Week 24 Checkpoint: FDE Simulation I
- Recall Week 24's toolkit: hats, Mom Test rules, requirements sections, proposal bones, demo arc
- Convert a fresh, unseen discovery transcript into a requirements doc under time pressure
- Produce a proposal with an architecture sketch and a demo plan for the same engagement
- Grade the full package against the four-dimension FDE rubric and extract targeted drills
| Spaced-rep: full due deck for Days 162–167 | 10 min |
| Phase 0: closed-book recall of the week's frameworks | 15 min |
| Phase 1: mine the Copperline transcript | 20 min |
| Phase 2: timed deliverable sprint (requirements → proposal → demo plan) | 55 min |
| Phase 3: rubric grading + objection preview | 15 min |
| Quiz + flashcards from today's gaps | 10 min |
Builds on: Day 163 — Discovery & the Mom Test · Day 164 — Requirements — the template · Day 165 — Proposals & architecture docs · Day 167 — Demo craft — the demo plan
Flight schools end each phase with a check ride: the examiner doesn't quiz you on aerodynamics — you fly, with the examiner in the right seat, through a route you have never seen, and the graded thing is the whole performance: planning, execution, judgment, communication. Skills that felt solid in isolation wobble the first time they run in sequence under a clock, and that wobble is precisely what the check ride exists to find while finding it is cheap.
Today is your first FDE check ride. You get a discovery transcript you have never seen — a real-feeling conversation with a freight company, full of pain, politics, red herrings, and numbers — and a timer. From it you produce the full deliverable chain of an engagement's opening week: the requirements doc, the proposal with an architecture sketch, and a demo plan. Then you become the examiner: grading your own package against the same four-dimension rubric (discovery, scoping, communication, feasibility) that FDE interviewers use. The review-day machinery runs underneath as always — the spaced-repetition warm-up and the closed-book recall of the week's frameworks are what make the templates available to you at speed, because in the client room, and in the interview that simulates it, you don't get to look things up.
This simulation IS the FDE interview, rehearsed: the case round at AI labs and applied-AI startups hands you exactly this — a transcript or a role-played customer, then asks for scoping, a proposal, and a plan, graded on precisely today's four dimensions. It is also the honest audit of Week 24: gaps that survive today's rubric would otherwise surface in a real interview or, worse, a real engagement. Day 174 re-runs the cycle harder (objections, mid-course scope change); today establishes your baseline score so that improvement is measurable, not felt.
Guided practice
Phase 0 — closed-book recall of the week
15 minEditor and notes closed. From memory, write:
- The three hats and the engagement lifecycle with each stage's deliverable (Day 162).
- The three Mom Test rules and the five extraction slots (Day 163).
- The eight sections of the requirements doc; the anatomy of a logged assumption; the v0 test (Day 164).
- The proposal skeleton; the three C4 altitudes and their readers; a well-formed estimate (Day 165).
- The demo arc, the live-failure fallback ladder, and the cutting checklist / one-question rule (Days 166–167).
Open your notes; diff; convert every gap into a flashcard BEFORE starting the sim — today's wobbles are Day 174's drills.
Phase 1 — the transcript (your working material)
20 minThis is the simulation transcript. Mine it per Day 163: mark slots, quote evidence, map stakeholders, and — new today — list what the interviewer FAILED to ask (there are at least three significant misses).
Discovery call — Copperline Logistics. Present: Elena Marsh (COO, economic buyer per the AE), Tomas Rieder (Quote Desk Lead), Farid Haddad (Head of IT Security, joined late). FDE: you-know-who. FDE: Elena, thanks for making time. Could you walk me through what happened the last time a quote request came in? ELENA: Tomas lives that, so — Tomas? TOMAS: Yesterday, typical one. A forwarder emails us: 40-foot container, machine parts, Rotterdam to Ho Chi Minh City, what's the rate and what paperwork does the shipper need. The rate part is fast — rate cards are in CargoWise, two minutes. The paperwork question is the time sink. Machine parts means HS-code questions, export-control screening, certificate-of-origin rules that changed in March... I keep a folder of PDFs — customs circulars, our own how-to docs, carrier bulletins. I searched it for maybe twenty-five minutes, then honestly I walked over and asked Ingrid, because she just knows. FDE: How many requests like that per week, and who handles them? TOMAS: The desk gets maybe 120 quote requests a week. Half are rate-only, quick. The other half have at least one compliance or documentation question. Me, Ingrid, and Stefan handle those — and Ingrid is the bottleneck, every hard one funnels to her. FDE: What happens when the answer is wrong? TOMAS: February. A shipment to Türkiye went out quoted with the wrong ATR certificate guidance — customer's goods sat in customs eleven days. We ate about €18K between the demurrage claim and the goodwill credit, and the customer moved half their volume to a competitor. That one was Stefan following a 2023 circular that had been superseded. FDE: Ouch. What did you change after that? TOMAS: Ingrid now reviews anything touching Türkiye, Egypt, or Brazil. Which makes her more of a bottleneck. She's also retiring in March, which nobody wants to talk about. ELENA: Which is why you're here. Our quote turnaround is the number I own — we promise same-business-day and we're at 60%. Every point matters; two big RFPs this year asked for our documented turnaround stats. FDE: Elena, if this worked, what would you look at in six months to call it a success? ELENA: Same-day rate above 85%, and the compliance answers not getting worse when Ingrid goes. I'd take those numbers to the board. FDE: Would an AI assistant that answers the desk's documentation questions from your circulars be valuable? ELENA: If it's right, obviously. If it's wrong we're back in Türkiye. [laughs] FARID: [joining] Sorry — context? ...Right. Before anything touches our documents: we're mid-audit for ISO 27001, our customers are EU shippers, and customer shipment data cannot leave the EU. The circulars folder also has carrier contracts mixed in — those are commercially confidential. Nothing goes to a US model provider without my sign-off, and honestly the answer is probably a European region or nothing. FDE: Understood. Tomas, how current is the circulars folder? TOMAS: [pause] Ingrid updates it when she notices something. The March origin-rule change took us three weeks to notice. FDE: One more — who'd actually type questions into a tool like this? You three, or the whole desk? TOMAS: The desk is eleven people. The eight juniors mostly do rate-only quotes. If the tool were good, juniors could take compliance questions too. That's the actual dream — Ingrid's knowledge not walking out the door in March.
Extraction targets: fill the five slots with quotes; map Elena / Tomas / Ingrid / Stefan / Farid / the eight juniors onto buyer-user-champion-blocker (careful: Ingrid is simultaneously the bottleneck, the quality bar, and a departure risk — what does that make her in your plan?). The interviewer's misses to find (minimum): never asked to SEE the folder (corpus size? formats? how much is contracts-vs-circulars?); never asked what "compliance question" volume looks like end-to-end (is 25 min typical or worst-case? no baseline distribution); never asked about CargoWise integration expectations or where the desk would want answers to appear; never validated the March deadline's hard constraints (what must be true BEFORE Ingrid leaves — capture her review judgments? by when?). Note also the one MT-BAD question and what it produced ("if it's right, obviously" — nothing).
On your own
Objection preview (Day 174 foreshadow)
10 minAfter the sprint, before grading: write the two hardest objections this room would raise to YOUR proposal, and answer each in ≤3 sentences.
Constraint: one must come from Farid (data boundary — your architecture had better already answer it; if your sketch sends carrier contracts to a US-region API, you have failed feasibility before he speaks) and one from Elena (likely: "how does this fix the Ingrid problem by March?" — does your v0 capture her review judgments as labeled data, or does her knowledge still walk out the door?).
Hints: strong answers cite your own artifacts — the residency choice in the container sketch, the review-queue-as-golden-set design where every Ingrid correction becomes eval data.
The graded simulation package
Phase 2, timed — 55 minutes, templates allowed, transcript allowed, this week's finished examples NOT allowed. Produce in docs/sim1/: (1) requirements.md — all eight sections for Copperline; every FR traced to a transcript quote; assumptions log covering at least: corpus size/formats, contract-vs-circular split, CargoWise integration path, question-volume baseline, Ingrid's availability for labeling; v0 must move Elena's same-day metric AND start capturing Ingrid's judgment before March; (2) proposal.md — exec summary, approach with context + container sketches (the trust boundary and EU-residency choice drawn explicitly; freshness pipeline for the circulars — the three-week-stale March incident is your evidence), ranged estimates, ≥5-row risk register (superseded-circular risk gets a row: retrieval must prefer current documents — recency weighting, Day 116), dated ask; (3) demo_plan.md — the one question the demo must retire (credibly: "can it answer a real HS-code/certificate question CORRECTLY, with the current circular cited?"), the magic moment, what is honestly faked, the failure play. Then Phase 3: score all four rubric dimensions 0–3 with one written line of evidence per score, and list your two lowest dimensions with a named drill for each before Day 174.
Common mistakes & misconceptions
- Inventing requirements the transcript never supports ("juniors need a mobile app"). Traceability is the discovery dimension's whole grade; if you can't quote it, log it as an assumption or leave it out.
- Scoping v0 to the fun build (chat over all documents!) instead of the buyer's metric and the March clock. Elena funds same-day-rate and knowledge-capture; everything else is v1.
- Treating Farid as an obstacle instead of a requirement source. The EU-residency and contract-confidentiality lines are hard constraints; an architecture that ignores them scores feasibility zero regardless of elegance.
- Missing the freshness trap. The February incident WAS a stale-document failure; a RAG design without recency weighting and an update pipeline re-ships the €18K bug at scale.
- Missing the Ingrid design. Her retirement is the engagement's real deadline; a v0 whose review queue converts her corrections into labeled eval data solves knowledge-capture as a side effect — examiners specifically look for this move.
- Grading yourself gently. The rubric only produces improvement if the 1s are called 1s; interviewers will not round up, so today should not either.
Q1. The February €18K incident (superseded circular followed) translates into which non-negotiable architecture requirement?
Q2. Farid's statement — EU shippers, ISO audit, contracts mixed into the folder, "nothing to a US provider without sign-off" — should appear in your package as…
Q3. Ingrid retires in March. The strongest v0 response to this is…
Go deeper — curated resources
- Re-run the sim with a partner as examiner — Hand a friend the rubric and your package; their questions will differ from yours in exactly the dimensions you graded too gently. Day 175's debrief is richer with an outside score to reconcile.
- Recall reconstruction done closed-book; gaps became flashcards
- Transcript mined with quoted slots, stakeholder map, and ≥3 interviewer misses found
- All three deliverables produced inside the 55-minute box and committed to docs/sim1/
- Rubric self-score ≥8/12 with evidence lines (below 8: schedule the named drills before Day 174); quiz ≥ 2/3
← Back: Every phase reuses a specific day: mining is Day 163, the requirements sprint is Day 164's template at speed, the proposal is Day 165 compressed, the demo plan is Day 167's one-question discipline — and the architecture judgment underneath is Day 160.
Forward →: Day 174 re-runs the full cycle with written objections and a mid-course scope change; Day 175 debriefs both simulations and rewrites your weakest deliverable; Day 179's FDE case drill and your real interviews are this exercise with strangers watching.
Unlocks: D174 FDE Simulation II — Full Cycle · D175 Week 25 Checkpoint: Debrief & Gap Closure