How I structure my deep work sessions when testing prompts at scale

One thing I've learned after months of prompt engineering:
unstructured testing sessions are a productivity killer.

When you're iterating on 20+ prompt variations, it's easy to
lose track of time — and worse, lose track of which version
you were actually testing when you got that good result.

My current workflow:

1. Set a fixed time block (usually 25–40 min) before starting
2. Define ONE variable to test per session — temperature, tone, structure, etc.
3. Log outputs in a simple table as I go
4. Hard stop when the timer ends, review before continuing

For the timer part, I've been using https://fullscreencountdowntimer.com/
— it's fullscreen, distraction-free, and keeps me honest
about when a session actually ends. Small thing, but it
changed how focused my testing feels.

Anyone else use time-boxing for prompt iteration?
Curious what workflows are working for people here.