session · operate & reflect
Signal reading session
60 minutes on the calendar-committed check date — the date that lived in the Feature Brief from the start. Witness the change. Write the five-line reading. Pick one of four outcomes. The session that turns a prediction into a check.
When
- On the date in the brief — not the day before, not the week after. The discipline is calendar-committed.
- At least 30 days after release for most predictions — enough time for behaviour to stabilise out of novelty.
Who
- PO — facilitates; writes the reading.
- Tech Lead — confirms the metric is being measured correctly.
- CS Lead — adds the voice-of-customer paragraph within 24 hours.
Time-box
60 minutes. Enough to witness, measure, and write. Anything longer drifts into interpretation; the Retrospective does that.
Inputs
- The Feature Brief's prediction (verbatim).
- The baseline number (captured pre-release).
- The leading and lagging signal dashboards.
- The helpdesk reading from the past 4 weeks.
- Access to the named person (for direct observation if the prediction is about a moment).
Agenda
| Time | What |
|---|---|
| 0–10 min | Read the prediction aloud verbatim from the brief. The room hears what was promised, exactly. |
| 10–30 min | Witness the change. If the prediction is about a moment ("Gal grades in under 15 minutes"), watch Gal. If it's about a metric ("completion rate above 80%"), read the dashboard with the team. Observation over report. |
| 30–45 min | Write the reading. Five lines: prediction · baseline · target · measured · gap. Use the Signal reading template. |
| 45–55 min | Pick the outcome. Met · Missed honestly · Refuted · Not checked. One of four. "Kinda met" is hedging — pick. |
| 55–60 min | Sign and file. PO signs; reading lives next to the brief; calendar entry for the next session (Retrospective). |
Outputs
- The signal reading (template) — filed next to the Feature Brief.
- The outcome — recorded, public.
- (Within 24h) CS Lead's voice-of-customer paragraph.
- The reading as input to the retrospective scheduled within the week.
What good looks like
The reading is five lines and no flourish. "Predicted 15 minutes. Baseline 47 (n=12). Target <15. Measured 11m20s (n=8). Gap: better than predicted; the keyboard shortcut absorbed more time than the deep-link navigation we built for." That is a reading. "We've seen improvement" is not.
The team treats the reading as model correction, not validation. A Met outcome is documented; a Missed honestly outcome is celebrated as the most valuable kind of learning; a Refuted outcome triggers a serious portfolio conversation; a Not checked outcome is the only worthless outcome.
Anti-pattern
The check drifts off the calendar, the prediction "wasn't quite ready to check yet." Weeks pass; the chance to honestly read fades; outcome ends up as Not checked. Fix: the date in the brief is the commitment. If the metric isn't ready, the reading says so — that is the finding, not an excuse.
A second anti-pattern: reading by report, not by observation. The PO opens the dashboard, sees the number, writes the reading from the desk. The number says success; the lived experience disagrees. Fix: for any prediction about a moment, witness the moment. The number tells you how much; only observation tells you what is happening when the number says so.
A third: the outcome hedges. "Kinda met but with some caveats." The PO picks one of four. The caveats go in the gap line. The outcome line is the verdict.
See also
- Canon — After We Build · Signal and the Prediction
- Template — Signal reading
- Practice — Signal reading
- Principle — Not checked is the only worthless outcome
- Next session — Retrospective