Overview

The by-year charts start at 2003 because the record does. Chicago's licence data begins in 2002, so every restaurant already trading that day enters it with a 2002 start: 4,163 of them, against roughly 700 in a normal year. That bucket is not an opening cohort, it is a census of what already existed, and it flatters itself: it survives four years at 63.9% against 47.4% for 2003, because those businesses had already proven they could trade before the record ever saw them. The same floor bounds what the survival model is trained on. The counts above are unaffected. A restaurant that opened in 2002 is a real restaurant.

The last few years read low because closures take years to surface, not because fewer restaurants closed. A closure here is dated at the earliest evidence we hold: the licence lapse, an inspector recording the business out of business, or the operator announcing it, whichever came first. That is the tightest bound the record supports, because every one of those signals is a date by which trading had already stopped. What it cannot do is tell us sooner. Measured against cohorts whose outcomes are now fully known, the public record reveals about 18% of closures within a year and is only complete at three, so the bars nearest today are the ones still filling in. Read this chart as a decade rather than as one year against another, and do not read the right-hand edge as a recovery.