Capstone requirements and pick-two

Single course exit: pick-two and commit

The capstone is the integrated proof of ShipAI — not a new idea sprint. Requirements are intentional constraints: demonstrate depth in at least two pillars from the curriculum list, hit measurable success metrics, and ship a demo a stranger can run. Breadth without evals fails the rubric; a narrow vertical slice with honest metrics passes.

Pillars (pick ≥2):

1. Mini-LLM / SLM — fine-tuned or served small model beating baseline on narrow task
2. Agent — tool-using loop with stops, traces, skills/MCP optional but strong
3. Agentic workflow — multi-step graph, harness, durability or failure isolation
4. Industry pattern — gateway, authz MCP, Stripe-style verify, Shopify-style eval loop, etc.

Document choices in capstone/scope_freeze.md — changes after freeze require written rationale.

Pick-two decision guide

Combination	Strength	Risk
Agent + workflow	Production-shaped	Scope creep
SLM + agent	Cost/latency story	Integration time
Workflow + industry	Clear narrative	Less ML depth
SLM + industry gateway	Infra focus	Weaker UX demo

Align with proposal from synthesis module — if pivoting, explain why with metrics impact.

Callout — Pick-two ≠ two half demos: One unified product story where both pillars show in the same user journey beats two disconnected folders.

Success metrics (measurable)

Define 2–4 metrics with targets:

• Task success rate on golden set (e.g. ≥85% on 20 cases)
• p95 latency < 8s for primary flow
• Cost < $0.05 per successful task (estimated)
• Red-team ASR on critical cases = 0% after guardrails
• Verify pass rate on harness path = 100% on demo script

Avoid "works well" — numeric or binary thresholds only.

Demo plan

demo/script.md:

• Audience — hiring manager, PM, engineer (pick primary)
• Duration — 3–5 minutes live
• Steps — exact commands, seed data, expected output
• Failure fallback — recorded video if live API fails
• What you will not show — scope discipline

Rehearse once with timer; cut features until stable.

Scope freeze document

scope_freeze.md sections:

1. One-sentence product pitch
2. Pick-two pillars with file paths proving each
3. Metrics + targets
4. Demo script outline
5. Non-goals (copy from proposal, add any last cuts)
6. Freeze date + signature (your name)

After freeze, new features go to capstone/BACKLOG.md, not v1.

Grader-facing checklist

Mirror rubric in scope_freeze appendix: pick-two evidence paths, eval commands, demo timing, post link placeholder. Graders find artifacts faster — you score better.

Team capstone coordination

If team project, scope freeze assigns pillar owners and interface contracts — same protocol discipline as multi-agent message lesson applied to humans.

Scope change protocol

After freeze, changes require scope_delta.md entry: what changed, metric impact, date — audit trail shows discipline even in solo projects.

Putting it together in practice

ShipAI treats this lesson as executable curriculum, not reading alone. Before marking complete, trace one real request through your portfolio stack and label where this lesson's concepts apply — even if the first pass is messy. Document what broke in the module README; that gap list becomes your next sprint.

Compare your implementation against the industry callouts cited earlier without copying their scale. Name one deliberate simplification you kept (mock auth, SQLite not Postgres, single-region deploy) and one simplification you refuse to ship without (no eval gate, no trace on mutating tools, no fail-closed guardrail on exfil cases). That contrast is what interviewers and graders look for.

Callout — Teach back: Explain this lesson's core tradeoff to a peer in five minutes without slides. If you cannot, re-read the failure modes section and add an example from your own run logs.

Common questions and misconceptions

"Is this overkill for a side project?" Side projects can skip pieces; capstones and production cannot skip knowing the pieces exist. You may waive cost accounting in v1 but your architecture diagram should still show where it would attach.

"Should I rewrite from scratch?" Extend what you built in prior modules — graders reward evolution, not parallel unused folders. Link file paths in your checklist.

"Which metric matters most?" The metric tied to user harm or revenue: policy violations, failed refunds, silent wrong answers — not vanity leaderboard scores.

Extension paths after the micro-project

After the micro-project passes smoke check, choose one extension aligned with your capstone pillar: tighten eval coverage, add a chaos or red-team case, or wire observability into SSE streams. Extensions belong in BACKLOG unless scope freeze explicitly includes them — avoids capstone death by optional polish.

Engineering problem (staff framing)

Requirements define pick-two constraints so the build is demable and evaluable.

Diagram — Requirements → slices

flowchart TD
  Req --> Must --> Slice[Vertical slice]
  Req --> Nice

Precise definitions & mental model

MoSCoW, acceptance tests, demo script.

Tradeoffs — when to use what

Breadth vs depth of slice.

Failure modes (interview + on-call)

Unfalsifiable goals; no acceptance tests.

Production & OSS practices

Requirements frozen mid-capstone except bugfixes.

Micro-project: Scope freeze doc

Deliver finalized capstone/scope_freeze.md plus:

1. Updated demo/script.md with commands.
2. Metrics table copied to capstone README header.
3. Pillar mapping diagram (ASCII OK).
4. Checklist confirming proposal alignment or delta log.
5. BACKLOG.md for cut ideas.

Acceptance: independent reader knows exactly what ships in v1.

Checklist

• ≥2 pillars selected with repo paths
• Metrics have numeric targets
• Demo script with commands and timing
• Non-goals and freeze date recorded
• BACKLOG.md for deferred features