An analytics dashboard's icon set grew to 300 icons from several contributors and now mixes stroke weights, rounded and sharp corners, and some perspective drawings. How do you restore consistency?

## Why icon sets drift An icon set starts coherent because one person or a small team draws it. Drift begins as it grows: new contributors join, product teams draw icons under deadline, icons are borrowed from outside sets to fill a gap, and the earliest icons were drawn before the style was written down. Typical symptoms in an analytics dashboard's set of a few hundred icons: - **Mixed stroke weights**: some icons drawn heavier, some lighter, and some drawn at one canvas size and scaled to another, which thins or thickens their lines. - **Mixed corners and terminals**: rounded outer corners beside sharp ones, rounded line caps beside flat ones. - **Perspective intruders**: an isometric database cylinder or a shaded cube among flat, front-on shapes. - **Metaphor duplicates**: two pictures for one concept, often a symptom of the same process gap. ## Step 1: audit the whole set at real size Put **every icon on one sheet**, grouped by category, at the size it is actually used. Drift is a relational defect: an icon only looks too heavy **next to** its neighbours, so auditing icons one at a time, or zoomed in, misses it. For each icon record: | Check | Measure against | |---|---| | Stroke weight | The value most of the set uses at the reference canvas | | Terminals and joins | The majority cap and join style | | Outer corner radius | The majority radius on shapes of similar size | | Fill treatment | The variant rules (default and state variant) | | Perspective | Flat, front-on or simple side view | | Metaphor | The concept map, for duplicates | The output is a tagged list: which icons deviate, and on which attribute. ## Step 2: decide the canonical style Pick the target style **deliberately**, usually the one that most of the set and the most-seen icons already follow, since that minimises redraws and matches what users already recognise. Do not let the most recent contributor's taste decide it. Then write it as **measurable values**: stroke width and outer corner radius at the reference canvas, the cap and join choice, the variant rules, flat drawing, and a short documented list of allowed optical exceptions, for example a slightly lighter stroke on unusually dense glyphs. ## Step 3: redraw by exposure A few hundred redraws is weeks of work, so order it by **how often users see each icon**: 1. Navigation and global chrome, seen on every screen. 2. Table row actions and chart toolbar icons, repeated many times per screen. 3. Icons on secondary settings and rarely visited pages. This delivers the visible improvement early and lets the long tail follow over later releases, tracked against the audit list so nothing is forgotten. ## Step 4: ship without breaking consumers Keep each redrawn icon's **identity**: its name, its meaning and its footprint. If only the drawing changes, consuming teams on the web and on native mobile receive the fix as an ordinary update with no code changes, and the change is purely visual. An icon whose metaphor must change is a different kind of change and is handled as one, with notice to the teams that use it. ## Step 5: stop it from happening again The audit is wasted if the process that caused the drift is unchanged: - A **starter template** with canvas, stroke, caps and joins pre-set, so every new icon begins on-spec. - A **review checklist** for new icons covering each attribute in the table above, run before an icon is published. - **Automated checks** on the vector sources where practical, for example a script that flags stroke widths or corner values outside the spec. - A **side-by-side preview** of each new icon among existing ones at real size as part of review. - A **policy on external icons**: borrowed icons are redrawn to the house style, never dropped in as they are. ## What not to do Resist normalising everything with a blind batch operation. Forcing one stroke width onto every path breaks icons with deliberate optical exceptions, cannot fix corners, perspective or metaphors, and hides the process gap that caused the drift. Drift is a design problem with a process cause, and the fix has to address both.