video seo

deploy/Caddyfile.example

@@ -208,14 +208,25 @@ catcrafts.net {
 
 	# ── access log ────────────────────────────────────────────────────────
 	#
-	# Feeds the GoAccess analytics (see deploy/catcrafts-analytics). Retention
-	# sized for that: ~15 MB/day means a roll roughly weekly; 52 compressed
-	# rolls ~ a year of raw logs (~6 MB each gzipped). roll_keep_for must be
-	# explicit — lumberjack's default silently deletes after 90 days.
+	# Feeds the GoAccess analytics (see deploy/catcrafts-analytics), and is the
+	# only record of everything those aggregates discard: Sec-Fetch headers,
+	# Range behaviour, query strings, per-IP request sequences. Every bot fleet
+	# identified so far was caught in the raw log, not the report.
+	#
+	# roll_keep is a COUNT, so retention is really min(roll_keep × roll_size,
+	# roll_keep_for) — a byte budget, not the time window it reads as. 52 rolls
+	# is 5.2 GB, which is under a year for anything busier than 14 MB/day, and
+	# nothing here has ever been that quiet: 19-24 MB/day on a slow week, and
+	# 617 MB/day at the Hacker News peak, where 52 rolls is EIGHT DAYS. 365
+	# rolls (~2.6 GB gzipped, at the ~14:1 these compress to) makes
+	# roll_keep_for the constraint that actually decides, up to ~100 MB/day.
+	#
+	# roll_keep_for must be explicit — lumberjack's default silently deletes
+	# after 90 days.
 	log {
 		output file /var/log/caddy/catcrafts.net.log {
 			roll_size 100MiB
-			roll_keep 52
+			roll_keep 365
 			roll_keep_for 8760h
 		}
 	}