Campus wireless is where the gap between a designed network and a deployed one shows up fastest. A coverage-based design will produce a signal everywhere and a network that performs badly in exactly the places it matters most.
Coverage is the easy half
Signal strength is straightforward to predict and straightforward to demonstrate. It is also not what users experience. Performance in dense environments is governed by capacity and interference, both of which get worse as you add access points to improve coverage.
The failure pattern is consistent: a design based on coverage produces too few access points for a lecture hall or trading floor and too many in corridors, where they interfere with each other and with the rooms that needed the capacity.
Design to device density
The useful input is not floor area. It is how many devices will associate in each space at peak, and what they will be doing. A three hundred seat auditorium during a session and the same room empty are entirely different problems.
That means gathering occupancy data per space, assuming two to three devices per person, and designing capacity per area rather than distributing access points evenly. It also means accepting that some spaces need directional antennas and reduced power rather than more coverage.
The things that get missed
Roaming. Users move. A design that performs well at every fixed point can still handle transitions poorly, which presents as dropped calls and stalled sessions while walking between buildings. Voice and video traffic are far less tolerant of this than data.
Interference sources. Other networks in adjacent buildings, and non-network sources in industrial and clinical environments. A predictive design cannot see these; a site survey can.
The wired side. Every access point needs switch capacity, PoE budget and uplink bandwidth. Wireless designs that ignore the access layer produce a switch replacement project nobody budgeted for.
Validation. A post-installation survey against the design is the only way to know what was actually delivered. Predictive modelling is a design tool, not evidence.
A note on sequencing
Survey, design, deploy, validate — and treat validation as a project deliverable with acceptance criteria. Wireless is the one network layer where the difference between the drawing and the building is large enough to matter.