part four · assumption surfacing
Assumption Surfacing
Naming what we believe but haven't witnessed.
An assumption is a claim the team is making about the world that has not been witnessed in field observation. Discovery's job is to separate the claims from the witnessing — to know which is which, and to hold the unwitnessed claims as honestly named uncertainty rather than as silent confidence.
Why this is a separate practice
Most teams have assumptions. Few teams write them down. The ones that don't write them down do not have fewer assumptions — they have invisible ones, which the chain cannot test or update.
Assumption surfacing is the deliberate act of saying: here is what we currently believe, and we have not yet seen it, and we are not going to act as if we have until we have.
Three states
Every claim sits in one of three states.
| State | Meaning | What happens next |
|---|---|---|
| Witnessed | We have observed this directly | Brief can rely on it |
| Inferred | We have evidence but not direct observation | Cite the evidence in the brief; flag for cycle check |
| Not witnessed | We believe this but have not seen it | List explicitly in the brief's assumption space |
A brief whose claims are all in the witnessed column is a strong brief. A brief that mostly contains not witnessed claims is a Discovery brief that has not yet finished Discovery — the work of moving claims from not witnessed to witnessed is what's left.
Writing assumptions in the brief
A brief carries its assumptions in a labelled section. They sit there visibly, not hidden in the prose.
Brief: Grading Flow v2 (excerpt)
Assumptions (witnessed):
• Graders use a second monitor for the rubric. (Observed in 6/6 sessions)
• Hebrew names cause display issues. (Observed in 4/6 sessions)
• Graders work in batches of 5 with short breaks. (Observed in 5/6 sessions)
Assumptions (inferred):
• Graders prefer keyboard shortcuts over mouse nav. (User comments in 3/6 sessions; not directly observed)
• The 47-min mean cycle time will fall to <15 min. (Modelled, not observed yet)
Assumptions (not witnessed):
• Graders will adopt the new flow without training. (We have not tried it)
• Hebrew name fix will hold under all unicode forms. (We have not tested non-grader names — student names — in production load)
• The IT admin can roll the change out smoothly. (We have not talked to the admin)Each not-witnessed assumption is a candidate for a discovery question or a cycle check.
The cycle's job, against the assumption list
A cycle either witnesses an assumption (moving it from not witnessed to witnessed), or it does not. The model update at the end of the cycle (Volume V Part 6) walks the list and marks each one.
This is the chain's most reliable form of learning. Assumptions that are confirmed across multiple cycles compound the team's confidence. Assumptions that are contradicted produce sharper briefs. Assumptions that remain not witnessed across multiple cycles are flagged — they are usually the assumption hiding the largest risk.
Why teams resist this
Three reasons, all wrong, all common.
- Listing assumptions makes us look uncertain. Listing assumptions makes the team honest. Hiding them makes the team confident in the wrong direction.
- We don't know all of them. You don't have to. List the ones you can name. The retro will surface others.
- It feels like overhead. It is the cheapest insurance the chain produces. A two-minute list at brief time, that the cycle later resolves, is one of the highest-leverage practices in the corpus.
Assumption space and prediction space
A prediction (Part 9) is an assumption about a change. An assumption is a claim about a state. They are kept separate because their checks are different.
- A prediction is checked against measurement: did the time fall to under 15 minutes?
- An assumption is checked against observation: did the IT admin actually roll the change out smoothly?
Both are checked. Both feed the model update. They are different artifacts because they answer different questions.