Recipe catalog / safety-incident-kind

Classify a safety event

What kind of safety event does report describe: a near miss, a first-aid injury, a medical-treatment injury, property damage, an environmental release, or an unsafe condition?

You need to sort free-text safety reports from workers, supervisors, or contractors into a fixed set of event kinds so each can be routed to the right investigation, form, or notification path.

Explore this recipe interactively · Source and implementation guide

Use safety-incident-kind in TypeScript

Install with npm install jev-recipes. Requires Node.js 22.9 or newer and ES modules. Set TYPESAFE_API_KEY in your server environment for live calls, which send input to TypeSafe and use API quota. See the installation guide.

import { safetyIncidentKind } from 'jev-recipes/safety-incident-kind';

const result = await safetyIncidentKind({
  "report": "Warehouse B, 14:20. A forklift reversing out of aisle 7 came within about a metre of a contractor walking through the marked pedestrian crossing. The driver stopped when the spotter shouted. No contact, no injuries, nothing damaged. The contractor was not wearing a high-visibility vest and the mirror at the aisle end is missing.",
  "minConfidence": 0.8
});
console.log(result);

Input contract

{
  "input": {
    "$schema": "https://json-schema.org/draft/2020-12/schema",
    "type": "object",
    "properties": {
      "report": {
        "type": "string"
      },
      "minConfidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      }
    },
    "required": [
      "report"
    ]
  },
  "result": {
    "$schema": "https://json-schema.org/draft/2020-12/schema",
    "type": "object",
    "properties": {
      "model": {
        "type": "string"
      },
      "usage": {
        "type": "object",
        "properties": {
          "input_tokens": {
            "type": "integer",
            "minimum": 0,
            "maximum": 9007199254740991
          },
          "output_tokens": {
            "type": "integer",
            "minimum": 0,
            "maximum": 9007199254740991
          }
        },
        "required": [
          "input_tokens",
          "output_tokens"
        ],
        "additionalProperties": false
      },
      "status": {
        "type": "string",
        "enum": [
          "ready",
          "review"
        ]
      },
      "verdict": {
        "type": "string",
        "enum": [
          "near_miss",
          "first_aid",
          "medical_treatment",
          "property_damage",
          "environmental_release",
          "unsafe_condition",
          "unclear"
        ]
      },
      "confidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      },
      "probabilities": {
        "type": "object",
        "propertyNames": {
          "type": "string",
          "enum": [
            "near_miss",
            "first_aid",
            "medical_treatment",
            "property_damage",
            "environmental_release",
            "unsafe_condition",
            "unclear"
          ]
        },
        "additionalProperties": {
          "type": "number",
          "minimum": 0,
          "maximum": 1
        },
        "required": [
          "near_miss",
          "first_aid",
          "medical_treatment",
          "property_damage",
          "environmental_release",
          "unsafe_condition",
          "unclear"
        ]
      }
    },
    "required": [
      "model",
      "usage",
      "status",
      "verdict",
      "confidence",
      "probabilities"
    ],
    "additionalProperties": false
  }
}

Saved example result

This hand-authored response demonstrates the contract. It is not a model accuracy measurement. Run it without an API key: npx jev-recipes demo safety-incident-kind.

{
  "model": "demo-fixture",
  "usage": {
    "input_tokens": 0,
    "output_tokens": 0
  },
  "status": "ready",
  "verdict": "near_miss",
  "confidence": 0.87,
  "probabilities": {
    "near_miss": 0.87,
    "first_aid": 0.01,
    "medical_treatment": 0.01,
    "property_damage": 0.01,
    "environmental_release": 0.01,
    "unsafe_condition": 0.07,
    "unclear": 0.02
  }
}

Evaluation evidence

Use the evaluation guide to measure this decision on your own labeled cases.

Limitations

• Classifies what the report describes, not what happened. An understated report yields an understated kind.
• Does not decide recordability, reportability, or lost-time status under any regulation. Map kinds to your regulatory and notification rules in application code.
• A report describing several events is classified by the most consequential one; split multi-event reports in code when each matters.

Related recipes

• incident-severity-wording: Use incident-severity-wording to grade how severe an operational outage is, rather than what kind of workplace safety event occurred.
• failure-kind: Use failure-kind when the categories are caller-supplied rather than this fixed set of safety event kinds.