AI SDLC & Release Train

The release-train anatomy (stages, gates, owners), the four ownership axes, encoding the train as code, and where the train sits in the bRRAIn Build Methodology. A scenario practices ownership calls under pushback; the lab replaces a `make deploy` script with a train.

1. Pretest: train design Diagnostic pretest · 5 min
2. The release-train anatomy — stages, gates, owners Reading · 15 min
3. AI gate vs human gate — the deciding axes Reading · 15 min
4. Encoding the train as code (config + scripts) Worked example · 15 min
5. Where the train sits in the bRRAIn Build Methodology Reading · 15 min
6. Scenario: allocating gate ownership for a new train Scenario · 15 min
7. Lab 1: Replace make deploy with a release train AI role-play lab · 45 min
8. Retrieval: 8 questions across the module Retrieval check · 10 min

Tests, types, schema and audit as deterministic AI-owned gates — and the pass criteria that make them catch what AI-generated code gets wrong. A scenario dissects a green build that should have been red; the lab specifies four gates for a staff engineer who blames the agent.

1. Pretest: gates AI owns Diagnostic pretest · 5 min
2. Tests — what AI can run, what AI cannot Reading · 15 min
3. Types — leveraging the type system as an AI gate Reading · 15 min
4. Schema — migrations + drift detection Worked example · 15 min
5. Audit — the AI-touched commit trail Reading · 15 min
6. Scenario: the green build that should have been red Scenario · 15 min
7. Lab 2: Write the pass criteria for four deterministic gates AI role-play lab · 45 min
8. Retrieval: 8 questions across the module Retrieval check · 10 min

Architecture, customer impact and threat-model gates: why they stay human, what AI prepares for them, and how to design a review surface that produces real decisions in about two minutes. A scenario fixes a gate that became a rubber stamp; the lab designs the customer-impact gate for a breaking API change.

1. Pretest: gates humans own Diagnostic pretest · 5 min
2. Architecture decisions — why AI cannot own these Reading · 15 min
3. Customer impact statements — why AI cannot own these Reading · 15 min
4. Threat-model deltas — when AI assists vs when AI must not Reading · 15 min
5. Designing a human-gate UI that AI prepares Worked example · 15 min
6. Scenario: the human-gate queue that turned into a rubber stamp Scenario · 15 min
7. Lab 3: Design the human gate for a breaking API change AI role-play lab · 45 min
8. Retrieval: 8 questions across the module Retrieval check · 10 min

Build → stage → migrate → swap → smoke → declare; `.prev` rollback and when it is not enough; attribution records; smoke beyond /healthz. A scenario handles a smoke failure on one binary of three; the lab is a deploy-and-rollback game day.

1. Pretest: deploy choreography Diagnostic pretest · 5 min
2. The deploy choreography — build → stage → migrate → swap → smoke → declare Reading · 15 min
3. Rollback — .prev fallback, why it exists Reading · 15 min
4. Attribution — who ran the deploy, what artefacts it produced Reading · 15 min
5. The smoke-test gate — minimum viable post-deploy check Worked example · 15 min
6. Scenario: smoke fails on one binary of three Scenario · 15 min
7. Lab 4: Deploy and rollback game day AI role-play lab · 45 min
8. Retrieval: 8 questions across the module Retrieval check · 10 min

Kill-switch, allowlist and gradual-rollout flags; forward-compatible, reversible, idempotent migrations; the IF NOT EXISTS trap; coordinating flags and migrations across binaries and trains. A scenario sequences a rename against a deadline; the lab plans the flag-and-migration pair.

1. Pretest: flags and migrations Diagnostic pretest · 5 min
2. Feature flag patterns — kill-switch, gradual rollout, allowlist Reading · 15 min
3. Migration rules — forward-compatible, reversible, idempotent Reading · 15 min
4. The IF NOT EXISTS trap — a case study Worked example · 15 min
5. Coordinating flags + migrations across multiple binaries Reading · 15 min
6. Scenario: one release, a rename, a flag and a Friday deadline Scenario · 15 min
7. Lab 5: Ship an unfinished feature safely — flag and migration plan AI role-play lab · 45 min
8. Retrieval: 8 questions across the module Retrieval check · 10 min

What AI uses from metrics, logs and traces; the minimum-viable observability set per binary; page versus email; post-deploy classification and when it may act alone; and the blameless release retro with gate analysis. The lab triages five post-deploy signals.

1. Pretest: observability and retros Diagnostic pretest · 5 min
2. Metrics, logs, traces — what AI uses for self-diagnosis Reading · 15 min
3. The minimum-viable observability set per binary Reading · 15 min
4. Alert design — what should page, what should email Reading · 15 min
5. Post-deploy regression detection — automated vs human-triaged Reading · 15 min
6. The retro format — distinguishing AI vs human cause for each notable event Worked example · 15 min
7. Lab 6: Post-deploy triage — classify five signals AI role-play lab · 45 min
8. Retrieval: 8 questions across the module Retrieval check · 10 min

Identities, least privilege, separation of duties, secrets and tamper-evident audit on AI-driven deploy paths, then a worked case study of how bRRAIn ships its own brain: the release artifact, brrain upgrade (-check, -dry-run, -pin, -rollback), supervised auto-rollback, in-place Upgrade and Update, version agreement and release notes. The lab plans a brain upgrade.

1. Pretest: zero-trust and the bRRAIn case Diagnostic pretest · 5 min
2. Zero-trust on AI-driven deploy paths Reading · 15 min
3. Case study: how bRRAIn ships its own brain — the release artifact and brrain upgrade Worked example · 15 min
4. Case study: in-place upgrades — why the address must not change Worked example · 15 min
5. Version agreement across surfaces, and release notes after every deploy Reading · 15 min
6. Scenario: planning a brain upgrade during quarter-close Scenario · 15 min
7. Lab 7: Plan a bRRAIn brain upgrade AI role-play lab · 45 min
8. Retrieval: 8 questions across the module Retrieval check · 10 min

Running release sessions with the MCP-first bRRAIn session method (git-based open.md / closure.md as the alternative), writing decision records the next train can use, inheriting a train's rules, then the capstone brief, anchor exemplars and the capstone role-play.

1. Pretest: release memory and capstone Diagnostic pretest · 5 min
2. Release memory: running the train with the bRRAIn session method Reading · 15 min
3. Writing release decision records the next train can use Worked example · 15 min
4. Scenario: the new train owner and the rules nobody explained Scenario · 15 min
5. The capstone brief — the release review board Reading · 15 min
6. Anchor exemplars — what 92, 78, 71 and 54 look like Reading · 15 min
7. Retrieval: 8 questions across the module Retrieval check · 10 min
8. Capstone: the Fernhill Freight release review board AI role-play lab · 45 min