I Cleaned 455 Articles, 474 Numbers Stayed in the English Twin

Description and use

An automated sweep cleaned 455 Turkish articles, deleting 3,222 lines of fabricated content, while 474 percentages remained in English twins.

Problem

Automated content pipeline accumulated fabricated figures, invented scenes, and fake command output in a bilingual blog, requiring a sweep to de‑fab the content.

Outcome

The sweep cleaned 455 Turkish articles, deleted 3,222 Turkish lines, and identified 474 percentages that persisted in English twins.

Workflow

• Run automated sweep script (scripts/measure-defab-coverage.mjs) to identify fabricated content.
• Commit pilot with rule definitions.
• Execute four batch commits to delete fabricated lines.
• Run second wave commit to finalize cleanup.
• Measure coverage and report 455 Turkish files cleaned, 474 numbers remaining in English twins.

Evidence

Six commits: a pilot (09:53), four batches (10:19–10:55) and a second wave (13:42).

Source

The sweep touched 455 Turkish article files. The number of English files it touched: zero.

Source

474 numbers, across 213 articles.

Source

The pilot commit's message names four classes: fake numbers, fake command output, piles of anecdotes, repeated credits.

Source