part six · the model update

The Model Update

The step most teams skip. Where learning survives the conversation.

Events in this phase

Held immediately after the retrospective. Owned by the PO. Output is a written change to a chain artifact — the model the team carries into the next cycle. Visible to anyone the model affects — usually the whole trio plus leadership.

The signal reading recorded what happened. The retrospective recorded what the team is changing about how it works. The model update records what the team now believes about the world that it didn't believe before, and writes that belief into the artifact that will shape the next cycle.

This is the step most teams skip. The conversation happens. Insights are spoken. People nod. The next cycle begins, and somewhere — in a brief, in a story, in an estimate — someone makes the same wrong assumption that was named two weeks ago. The reason is not that the team didn't listen. The reason is that nothing was written down where the next cycle would see it.

Four moves

The model update is mechanical. Four moves, in order.

1. Close the assumptions that were witnessed

Volume II briefs name assumptions: what we believe but haven't witnessed. After this cycle, some of those assumptions have either been confirmed (witnessed in the field), contradicted (witnessed and wrong), or remain unwitnessed. Close the ones that have moved. Mark them in the brief that owns them.

Assumption (Volume II brief, Initiative #INV-204):
"Graders prefer keyboard shortcuts over deep links."

Status (post-cycle):  CONFIRMED
Evidence:             8 of 8 graders observed using ⌘+Enter to advance.
                      Deep links were used by 1 grader once during the cycle.
Closed:               2026-05-05

2. Add the assumptions you didn't have

The cycle surfaced things the brief did not predict. Some of these are leftover not witnessed items. Some are new — facets of the world the team had not considered. Add them, with their status. The next brief that touches this initiative inherits them.

New assumption (added 2026-05-05):
"Graders work in batches of five exams, then take a 2–3 minute break."

Status:               NOT WITNESSED at decision time, OBSERVED during check
Implication:          Anything that interrupts a batch produces friction
                      out of proportion to its size.

3. Append the signal reading to the brief

Not a copy. A link. The brief now points at the check that was promised before the cycle ran. Anyone who reads the brief next cycle reads the result alongside the prediction.

4. Sharpen the open questions

Every brief carries open questions that need to resolve before the next phase. The cycle answered some of them — strike them. It refined others — rewrite them. It surfaced new ones — add them. The set of open questions in the brief is now the agenda for the next discovery.

What gets updated, beyond the brief

The model lives in more places than one document.

• Templates — if the cycle showed that the brief template was missing a question, add the question. Future briefs inherit.
• Checklists — release gate, DoR, postmortem — gain new items if the cycle surfaced a chain-level gap.
• Domain glossary — new terms learned from the field go into the project's domain glossary so the next story uses the same word.
• Persona notes — Dina's actual day, observed, replaces Dina's described day.

What separates a model update from a wiki entry

A wiki entry is information. A model update is a change to an artifact that the next cycle will use without anyone remembering to look. The test: if the PO who wrote the update were hit by a bus, would the next cycle still inherit the learning?

If yes — model update. If no — wiki entry.

The corpus is built on the assumption that the next cycle will not remember to look. Everything important must be where it is needed, when it is needed.

Why this is the step most teams skip

Three reasons, all real, none acceptable.

1. It feels redundant — we just talked about it in the retro. The talk does not survive. The artifact does.
2. It is unrewarding in the moment — there is no audience, no celebration. The reward is paid out next cycle, by the gap that doesn't open.
3. No one owns it by default — and that is exactly why the corpus assigns it to the PO. Without an owner, it is no one's job, which means it doesn't happen.

The team that does this consistently is the team whose third cycle is meaningfully better than its first. The team that skips it has done six cycles' worth of work, and is still running on its first cycle's model.

Resolution gate — model written down

Enough to know learning survived.

Assumptions in the brief are closed or annotated. New assumptions are recorded. Signal reading is linked from the brief. Open questions are sharpened. At least one template, checklist, or glossary file changed.

Part 7 — The Ongoing Relationship →