "intent": "document_processing",
          "probability": 0.08959612528781084
        }
      ]
    },
    {
      "text": "Prepare a privacy-minimised review pack and route the uncertain high-impact request to a named manager for an explicit recorded decision.",
      "expected_intent": "human_escalation",
      "predicted_intent": "human_escalation",
      "matches_expected": true,
      "top_probabilities": [
        {
          "intent": "human_escalation",
          "probability": 0.17873971468427835
        },
        {
          "intent": "workflow_automation",
          "probability": 0.1528334966928371
        },
        {
          "intent": "analytics",
          "probability": 0.13204253575995994
        }
      ]
    },
    {
      "text": "Synchronise approved synthetic records between two sandbox systems with schema validation, idempotency keys, bounded retries and reconciliation totals.",
      "expected_intent": "api_integration",
      "predicted_intent": "api_integration",
      "matches_expected": true,
      "top_probabilities": [
        {
          "intent": "api_integration",
          "probability": 0.524774023979317
        },
        {
          "intent": "analytics",
          "probability": 0.07249582870387707
        },
        {
          "intent": "knowledge_retrieval",
          "probability": 0.07125293418771814
        }
      ]
    },
    {
      "text": "Generate a monthly operational summary with reporting period, volumes, overdue work, definitions, source coverage and data-quality caveats.",
      "expected_intent": "reporting",
      "predicted_intent": "reporting",
      "matches_expected": true,
      "top_probabilities": [
        {
          "intent": "reporting",
          "probability": 0.4905433258815887
        },
        {
          "intent": "analytics",
          "probability": 0.09335855933097363
        },
        {
          "intent": "exception_handling",
          "probability": 0.0820343071940959
        }
      ]
    }
  ]
}