jevrecipes

Recipe catalog / paragraph-boundary

Recover a paragraph boundary

Do two adjacent extracted text fragments continue one paragraph or belong to separate blocks?

Extraction preserved reading order but lost the difference between a hard line wrap and a paragraph break.

Explore this recipe interactively ยท Source and implementation guide

Use paragraph-boundary 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 { paragraphBoundary } from 'jev-recipes/paragraph-boundary';

const result = await paragraphBoundary({
  "left": "To finish installing the package, run",
  "right": "the command shown in the next example.",
  "minConfidence": 0.8
});
console.log(result);

Input contract

FieldTypeNeeded
leftstringRequired
rightstringRequired
beforestringOptional
afterstringOptional
contextstringOptional
minConfidencenumberOptional
Full input and result schemas
{
  "input": {
    "$schema": "https://json-schema.org/draft/2020-12/schema",
    "type": "object",
    "properties": {
      "left": {
        "type": "string"
      },
      "right": {
        "type": "string"
      },
      "before": {
        "type": "string"
      },
      "after": {
        "type": "string"
      },
      "context": {
        "type": "string"
      },
      "minConfidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      }
    },
    "required": [
      "left",
      "right"
    ]
  },
  "result": {
    "$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": [
          "continue",
          "separate",
          "unclear"
        ]
      },
      "confidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      },
      "probabilities": {
        "type": "object",
        "propertyNames": {
          "type": "string",
          "enum": [
            "continue",
            "separate",
            "unclear"
          ]
        },
        "additionalProperties": {
          "type": "number",
          "minimum": 0,
          "maximum": 1
        },
        "required": [
          "continue",
          "separate",
          "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 paragraph-boundary.

{
  "model": "demo-fixture",
  "usage": {
    "input_tokens": 0,
    "output_tokens": 0
  },
  "status": "ready",
  "verdict": "continue",
  "confidence": 0.96,
  "probabilities": {
    "continue": 0.96,
    "separate": 0.02,
    "unclear": 0.02
  }
}

Evaluation evidence

Earlier-evaluator measurement

typesafe-ai/jev / 2026-09-27 / 167 held-out cases

Scoring revision 1.

93%All-case accuracy
85Sent for review
0Failed calls / cases

82 ready decisions, with 99% accuracy among those decisions.

95% case-level interval: 88% to 96%. Related synthetic cases are correlated.

Experimental: declared document acceptance policy not met

This measurement uses an earlier or unverified recipe or evaluator version. Rerun with the current recipe and evaluator before treating these numbers as current.

Full report and missesDataset and labels

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

Limitations

Related recipes