The crontab called `bash -lc '... node ...'`, and every part of that failed silently. node was not on the PATH -- fnm is wired into fish, not a bash login shell -- so three publishes a day died before they started, and the path an interactive shell reports is an fnm multishell under a PID directory, per-session and useless in a crontab. Nothing updated the automation checkout, so it ran whatever it was sitting at. Nothing committed the ledger a publish mutates, so lith never learned about a new reel. And none of it announced itself: three dead runs today left ninety bytes in a log nobody reads. oskiewar-clockwork.sh resolves node from the installations, fast-forwards the checkout, runs the slot or the insights pull, pushes the ledger, and leaves a timestamped line plus a heartbeat carrying the exit code. The ledger only reaches lith with a deploy, so git alone would leave the wall quoting the last deploy's figures forever. Insights are posted to /api/oskiewar-stats as well and stored per reel, and the GET merges stored rows over ledger rows so the newer number wins. That POST needs the wall key, the same secret that is allowed to read the figures back out.