Quick answer: Launch-day test plans usually fail because they're aspirational. A 3-hour, 6-person, room-by-room playthrough plan that runs in T-24h catches more than 30 pages of spec.
By launch day, the QA spec is two months old. A pragmatic 3-hour structured playthrough finds more than the spec ever will.
3 hours, 6 testers
Three pairs: gameplay, UI, performance. Each pair owns 20 minutes per area; the rotation covers every area twice.
Build the matrix
Rows = areas, columns = pairs. Each cell is one 20-minute pass. The matrix lives on a shared doc; cells turn green as passes complete.
Triage in real-time
One engineer-on-call. New bugs land in a 'launch-day' tag in the tracker; engineer triages every 15 minutes. Sev-1 = stop launch; sev-2 = hotfix candidate; sev-3 = note for patch 1.
Time-box decisions
T-2h: review the launch-day tag. Sev-1 count drives go/no-go. If >0, push launch; if 0, ship and accept the sev-2 list for patch 1.
“Launch day quality is what your final playthrough catches, not what your spec claimed to cover.”
Save the matrix template. The next launch's plan is a copy-paste plus 30 minutes of customization.