The time axis needs to stretch from seconds to months depending on what's being diagnosed

What it is

An adjustable time-axis scale lets an operator move freely between second-scale transients, hour-scale control-loop response, and month-scale drift, matching the window to whatever diagnostic hypothesis is currently being tested, instead of being locked to one fixed window size. Zooming itself has to preserve time semantics clearly — zooming in should reveal detail within the same real interval, not silently switch to a different one — and it has to allow a quick return to a shared reference window, so an operator does not lose track of which stretch of history they are even looking at after a few zoom operations.

Why it happens

A fixed time window is not fair to every kind of signal: set it too short and a drift that only becomes visible over months looks like a flat horizontal line with no discernible change; set it too long and a transient lasting a few seconds compresses into a spike too small to read, equally invisible. What multiscale viewing actually does is change the visual ratio between signal and noise by changing the window size — the same underlying data reads clearly at the right window and gets swamped by noise or by the compression of the time axis itself at the wrong one. An easily missed detail here: if a zoom operation does not preserve an event's absolute position on the time axis, an operator who zooms in or out loses track of how the segment now on screen relates in time to the segment they were just looking at, and the before/after correspondence breaks.

Where it stops holding

Too much zoom freedom has a cost of its own: an operator can keep adjusting the window until it happens to produce a shape that supports a conclusion they already believe, a form of window-shopping confirmation bias that is hard to catch after the fact. Smoothing methods such as a moving average also do not always produce consistent results across scales — recomputing the same raw data at a different window can change the apparent shape of a trend, something easy to overlook when comparing across scales. Regulatory records may additionally require fixed sampling rates and immutable raw data, a constraint that is a separate concern from flexible diagnostic zooming and cannot be satisfied by the same processing pipeline.

Applying it

Offer a handful of presets tuned to common diagnostic scenarios plus a continuously adjustable zoom control to cover whatever falls between them; at every scale, keep the current window's start and end time, the data's sampling resolution, and a thumbnail overview highlighting where the current zoomed range sits within the full history continuously visible on screen. Validate using historical cases with a known answer whose fault signal genuinely spans multiple time scales, and have operators choose their own window and explain the choice, checking whether their selection process misses context that only exists at a different scale.

Related

• Same group: Y3.07.1 Trend display for rate and direction · Y3.07.3 Misleading multiple axes · Y3.07.4 Event-centered historical navigation
• Nearby: Y1.03 Trends and rate of change · Y1.02 Overview and detail
• Search terms: multiscale trend · adjustable time-axis scaling · diagnostic time horizon · trend window selection