Agile Coach

In the chosen window (14, 30, 60 or 90 days), it builds a portrait of the team from seven declared sources: the requests' actual flow (lead time, cycle time, throughput, WIP, rework, time per stage), the velocity of closed sprints, the current sprint and its goal, the ceremonies recorded as held, the retrospective notes, the pulse and the backlog's WSJF scores.

Where there is no data, the portrait literally records NOT MEASURED — and the coach is required to say it wasn't measured, instead of filling the gap with a nice sentence.

Doing is what the team does: the ceremonies happen, the board is updated, the retro is held. Being is how the team is: retro actions move forward, WIP is respected, climate is declared, people say what they think.

The classic case is the team that religiously holds every ceremony and changes nothing: high Doing, low Being. The opposite also exists — a healthy team that records nothing and therefore can neither show nor learn.

That's why "we hold the ceremonies, but is it helping?" is this screen's question, and not "are we holding the ceremonies?".

What the coach returns are four blocks: what the numbers say, the Doing × Being reading, the point of greatest leverage (just one thing) and three questions for the next retro, each tied to the number that prompted it.

The rules it is required to follow: (1) every number cited must be in the portrait — no estimating or rounding for effect; (2) where the portrait says NOT MEASURED, say it wasn't measured and what the team would need to record; (3) don't judge people or infer mood or motivation — the only climate data allowed is the score declared in the pulse; (4) if the data isn't enough, say so instead of producing a generic opinion.

The opinion changes nothing in the system. It speaks to the team, not to the board of directors.

Every opinion is stored together with the portrait that originated it, and the "Calculation breakdown — the numbers the coach saw" panel opens right below. This serves two purposes: checking a statement that sounded odd, and comparing today's opinion with the one from two months ago knowing exactly what data each one had in front of it.

If AI isn't active for the company, the screen is still useful: there is no opinion, but the team's portrait in numbers is still there — it's what the opinion would come from.

The coach needs a defined team. Without a room, it would read the whole operation as if it were a single team, and the opinion wouldn't be useful to anyone — that's why the screen refuses to generate it.

Another thing to know: the combo's window (14/30/60/90 days) is independent of the window configured in the room. It always starts at 30 days. A window that's too short for a slow-paced team produces "insufficient data"; one that's too long mixes two different realities.

Two actions: Take to the retro (marks the opinion as accepted, so it becomes an agenda item) or discard. Discarding is a legitimate and frequent answer — an opinion that doesn't fit the team's moment should be discarded, not archived with guilt.

The latest opinions are listed with date, window and model used. Their sequence over the months usually says more than any one of them alone.