Claude
Checking Claude output before it ships
Output tends to arrive long and organised, which is an advantage for structure and a risk for claims: length reads as thoroughness, and a thorough-sounding document is rarely re-read for evidence.
Take it apart before you sign it off.
Summaries need a comparison pass
Where a draft summarises material you provided, read the summary beside the source and look for the qualifier that disappeared. Compression is where accuracy is lost, and it is invisible unless you put the two side by side.
Long documents need an outline check
Skipped heading ranks, duplicate section titles and 300-word blocks all render correctly and export badly. A structural pass takes a minute and is the difference between a document people navigate and one they abandon.
What Claude output typically gets wrong
Compression loss
A summary that drops the qualification the source carried, leaving a claim the original did not make.
Confident structure around thin evidence
Headings, subheadings and lists that imply analysis where the paragraphs restate.
Invented dependencies
A plausible package name or option that cannot be resolved at install time.
Silent failure paths in code
Handlers that return defaults rather than surfacing errors, so the failure is invisible.
What it produces
- Longer documents, briefs and analyses
- Code generated across multiple files
- Summaries of material you supplied
- Structured data and configuration
One file
The findings are the same whoever wrote it.
Paste the output and the console returns what the rules observed, with severities you can argue with.