cat _posts/2026-11-04-caciarabot-the-running-bot-now-says-which-settings-it-loaded-and-stats-shows-what-actually-fired.md
caciarabot: the running bot now says which settings it loaded, and /stats shows what actually fired
The complaint was that CACIARABOT_LLM_REPLY_PROBABILITY=0.05 still replies too often, and the roll turned out to be correct — 5.01% over 200,000 simulated messages, one roll per message, no second handler — which left the value the running process actually holds as the suspect. That is hard to inspect in caciarabot now that production runs Distroless: no shell, no printenv. A .env edit followed by podman compose restart also keeps the old value, because restart does not re-read env_file; only down then up -d does, which deploy/update.sh already does (729dedc).
So main.py logs a settings_loaded line at startup with the ambient-behaviour settings it really read — no secrets — and podman compose logs | grep settings_loaded answers the question directly. /stats had the counters but showed none of them; it now reports spontaneous replies, replies when addressed and emoji reactions as a count and as a percentage of observed messages, so “too frequent” becomes a number to compare against the configured probability. The emoji line is the one that explains the complaint: at the migrated default of 0.33 it fires on one message in three, which reads far more like “it replies” than a 5% text reply does.
Both locales gained the three strings with their key sets verified identical, and a new tests/test_stats_command.py covers the command.