Reliability, Framed Honestly

Reliability for compute is availability plus predictability plus graceful failure; the burner is honest about where it is strong and where it is not yet.

Three things reliability means

Operators fold three properties into the word reliability. Availability is how much of the time the source is up. Predictability is whether outages are scheduled and bounded rather than sudden. Graceful failure is whether a fault degrades output smoothly or drops it instantly. A source can be strong on one and weak on another.

On raw availability the burner is not yet where a data center needs it — that is the availability gate, and this site never softens it. On predictability, a fusion plant's dominant downtime is planned maintenance of magnets and vacuum systems, which can be scheduled and coordinated with the load. On graceful failure, the design intent is that a plasma trip drops one member of a hybrid while the battery and grid carry the load through the transition.

Honest framing changes how the source is sold. It is not a Tier III box you drop in and forget. It is a firm, low-water base layer whose outages must be planned around and backed up. Presenting it that way is what earns trust with the engineers who will actually review it, and it is the only framing consistent with a machine that has not yet been built.

Reliability, then, is a system outcome the operator engineers, using the burner as a predictable, gracefully-degrading base member — not a guarantee stamped on a single reactor.

• Reliability = availability + predictability + graceful failure
• Burner: weak on raw availability, designed for predictability and graceful degradation
• Planned maintenance dominates fusion downtime
• Trust comes from stating the gate, not hiding it