Practice / A diagnostic
The hypothesis ratio
A five-minute way to tell whether a build relationship is resolving ambiguity or returning it.
The formula#
On any ticket, thread, or pull request, count only the building side’s comments. Ignore the answers coming back.
- H — comments containing a hypothesis. Look for “we inferred”, “we propose”, “our take”, “we will apply X unless you say otherwise”, or a filled-in table where the only ask is “confirm?”
- D — comments that only ask. Look for “please clarify”, “which is correct?”, “what happens when…”, or a table with empty cells.
- The ratio is H divided by (H plus D).
Reference ranges#
The thresholds aren’t arbitrary. They correspond to the point at which one product owner can keep up with the flow of ambiguity coming back.
- 70 percent and above — the building side is resolving ambiguity. The other side confirms or redirects. This is the target.
- 50 to 70 percent — judgment on some of it, and the rest comes back. Workable, and slower than it looks. Probably most of the industry.
- 30 to 50 percent — occasional hypothesis, dominant posture is “please decide”. Every ticket accrues calendar time.
- Below 30 percent — most decisions are coming back for one person to make. Worth a conversation about how the work is shared.
Try it on a thread of your own
How to count quickly#
Read the first two or three sentences of each comment. A verb of proposal counts as H. An opening of “please clarify” or “could you advise” counts as D. If a comment carries one hypothesis and three unresolved questions, count it where the center of mass is. When genuinely unsure, count it as D.
A twenty-comment thread takes about five minutes. Run it monthly on whatever is busiest and watch the trend rather than the number.
Why it’s a leading indicator#
By the time a ticket has spent three months in transit with thirty comments, the cost is already paid in calendar time. The ratio tells you the operating mode at comment three, while the shape can still change.
It’s also resilient to noise. One clarifying question can be entirely legitimate. Ten in a row with no hypothesis is a pattern.
Limits#
Good engineers ask questions. The ratio measures whether the questions arrive with judgment attached or in place of it.
Genuine business questions that nobody outside the room could infer are legitimate escalation. Those should stay.
Use it privately, to understand a relationship and to make the case for changing how it works. Pasting the number into a ticket wastes it.