JSONL Auto-Fixer

updated 4 May 2026

100% client-side. Your file never leaves this page.

What it fixes

Every input line passes through this pipeline. Each step only touches the line if a transformation actually applies, and the final output is re-parsed before being accepted — so the fixer won't silently produce something that looks like JSON but isn't.

1. Already-valid lines pass through unchanged. No edits when the line already parses.
2. UTF-8 BOM at the start of a line — common when a file was saved from Notepad on Windows.
3. Leading and trailing whitespace on each line.
4. Leading garbage before the first { or [ (e.g. log prefixes, banner text).
5. Smart / curly quotes “ ” ‘ ’ normalised to straight ASCII " '.
6. JS-style comments — // line and /* block */ stripped.
7. NaN, Infinity, -Infinity, undefined replaced with null (these aren't valid JSON values, but JS often emits them).
8. Single-quoted strings and keys rewritten to use double quotes, with embedded double-quotes escaped.
9. Unquoted object keys like {id: 1} rewritten to {"id": 1}.
10. Trailing commas before } or ] removed.

If a line still won't parse after all of that, the fixer leaves it out of the output and reports it in the error list with the column and a suggestion. It will never substitute a partially-broken record with a guess — silently corrupted data is worse than data you know is broken.

How to use it

1. Paste the broken JSONL into the Input pane, or drop a .jsonl file onto the drop zone.
2. Click Fix automatically. The repair pipeline runs over every line.
3. The Output pane fills with cleaned JSONL; the What was fixed summary lists each transformation and how many lines it touched.
4. Anything unfixable shows up in the error list with line, column, the offending source, and a suggested fix you can apply by hand.
5. Click Download fixed or Copy to grab the cleaned file.

Example

Input — five common breakages on consecutive lines:

{"id":1,"ok":true}
{'id':2,'ok':true}
{"id":3,"ok":true,}
{"id":4,"score":NaN}
{id:5, ok:true}

Output:

{"id":1,"ok":true}
{"id":2,"ok":true}
{"id":3,"ok":true}
{"id":4,"score":null}
{"id":5,"ok":true}

What it deliberately won't do

• Won't infer missing fields. If a line is missing a required key, that's a schema problem — use the OpenAI / Anthropic validator or schema inferrer.
• Won't close unterminated strings. Guessing where a string was supposed to end would risk merging two records.
• Won't unwrap a JSON array on a line. One record per line is the convention; if you have [{...}, {...}] on a line, use the JSONL ↔ JSON converter first.

Related tools

• JSONL ValidatorJust check syntax — don't modify the file.
• JSONL FormatterPretty-print or minify each line.
• LLM dataset validatorSchema-aware checks for OpenAI and Anthropic fine-tune files.
• Schema inferrerWalk the file and emit a JSON Schema.

Frequently asked questions