Cohort Session 3 Build Run and Loops - akash-coded/aws-bedrock-agentcore-strands GitHub Wiki

Session 3 · Build, run and the loops

How is it built in proven slices, run in production, and kept honest by the eight loops?

Part of the Cohort Kit. Ninety minutes. The session that turns the spec into a plan: by the end the room has a bolt order for the first slice, knows which loops close by themselves and which three nobody is waiting on, and has the evidence pack for the P1 to P2 hand-off written out.

Pre-reading (about 40 minutes)

P2 · Build and prove — bolts, the golden set, and proving each slice before the next
P3 · Run and learn — shadow, cut-over, the two numbers, and what the model changes about operations
The eight loops — which loops close themselves and which need an owner
The evidence pack — the minimum set of artefacts at each hand-off

Bring. The spec from session 2's homework, with the reviewer's mark on the one thing that could not be built as written.

You will leave with. A bolt plan for the first slice, a named owner for each of the three loops that does not close itself, and the P1 to P2 evidence pack as a checklist.

Run-sheet

Minutes	Block	Do this	Material
0–10	Recap	What is a bolt? What closes the trust loop? Which three loops need an owner? What crosses the P1 to P2 hand-off?	—
10–35	The idea	The loop map on screen: ask the room which loops the team already runs, under other names. Then the Day 30 episode, where the first bolt ships by four in the afternoon, and the evidence page, so the room sees what "done" looks like	Loop map · Day 30 · the first bolt · The evidence pack
35–65	Exercise	Pairs do D1 · ordering bolts by dependency on the team's spec, then D4 · a bolt that cannot be built alone. Then B2 · chained probability on the candidate's own step count, so the room feels why short chains matter before anyone proposes a long one	D1, D4 · B2 · Sprint-to-bolt planner
65–80	Decision	The room fixes the first three bolts, in order, each with the check that proves it; names an owner for the drift, cost and learning loops; and ticks the P1 to P2 evidence pack against what actually exists	The minimum artefact set · Templates · engineering
80–90	Close	Homework named. One sentence each: which loop does your team currently not run, and who should own it?	—

Facilitator notes

A bolt is not a small story. The test is "can it be built alone and proven alone". The room will propose bolts that depend on two others; D4 exists to catch that. Draw the dependency arrows on the board.
B2 lands hardest on the architects. Four steps at 90% is 66% end to end. If the candidate's design has six model calls in a row, this is where somebody says so. Let them.
The three loops nobody is waiting on are drift, cost and learning. Every team has an owner for the build loop and none for these. Names on the board, not roles.
Ten-minute version: the loop map with the room naming what it already does, then D1 on the spec.

Homework

The bolt plan for the first slice, in the engineering lead's template, with the proof for each bolt written before the bolt. If the team has a repository, the first bolt is opened as a ticket with the proof in its description.