The first 72 hours, from pallet to stable hashrate

Hardware failures are not evenly distributed over a machine's life. They cluster hard at the beginning, which is why burn in exists as a concept. The way you handle the first three days decides how much of that early failure lands inside a claim window and how much lands on you.

1. Receive before you unpack

Count cartons against the packing list while the driver is still there. Photograph any damaged packaging immediately, because damage attributed after the freight leaves is nobody's problem but yours. Note the serial numbers as you unpack rather than later from memory, and keep the packaging until the units have run.

2. Let them acclimatise

A container that crossed an ocean or a truck that crossed a winter arrives cold. Powering up cold hardware in a warm room condenses water inside it. Give the units several hours at room temperature before anything is switched on. This single step prevents more dead on arrival units than any other, and it costs nothing but patience.

3. Power up in stages

Do not energise the whole batch at once. Bring up one shelf, confirm it, then the next. Powering up in stages keeps inrush current inside what the breakers expect, and it means a fault is traceable to a handful of machines rather than to a room. It also gives you a working reference to compare the rest against.

4. Network and pool before scale

Set up the pool, the worker naming scheme and the monitoring on the first shelf, and get it right there. A naming scheme invented on shelf one and applied consistently is the difference between finding a failing unit in seconds and walking the aisles with a thermal camera. Fix static addressing or DHCP reservations now, not after 200 machines have taken random leases.

5. Watch the first day closely

• Hashrate against nameplate per machine, not per fleet. A fleet average hides three dead boards.
• Chip temperatures and the spread between them. A wide spread inside one machine usually means airflow, not silicon.
• Rejected shares. A rising reject rate is normally network or pool configuration, not hardware.
• Intake temperature at several points in the room. The first real test of whether your hot and cold separation works.
• Anything that restarts on its own. Note it, do not wait for a pattern.

Raise problems inside the window

Every quote states a period within which a discrepancy or a dead unit can be reported. That window is short by design, because after it nobody can distinguish a factory fault from a site fault. Test the whole batch inside it, including the units you did not plan to deploy immediately. A spare that turns out to be dead in month four is a spare you paid for.