01-projects/phdata

Agent Brigade teaching diagrams — L1–L4 draft set (shared spine + timeline grammar)

2026-07-05·working-draft·status: draft-for-founder-review
agent-brigadeeducationphdataslideware

Agent Brigade teaching diagrams — L1–L4 draft set

Draft rebuild of the founder's 4-level education diagrams (his Miro draft, shared 2026-07-05 morning). Audience: both phData-internal and client-facing. Renders: assets/2026-07-05-brigade-teaching-L1..L4.png, editable source assets/2026-07-05-brigade-teaching-diagrams-source.html (single HTML file, JS-templated SVG — same shape library draws all four levels).

The three design rules applied (why these read as a progression)

  1. Shared spine. Order pinned far left, Stations far right, on every level. Each level inserts new machinery into the same middle gap: Expo (L2) → Ticket + Rail (L3) → Steward + Cellar (L4). L1's long empty arrow is the point — the audience watches the pass get built between request and work.
  2. Same grammar on every level. L4 worked because its timeline narrates the numbered arrows. Now every level has: numbered arrows → a Timeline block → a "Where this level breaks" ribbon that hands off to the next level. The ribbon is the lab hook: the break line is exactly what the learner should experience at the end of that level's lab.
  3. The deciding role gets visual weight. Expo is solid red with "DECIDES · advance / refire / kill" — no longer the same trapezoid as a Station. (Exits deliberately compressed from the canonical five to three for slideware; the full exit set stays in the lab material.)

Paste-ready timelines + break lines

L1 — Agent Skill (one skill, one task)

  1. An Order is submitted straight to a Station — a skill: procedure + judgment + an output contract.
  2. The Station executes its procedure.
  3. The Station responds with the finished artifact.

Breaks: three related Orders arrive — a human routes, sequences, and quality-checks by hand. The human is the orchestrator. → L2

L2 — Agent Orchestration (an expo coordinates the stations — and decides)

  1. An Order is submitted to the Expo at the pass.
  2. The Expo breaks it into Station work and dispatches.
  3. Work returns; critics advise; the Expo decides — advance, refire, or kill.
  4. The Expo responds with the assembled result.

Breaks: Orders arrive while no one is at the pass — every run is hand-fired, and nothing remembers what was built. → L3

L3 — Agent Brigade (tickets on a rail — the backlog works itself)

  1. The Order is written onto a Ticket — scope, context pointers, acceptance criteria.
  2. The Ticket hangs on the Rail.
  3. The brigade pulls Tickets in turn; gates check every Ticket at the door.
  4. Expo dispatches, Stations work, the Expo decides — the Ticket collects the build record as it travels.
  5. The response returns with receipts: the marked-up Ticket is the audit trail.

Breaks: Tickets arrive thin, context is gathered by hand, and finished work has no home a second requester can find. → L4

L4 — Agent Brigade, Operationalized (the full house: self-serve in front, disciplined in back)

Founder's 9-step timeline kept nearly verbatim (it was already right). Additions: FRONT OF HOUSE / BACK OF HOUSE zone labels with the Rail on the seam, and the closing ribbon teases the cross-brigade layer:

Where this goes: requesters multiply, brigades multiply — who pairs an Order to the right brigade? Menus, discovery, the factory: The House (candidate L5).

Level ↔ live artifact map (for the labs)

Level Runnable artifact today
L1 any station skill (variance-analysis + its execution-eval receipts)
L2 fire mode — expo + stations invoked directly, no rail
L3 service start on the rail (walk, gates, ticket markup, close-out)
L4 the 7/3 full hands-off loop (steward agent → walker → independent critic → expo decision)
L5 (candidate) factory artifact:brigade ticket — Website Brigade assembly, menus/discovery

Shape grammar (founder's design, sharpened 2026-07-05 ~9:10am dialogue)

Two narratives, deliberately sequenced — Act 1 = power progression (L1→L4), Act 2 = architecture reveal. The shape grammar carries Act 2 silently through Act 1.

Built same morning (founder go ~9:13am)

Round 3 (founder feedback ~9:30am, built ~9:35am)

Round 4 (founder go ~9:43am, delivered ~9:50am) — three drill-in/finale slides

Feedback round (2026-07-05 ~4pm) — two reactions to the PDF

Founder shared a skeptical Claude critique + his dad's enthusiastic one.

Skeptic = a pass wearing a frown. It correctly extracted the load-bearing engineering ideas as "genuinely good practice" and named them (ticket-as-audit-trail, pointers-not-copies, bounded bounce-back, compounding cellar, common surface). Two knocks:

Dad = nailed the value story. Points 3-6 are the business hooks (existing platforms / common surface / compounding cellar / self-building factory). Steal his Yum Brands analogy for the factory slide: KFC/Taco Bell/Pizza Hut + a team that spins up new restaurant teams = the factory brigade. Better anchor than the current ghost-hexagon.

Decision surfaced to founder: the two reactions = Act 1 (value, deck nails it) + Act 2 (receipts, missing). Next artifact = a short EVIDENCE APPENDIX answering the skeptic point-for-point (Lenovo end-to-end · 11 shakedown defects fixed w/ regression tests · execution-eval lift numbers · Website dogfood). Offered to build; awaiting founder go. Also queued: Yum-Brands factory-slide swap · plain-terms glossary pass.

Open items for founder

Related: [[2026-07-03-brigade-nine-step-receipts-lenovo]] · BRIGADE-INTERFACE.md (ray-plugins) · [[2026-07-03-hex-deployment-matrix]]