Reading Rejected Shares on a BTC-Only Pool Dashboard

When a client called us about “pool luck” last spring, their dashboard showed 4.2% rejected shares across four identical units. Pool variance does not cluster that tightly — something local was wrong.

Step one: match firmware profiles

We remote-checked each unit. Two ran a custom overclock from a Telegram profile; two ran stock. Comparing them was meaningless until all four matched. After reflashing to factory, rejections on three units dropped under 0.8%.

Step two: clock drift and network

Stale shares spiked every time their cheap switch rebooted. Replacing the switch was cheaper than RMAing boards that were never faulty.

Step three: board-level isolation

The remaining unit still rejected 3.1%. Lab swap of hash board three confirmed a failing ASIC chain. Client used our serial photos in a warranty claim.

Takeaway

Log rejected share rate per unit, not farm average. A single outlier in a homogeneous row is almost never the pool’s fault.