Exercise 2: A Checkpoint at 180 Days — Possible Solution ==================================================================== RESULTS ------------------------------ 90 days: 1 alternative, 0 reasons captured 180 days: 1 alternative, 0 reasons captured Identical results at 90 and 180 days - the count doesn't decrease further between these two checkpoints. WHY IT DOESN'T KEEP DECREASING INDEFINITELY ------------------------------ The chapter's own recall_fraction() function has an explicit floor: any delay beyond 30 days returns a fixed 0.15, rather than continuing to decay toward zero the longer the delay stretches. This floor represents a real, honest limit on the model: some information survives indefinitely regardless of how much time passes, because it isn't purely dependent on any one person's memory - committed code, chat history that's still searchable, ticket descriptions, and email threads all persist as artifacts even after nobody remembers the conversation itself. The model deliberately doesn't claim an ADR written a year late captures literally nothing; it claims a late ADR captures a small, roughly constant fraction of what a same-day one would, which stays genuinely useful as a rough estimate rather than a precise memory-decay curve. WHY THIS WORKS AS AN ANSWER ------------------------------ This is an important, honest limit on the chapter's own simulation: a model that let recall decay all the way to 0 as time went to infinity would overstate how bad extremely late documentation actually is. The floor keeps the model's own claim modest and defensible - "significant, real information is lost quickly, and what's left levels off at a persistently low but nonzero fraction" - rather than implying a year- old decision could never be documented at all, which isn't realistic either.