---
title: "Comment by OnlineProxy on \"Scraped Data Quality Is a Runtime Contract, Not a Cleanup Task\" | Hashnode"
description: "the main thing i’d flag is that normalized drops price_text, so reviewers of quarantined records can’t see what the parser actually got, which goes against the"
url: "https://hashnode.com/posts/scraped-data-quality-is-a-runtime-contract-not-a-cleanup-task/6ace024d2180f13ea0681ef5/comment/6acfdb2cde6f6a64a52b5217"
---

[Discussion](https://hashnode.com/posts/scraped-data-quality-is-a-runtime-contract-not-a-cleanup-task/6ace024d2180f13ea0681ef5)

Comment [Scraped Data Quality Is a Runtime Contract, Not a Cleanup Task](https://hashnode.com/posts/scraped-data-quality-is-a-runtime-contract-not-a-cleanup-task/6ace024d2180f13ea0681ef5)

[OnlineProxy](https://hashnode.com/@onlineproxyio)

[Oct 14](https://hashnode.com/posts/scraped-data-quality-is-a-runtime-contract-not-a-cleanup-task/6ace024d2180f13ea0681ef5/comment/6acfdb2cde6f6a64a52b5217)

the main thing i’d flag is that normalized drops price\_text, so reviewers of quarantined records can’t see what the parser actually got, which goes against the article’s own advice

the locale handling could also use some work, since the euro and pound signs don’t get stripped and comma decimals can silently turn into the wrong number. empty price\_history skips the plausibility check entirely, which is exactly where new products tend to hide bugs. a per-identity MAD check might help, but it’d need testing on real data first