"20% infill" is not a universal default that fits every part
Slicer defaults around 15-20% infill are a reasonable generic starting point, but they are tuned for an average case, not your specific part. A purely decorative object rarely needs that much; a load-bearing bracket often needs more wall count rather than more infill at all (see the Wall-versus-Infill Strength Planner); and a part with a wide unsupported top surface needs enough infill underneath regardless of load type, or the top layers sag or show gaps.
How the plan works
A base density band comes from your part's role: decorative parts start low, light functional parts land in a middle band, load-bearing parts start considerably higher. A wide top-surface span (over about 20 mm) pushes the recommendation up regardless of role, since the top layers need infill underneath to bridge over cleanly. Your time priority nudges the final number up or down within that logic.
Worked example
A light-functional part, 15 mm top-surface span (not especially wide), medium time priority: the tool recommends about 20% infill as a starting point, in the middle of the light-functional band, with no extra push from the top-surface span since it is not wide enough to need special treatment.
A common mistake
A frequent error is raising infill density as the first response to a part that fails structurally, when the actual weak point is often the wall count or the load direction relative to layer lines instead. Infill contributes real but limited strength for most loads; check the Wall-versus-Infill Strength Planner before assuming more infill is the fix. A quick way to sanity-check this is to look at where the part actually failed; if the break ran across a wall rather than through the internal lattice, more infill was unlikely to have helped in the first place.
Limitations
This compares density bands qualitatively using load type, top-surface span, and time priority as inputs; it does not calculate a safe load or a guaranteed structural outcome for a specific infill percentage.