The emit-time blocklist in batch_insert.py has grown to 38 patterns covering 19 provider domains (with and without @ variants). It started with a handful of obvious domains and evolved through: the "gmail" false-positive fix (allowing "Gmail" as platform name), the protonmail addition, the [provider domain]/[provider domain]/[provider domain] German provider expansion, and the [provider domain]/[provider domain]/[provider domain]/[provider domain]/[provider domain]/[provider domain]/[provider domain]/[provider domain] global provider additions. Each addition came from a discovered edge case. The blocklist is not static policy — it's accumulated learning encoded as code.