Start with the symptom, not the constant

When a Creep appears stuck, first separate command rejection, path failure, fatigue, and target/range logic instead of repeatedly calling moveTo().

When harvest() has no visible effect, confirm the return code, target type, range, WORK parts, and available resource before rewriting role logic.

When the Spawn queue produces nothing, separate busy state, Energy, body/name arguments, and RCL limits instead of blaming the queue generically.

Controller pressure is rarely one API problem; inspect ticksToDowngrade, Upgrader Energy, range, hauling, and the real upgrade return code as one chain.

When a Link does not transfer Energy, inspect cooldown, sender stock, receiver capacity, target identity, and the real transferEnergy() result together.

When Market deal/createOrder fails, inspect order arguments, Credits, resources, Terminal state, and transaction Energy cost rather than price alone.

High CPU usually has no single return code. Measure hotspots, bucket, and tick load first, then decide whether to reduce scans, cache work, or spread work across ticks.

When logistics stalls, identify whether withdraw, carrying, movement, transfer, or destination capacity is the broken stage and store each result separately.

When build() or repair() has no visible effect, inspect the real return code, target type, range, WORK parts, Creep Energy, and target state first.

When a Tower does not produce the expected action, confirm target, Energy, ownership, and the real action result before judging range falloff or combined Tower output.