A large first–steady gap makes users misjudge overall performance

What it is

Four seconds the first time, 0.4 s thereafter, and people may still file the product as “slow”. The other way around — instant first hit, then frequent 3 s stalls — and they say “unstable” or “it gets slower”. When first and steady sit too far apart, overall performance is misjudged: one or two samples set the product’s tempo, and those samples happen to be the least typical ends of the distribution. A large gap is not excused by “cold start is normal”. It is a calibration failure.

Why it happens

Global impression is sparse sampling plus peak memory. The first hit is almost always sampled — there is no product experience without it. Steady state needs many trials to enter the impression. If first is an extremely slow sample, the peak writes “slow” and later speed must repeat to overwrite it. If first is an extremely fast sample, it writes a “fast” anchor, and every later drop back to cold-start level is read as regression or fault. A gulf in either direction stops the internal model from representing the real distribution.

Demos and benchmarks step on this. The first hit on a sales floor is warmed to look like steady state; the customer’s first hit at home is a true cold start; reputation is set by the one at home. Labs measure only the hot path; production logs are full of every day’s first.

Where it stops holding

A small gap (say 1.2×) does not mis-set the tempo; people may not even feel two bands. A large gap whose first hit is clearly “first open after install / update” can be attributed to setup rather than product speed — if it happens only then, and something like “finishing setup” explains it. Every daily open as slow as first makes the explanation fail. Experts warm up on their own (tap once idly); misjudgement happens to people who do not.

Applying it

• Treat the first-to-steady ratio as an experience metric. Past about two or three times, pull first closer or make the difference a visible preparation, not a silent gulf.
• Time demos and screenshots on a cold start, or label “warmed”. Using warm as the default calibrates outsiders wrongly.
• Do not promise market numbers from steady state. What people meet every day is the first hit.
• How to check: two groups, one doing the main task immediately after cold start, one after warmup. Ask “is this product fast?” If the answers oppose, the gulf is already speaking for the product. Bring first into the same order of magnitude as steady; the answers should converge.

Related

• Same group: I1.07.1 The first response carries extra init cost and is usually slower than steady state · I1.07.2 Patience is higher for the first action than for later repeats · I1.07.3 Warmup or preload can move first-hit cost to a moment the user does not feel
• Nearby: I1.05 Latency jitter · I2.07 Perceived performance
• Search terms: performance miscalibration · first impression latency · cold versus warm gap