Historical docs estimated '95-98% Enron compatibility' without a harness. Now that the pipeline is consolidated behind mime::parse_message(), add a corpus runner that:
- downloads (or points at a local copy of) a public email corpus (SpamAssassin public corpus is small and redistributable; Enron for scale)
- parses every message, counting: parse success, rejected-by-policy, anomalies by kind/severity, decoded-body success rate, and time/memory
- publishes real numbers in docs (replacing the old estimates), and fails CI on regressions of the success rate on a pinned sample
The fuzz corpus can also be seeded from interesting failures found here.
Historical docs estimated '95-98% Enron compatibility' without a harness. Now that the pipeline is consolidated behind mime::parse_message(), add a corpus runner that:
The fuzz corpus can also be seeded from interesting failures found here.