At 03:12 your AI gives a confidently wrong answer. Nothing pages. It is up, fast and wrong, the same way for every user who asks. That part is identical at every organisation. What separates them is how long it takes somebody to find out.

Ask five people at your company which maturity stage your AI is at and you will get five answers, because each is measuring something different. This model gives them one number instead: the time between an AI system being wrong in production and a human knowing.

Stages 1 and 2 find out in about eleven days, when a customer mentions it. Stage 3 builds the measurement and brings detection down to four days, via a dashboard. Stages 4 and 5 get there in nine minutes. Most organisations are at Stage 2 and describe themselves as Stage 4.

The gap between days and minutes is three things, and two of them are not engineering: alerts on quality drift, a rollback performed in the last 30 days, and a named owner with on-call rotation. The most common blocker in every assessment we ran is the third. It costs nothing, and nobody owns it.

Two questions cannot be answered generously. Pick an answer from three months ago: can you reconstruct the input, model version, retrieved context and decision? And when was a rollback last actually performed? If the answer is 'we have never needed to', you do not have a rollback.

Stage 4 is the right target for most organisations. Stage 5 makes it reusable, and is only worth it once you run five or more production AI systems. The infographic walks all eight frames and ends with a 15-minute assessment that gives you your stage, the gate blocking the next one, and what clears it.