// the take
A Windows port and build-wrangling fork of CRM114, Bill Yerazunis' statistical message classifier best known as a Bayesian-era spam filter with its own mini scripting language. It's for someone who specifically needs CRM114's classifier algorithms (OSB, hyperspace, winnow, SVM, neural net) running natively on Windows rather than through Cygwin, or someone doing historical/academic work on classic text classification.
The classifier breadth is real: crm_osb_bayes.c, crm_osb_winnow.c, crm_osb_hyperspace.c, crm_expr_svm.c and crm_neural_net.c are separate, swappable implementations rather than one algorithm dressed up. The spamfilterjig subdirectory is a legitimate benchmarking harness with ROC curve generation (roc0.plot, plotroc.c) and bootstrap resampling (bootstrap2.c) for comparing classifier accuracy, which is more rigor than most spam-filter repos bother with. The mailfilter integration (mailreaver.crm, procmailrc.recipe) shows the project was built around a real deployment path, not just a toy.
Last commit is from 2011 — this is a 15-year-dead fork with 10 stars and no realistic chance it builds cleanly against a current MinGW/MSVC toolchain without patching the autotools config yourself. The README is actively broken: it's literally the github-markup gem's README pasted in by mistake, and the real crm114 README just says 'GROT GROT GROT' and lists 'write a proper introduction' as an open TODO, so there's no usable entry point into the custom CRM scripting language you'd need to write classify rules. CRM114's classifiers (Bayes, OSB, winnow) are mid-2000s spam-filtering techniques; for anything beyond nostalgia or academic comparison they're outclassed by basically any modern approach. No maintainer activity means any bug, including the memory-unsafe C parsing code typical of this era, stays unfixed.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →