Capstone
FDE capstone projects and acceptance
Choose one of four teaching scenarios and complete requirements, blueprints, implementation, acceptance and retrospective. Review learning through five-stage artifacts and decision evidence.
What should an FDE capstone project deliver?
A reviewable capstone explains the customer problem, data and permission boundaries, implementation, actual check results and handover. Choose one of four teaching topics and organize evidence across five stages. Use fictional or synthetic data and start with a rule-based program; real model evaluation and production adoption require separate authorization and evidence.
Teaching example · fictional data
Project example: equipment maintenance tickets
Suppose a dispatcher needs to record faults and confirm assignments. Start with a checkable rule-based program: ticket titles cannot be empty and states follow agreed transitions. This example explains evidence preparation; it is neither a delivered client project nor a completed solution.
- Scope: define dispatcher, technician and approver responsibilities, inputs, data sources and excluded features. Mark unknown conditions explicitly.
- Implement: start with the module workbook and runnable lab. Keep before-and-after output and explain which failure each change addresses.
- Accept: cover a valid ticket, an empty title and an invalid state. Record expected and observed results, code version and unverified scope; do not prefill PASS.
- Hand over: document startup commands, failure observation, an owner and stop conditions. The capstone still requires an AI step, human boundaries and an evaluation set. The foundation path may analyze labeled synthetic records; rule tests are not real model evaluation.
Before starting
Complete preparation checks and six module exercises, keeping case decision notes. Use your own practice project and fictional data; do not connect real payments, customer production or unauthorized data. Experienced learners still provide artifacts and checks for each stage.
Review the curriculum and readiness →Practice topics · choose one
- A
Internal training system: courses, enrollment, instructor counts and administrative permissions.
- B
Pet boarding platform: slots, deposits, merchant confirmation and booking conflicts.
- C
Factory maintenance system: fault tickets, assignment, repair records and role boundaries.
- D
Community group buying: campaigns, orders, inventory, settlement and reconciliation.
Required FDE evidence · check gaps before scoring
These requirements belong within the existing five stages, without extra points. Missing required evidence or unresolved red lines means incomplete/NEEDS_FIX; a total score alone cannot establish completed FDE delivery practice.
Scoping and access (requirements: 20)
Events, three stakeholder groups, source/access/approver/freshness/restriction inventory, scope and pause gates. Identify synthetic data or authorization; fictional clients must be labeled.
AI step and human boundaries (blueprints: 20)
Add classification, summarization or suggestions to a topic. Explain why AI is needed, a deterministic alternative, contracts and human confirmation. CRUD alone is insufficient.
Implementation and reproduction (implementation: 30)
Repository version, environment, four-part instructions, actual outputs and repairs. Build a reproducible software milestone before authorized model use; a rule example is not an LLM.
Model evaluation and red lines (acceptance: 20)
Per-case input, expectation, prohibition, actual output and decision; model/prompt/context/tool/dataset/judge versions and comparable baseline/candidate results. Red-line failures block release. The foundation-analysis path may use labeled synthetic records, without claiming real model evaluation.
Handover and adoption plan (retrospective: 10)
Independent successor reproduction, operation/recovery runbook, escalation owners and adoption users/tasks/window. Without observed adoption, submit a plan with blank results. Label self-rehearsal; independent handover remains unverified.
Distinguish two completion outcomes
Foundation path: synthetic scenarios, actual rule runs, analysis of given evaluation records and self-rehearsal form a foundation practice pack. FDE delivery capstone additionally requires authorized actual model runs, evaluation and independent handover review. Without these, report incomplete rather than inventing output or instructor grades. Neither is an industry credential or client production result.
Download artifact and review templates →Five-stage analytic review · 100 points
Scores describe only the submitted evidence. A missing required gate means INCOMPLETE; any red-line failure means NEEDS_FIX. This site does not grade automatically or issue credentials.
Requirements20 points
Problem, ownership, scope, data permissions, stop conditions and a reasoned disagreement record.
Band Level Anchor 18–20 Strong Scoping evidence is complete and reproducible; decisions stay within evidence. 14–17 Proficient Scoping evidence covers the main requirements with only minor non-decisive gaps. 10–13 Developing Scoping evidence is substantive but still lacks evidence, versions or boundaries. 0–9 Insufficient evidence Scoping evidence cannot be reviewed or claims beyond available evidence. Blueprints20 points
Inputs, outputs, human review, permissions, interfaces, criteria and independently verifiable milestones.
Band Level Anchor 18–20 Strong Blueprints and contracts is complete and reproducible; decisions stay within evidence. 14–17 Proficient Blueprints and contracts covers the main requirements with only minor non-decisive gaps. 10–13 Developing Blueprints and contracts is substantive but still lacks evidence, versions or boundaries. 0–9 Insufficient evidence Blueprints and contracts cannot be reviewed or claims beyond available evidence. Milestones and implementation30 points
Four-part instructions, traceable version points, actual commands and outputs, repairs and critical-behavior tests.
Band Level Anchor 27–30 Strong Implementation evidence is complete and reproducible; decisions stay within evidence. 21–26 Proficient Implementation evidence covers the main requirements with only minor non-decisive gaps. 15–20 Developing Implementation evidence is substantive but still lacks evidence, versions or boundaries. 0–14 Insufficient evidence Implementation evidence cannot be reviewed or claims beyond available evidence. Acceptance and quality20 points
Case provenance, version variables, routine and red-line results, affected regression and release decision.
Band Level Anchor 18–20 Strong Evaluation evidence is complete and reproducible; decisions stay within evidence. 14–17 Proficient Evaluation evidence covers the main requirements with only minor non-decisive gaps. 10–13 Developing Evaluation evidence is substantive but still lacks evidence, versions or boundaries. 0–9 Insufficient evidence Evaluation evidence cannot be reviewed or claims beyond available evidence. Handover and retrospective10 points
Independent reproduction, recovery and escalation, adoption plan, observed-result boundaries and next improvements.
Band Level Anchor 9–10 Strong Handover and retrospective is complete and reproducible; decisions stay within evidence. 7–8 Proficient Handover and retrospective covers the main requirements with only minor non-decisive gaps. 5–6 Developing Handover and retrospective is substantive but still lacks evidence, versions or boundaries. 0–4 Insufficient evidence Handover and retrospective cannot be reviewed or claims beyond available evidence.
Five non-compensable gates
- Customer problem, ownership and data authorization
- AI step, human review, permissions and I/O contract
- Implementation version, commands, outputs and repair trail
- Evaluation versions, per-case results and independent red lines
- Successor reproduction, recovery, escalation and adoption observation
Prepare your practice pack
- Requirements and blueprint files, an implementation repository, execution and verification commands.
- Stage decisions, actual acceptance results, failure handling and unverified scope.
- A retrospective: good decisions, improvements, what to redo and a learning plan.
Keep the pack in your repository or an agreed learning space. If no architecture drift occurred, report that rather than manufacturing failure. Fill evidence gaps without lowering criteria. Training inquiries can discuss learning or team review arrangements; an inquiry is not enrollment or certification.
Training inquiry →Complete practice, then choose your next step
Free workshops include examples, material and self-checks. Paid training is coming soon and is not currently offered. Discuss learning questions; paid enrollment and reservations are not accepted.