An early failure weighs more than an equivalent late one, because there is no success history to offset it

What it is

The same crash: a force-quit on first open, versus a force-quit in month three. The objective fault is the same. The first-day one hurts more, because there is not yet a sample of “it usually works” to offset it. An early failure weighs more than an equivalent late one. The hedge is a history of success, not time itself.

Late still hurts, but the estimate already holds positive samples; the failure is an update, not the first portrait.

Why it happens

Trust starts near blank. The first observation is weighted extremely hard and is written as the subject’s character. An early failure writes the character as “unstable”; later successes have to overturn that first stroke. A late failure meets an already-written “mostly works”; it changes degree. Muir’s update from experience, at sample size one, has no variance to eat; the point estimate is that one time.

Products also often give the roughest version to the first cohort. Early failure is therefore not a random landing; it is placed where the weight is largest. Stack the asymmetric unit update on the first stroke, and withdrawal can happen before any handover has been built — it looks like “no trust to collapse,” it is trust never formed.

Studying it

Insert the same kind of failure at trial 1 and at trial n (all successes before n). Measure the drop in handover, whether they continue, and how they describe “what it is like.” Independent variables: position of the failure, number of prior successes, whether the failure is attributed to “still in trial.” Dependent variables: ratio of drops, subject description (unstable / occasional), retention.

n must be large enough that a “mostly works” portrait has already formed. A “trial 2” that is too near still counts as early.

Where it stops holding

If the user arrives with outside experience (a colleague’s recommendation, a previous version), the first stroke is no longer blank and the early advantage narrows. If late is the first time they hit a high-stakes task, that “late” is still early in the consequence band. An incident severe enough to empty long-built trust does not care about position — that is the collapse threshold. A streak happens after samples exist, the opposite direction from “the first stroke was a failure.”

Applying it

• Put the most stable path on first use: a short task that can be finished, a slice of known high hit rate. Do not put untested capability on the first screen.
• Operate first-week failures as incidents, not as “early users should put up with it.”
• If the first stroke has already failed, give one completable success on the same task at once, specifically to offset that first portrait, rather than waiting for natural traffic.
• Check: compare next-day retention after “first-day failure” versus “the same failure at n.” A gap on first day means the early weight is real in the product; see whether the first screen is carrying untested capability it should not.

Related

• Same group: L5.10.1 One failure weakens trust more than one success strengthens it · L5.10.2 Users generalise a failure in one domain to the system's whole capability · L5.10.4 How a failure is handled can partly offset the damage; admitting the error beats downplaying it · L5.10.5 Restoring trust takes far more successes than the failures that caused the damage
• Nearby: L5.09 Overtrust and Trust Collapse · L5.04 Collapse of Trust · L5.03 Trust Calibration
• Search terms: early failure · first impression of reliability · no success history