WHAT THIS PHASE PRODUCES
You leave with
- A synthesis doc that grades the hypothesis against the pre-registered rule
- A list of the strongest dissenting evidence
- A written iterate-or-commit decision your advisors can read in five minutes
THE MENTAL MODEL
How to think about this phase
Synthesize is where the loop pays off. Done well, you leave with a graded bet and a written decision. Done badly, you leave with a folder of chat logs and a rationalisation.
The honest move is to grade the hypothesis against the rule you set in Test, not against the result you wish you’d gotten. If the rule failed, the bet is wrong as stated, even if a different bet inside the same general space might still be alive. Naming the wrong bet correctly is what makes the next loop sharp.
Look hardest at the dissenting evidence. The customer who hated it, the one who asked “what about X”, the one who paused for ten seconds before answering: those are the seams the next iteration must address. Pattern-matching only the wins gives you a confident wrong answer.
End with a written iterate-or-commit decision. Iterate means a refined hypothesis enters Loop 2 with the carry-forward artifacts wired up. Commit means the bet survived enough pressure to graduate into a real engineering effort. There is no third option called “keep going as is”. That is how loops become months.
COMMON TRAPS
What goes wrong in this phase
- Grading against a softer rule than the one you set in Test. This is the most common form of self-deception in the loop.
- Counting enthusiastic participants twice in your head and skeptical ones half. Your synthesis must use literal counts.
- Letting the decision linger. A loop without a written close becomes context the next loop has to relearn.
- Writing the synthesis doc only for yourself. Write it for an advisor. That constraint forces honesty.
WHAT GOOD LOOKS LIKE
A worked example
SETUP
From the Test example: 10 PM sessions, paste rate 7/10, share rate 2/10 against thresholds of 6 and 4.
OUTPUT
After Synthesize: a one-page doc that reads, in part, “Decision: iterate. The paste behavior is real (7/10, transcripts confirm they pasted live problems, not toy ones). The share behavior failed because the generated outline was useful but not branded enough to feel safe to forward. Loop 2 hypothesis: when the outline is rendered with the team’s shared vocabulary and a one-click ‘share to Slack thread’ flow, share rate clears 4/10 in another 10 sessions. Refined P0 next loop: vocabulary detection + Slack share. Carry-forward: the 10 transcripts, the 7 confirmed pasters as recruitable Loop 2 contacts.”