Writing

Notes from building the corpus.

Mostly the parts that were harder than expected — the undocumented pagination caps, the identifiers that live in prose, and the payload that would not fit in memory.

• GPSR: monitoring Safety Gate is easy, proving you did it is not
  The EU General Product Safety Regulation has applied since 13 December 2024. Checking alerts is the simple part — the requirement sellers fail is producing a dated, defensible record of what they checked and when. Measured against 64,455 EU recalls.
  7 min read
• Looking up a recall by barcode, and where the other four fifths went
  A barcode is the obvious way to ask whether a product has been recalled, and it works on 36,392 of the 176,198 recalls in this corpus. Which feeds carry barcodes, why most do not, and what to search by instead.
  10 min read
• Seven government recall feeds, and what it takes to make them agree
  Notes from normalising 176,000 product recalls from the EU, France and the US into one schema — the undocumented skip caps, a 14 MB download that decompresses to 294 MiB, why streaming it was not enough, and the amendments nobody announces.
  10 min read