A warehouse cutoff planning board that shows if every carrier will make it

The warehouse ships close to 3,000 orders a day across several carriers, and each carrier collects at its own time. The old board ran on a single 23:15 deadline. On one evening it read "on track" while hundreds of orders had already missed an earlier carrier's pickup.

The open queue was also wrong. Orders already shipped or with a cancellation request were counted as pickable, so the backlog was over-reported by roughly 40%. Shift leads were planning staff against a number nobody trusted.

Around the board, the daily work still ran on Excel and screenshots. The team leader rebuilt the day schedule by hand each morning, rosters arrived as images from a planning tool with no export, and SLA results were worked out after the fact.

I rebuilt the order count from the ChannelDock fulfilment center API. Every non-pickable order gets exactly one exclusion reason (on hold, pending, backorder, address check, future date, blocked, completed or cancelled), so the buckets always add up. A verification script checks the classifier against independent predicates. Carriers are grouped by shipment method ID, not name, because two "DHL" services leave on different runs.

The forecast uses 15-minute blocks. The baseline is the six most recent matching weekdays, weighted 30/25/18/12/8/7, with outlier capping per block. A live day factor adjusts it by how today is running, and may move at most 0.10 per quarter hour so every screen shows the same stable number. A deadline planner then plays the floor forward a quarter hour at a time, earliest cutoff first.

SLA emails run from a cron every 5 minutes, but the cron decides nothing. Each run asks for emails that are due and not yet sent, and claims each one in a Firestore transaction before sending, so two runs can never both deliver it. A 5-minute refresh job records order blockers all day, an hourly job stores snapshots, and a nightly job rebuilds missing history inside the API rate budget.