Explain a performance update only as far as evidence supports

Start with the measured operation. Page loading, completing a task and processing a file are different events. A number without a start point, endpoint and environment is not a usable creative claim. Ask the measurement owner for the details before drawing a chart.

Record the dataset, device or server conditions, repeated-run method and reported summary. Keep limitations beside the result. A demonstration recording may illustrate the operation but does not substitute for a measurement record.

• Operation and timing boundaries.
• Compared versions and configuration.
• Input size and relevant input characteristics.
• Number of observations and how the summary was computed.
• Variability, excluded runs and known limits.
• Owner approval for the exact claim wording.

Imagine a fictional report tool whose team says a new cache reduces repeated preparation work. No benchmark values are supplied. The verified behavior is that unchanged sections can be reused during a repeat preparation; the team has not established a general time reduction for every report.

A defensible announcement can explain the reuse behavior and its conditions. It should not display a fabricated stopwatch comparison, a percentage badge or an illustrative number that looks like a measured result.

• Unsupported: “Reports are twice as fast.”
• More precise when only behavior is verified: “Repeat preparation can reuse unchanged sections in the supported workflow.”
• Evidence still needed for a speed claim: comparable source runs, input definition, timing boundaries and approved summary.
• Visual alternative: mark reused and recomputed sections rather than animate invented elapsed times.

When a valid comparison is supplied, keep its scope in the scene. Identify the tested operation and input category, and label whether the value is a median, range or another approved summary. Use the supplied precision rather than adding decimal places to make a result look scientific.

Keep chart scales and timing treatments consistent. A cut that shortens one sequence more than the other can suggest a larger improvement than the measured record supports. If the film is illustrative, label the timing as editorial rather than measured.

The headline, chart, narration and thumbnail should express the same bounded result. A careful source note cannot rescue an unqualified “instant” in the title. Ask the measurement owner to review the final composition, because visual scale and labels affect the claim too.

If the evidence is inconclusive, communicate the change and the testing scope honestly. The useful outcome is a viewer understanding the update, not a numerical headline at any cost.