Part 4 · Play
Reading the learning screen and acting on it
Four questions to ask it, and what to change for each answer.
Who this is for
You open the learning screen, look at it, and close it again.
Before you start
- At least a few weeks of outcomes recorded, including silence.
How long
30 minutes, once a fortnight. Not daily.
The sequence
- 01
Ask first whether the system arm is beating the baselines. If it is not, nothing else on the screen matters yet.
/learning2 minutes
You should see: A comparison against three baselines that keep running on reserved budget.
- 02
Ask which hypothesis is leading, and check the interval rather than the mean. A wide interval means you do not know yet.
/learning5 minutes
- 03
Ask what is being killed and whether you agree. Nothing is killed below the minimum sample.
/learning5 minutes
- 04
Ask which channel produced the conversations, from the per-channel table rather than the total.
/learning5 minutes
- 05
Change exactly one thing: pause a hypothesis, change a segment, or sharpen a message. Then wait for enough resolved actions before reading it again.
/learning and /venture15 minutes
You are done when
- You can name the leading hypothesis and its rate.
- You changed one thing and know what you are waiting to see.
The commonest way this fails
You read it daily and change something each time. Every change resets the sample, so you never reach the floor where any of the numbers separate.