Certification Coverage Matrix
How the HungerSync world's systems map to every task statement in the Claude Architect cert.
HungerSync Case Series — Certification Coverage Matrix
Section titled “HungerSync Case Series — Certification Coverage Matrix”Purpose. Map the HungerSync world’s systems against every task statement in the two target certifications, so we can lock a case lineup that (a) covers both exams fully, (b) never forces a case to serve a cert it doesn’t naturally exercise, and (c) surfaces gaps before we commit. This is the planning substrate — the episode/ case map is not locked until this is approved.
Sourcing note: task-statement IDs and short paraphrased labels only. No verbatim exam text. Published Case artifacts will use original HungerSync scenarios built on this public skeleton, never the guides’ wording.
1. The two exams at a glance
Section titled “1. The two exams at a glance”Claude Certified Architect – Foundations (pass 720/1000)
| Domain | Weight |
|---|---|
| D1 Agentic Architecture & Orchestration | 27% |
| D2 Tool Design & MCP Integration | 18% |
| D3 Claude Code Configuration & Workflows | 20% |
| D4 Prompt Engineering & Structured Output | 20% |
| D5 Context Management & Reliability | 15% |
AWS Certified Generative AI Developer – Professional (AIP-C01) (pass 750/1000, 65 scored Qs, MC + multi-response)
| Domain | Weight |
|---|---|
| D1 FM Integration, Data Management & Compliance | 31% |
| D2 Implementation & Integration | 26% |
| D3 AI Safety, Security & Governance | 20% |
| D4 Operational Efficiency & Optimization | 12% |
| D5 Testing, Validation & Troubleshooting | 11% |
Interlock, not overlap. Claude is deep on agent loops / MCP / Claude Code and excludes cloud config, vector-DB internals, deployment, streaming, security infra. AWS lives substantially in those excluded areas. The cases exploit this: the heavy joint cases get a Claude reasoning layer and an AWS infrastructure layer; the single-cert cases carry the surface the other exam doesn’t test.
2. HungerSync system inventory (candidate case anchors)
Section titled “2. HungerSync system inventory (candidate case anchors)”| ID | System | Cert center of gravity |
|---|---|---|
| S1 | Passenger Ordering & Resolution Agent | Claude D1/D2/D5 + AWS D2 |
| S2 | Prediction & Dispatch multi-agent pipeline | Claude D1 + AWS D2 |
| S3 | Public-data & edge ingestion (ADS-B, weather, BTS, FIDS, NAS) | AWS D1.3 + resilience |
| S4 | Taste-Profile & Discovery (RAG) | AWS D1.4/1.5 (Claude excludes) |
| S5 | Vendor Enablement & Menu/Availability | Claude D4 + AWS D1.3/D3 |
| S6 | Voucher Rail & Settlement (financial, deterministic) | Claude D1.4/1.5 + AWS D2/D3 |
| S7 | Remote Operations / fleet monitoring (human-on-the-loop) | Claude D5.5 + AWS D4.3 |
| S8 | Platform build & CI/CD (team builds HungerSync) | Claude D3 (AWS thin) |
| S9 | Safety, privacy & guardrails (PII, allergen, injection) | AWS D3 (Claude thin) |
| S10 | Observability, eval & cost (ops center; token governance) | AWS D4/D5 + Claude D4.6/D5.5 |
3. Provisional case lineup (NOT locked)
Section titled “3. Provisional case lineup (NOT locked)”Tag = which cert is primary. “Exhibit” = business-architecture diagram reused.
| Case | Working title | Tag | Claude tasks | AWS tasks | Exhibit |
|---|---|---|---|---|---|
| C1 | The voucher that went to the wrong passenger | Joint | 1.1, 1.4, 1.5 | 2.1, 2.3, 3.2 | revenue-flow |
| C2 | get_flight vs lookup_booking |
Claude | 2.1, 2.3, 4.2 | 2.4, 5.2 | ecosystem |
| C3 | The forecast that forgot Concourse F | Joint | 1.2, 1.3, 1.6, 5.6 | 2.1, 2.5 | value-stream |
| C4 | The feed dies mid-storm | Joint | 2.2, 5.3, 5.6 | 1.2, 2.4, 2.1 | value-stream |
| C5 | The vendor that was already 86’d | Joint | 4.1, 4.2, 4.3, 4.4 | 1.3, 3.1, 5.1 | capability-map |
| C6 | What does this flight crave? (taste RAG) | AWS | 2.4 (resources) | 1.4, 1.5, 4.2, 5.1, 5.2 | journey |
| C7 | Nightly forecast vs the live floor | Joint | 4.5, 1.6 | 2.2, 2.4, 4.1 | value-stream |
| C8 | Rebuilding dispatch with Claude Code | Claude | 3.1–3.6, 2.5, 4.6, 5.4 | 2.3.5, 2.5.4 | (code) |
| C9 | The four-hour delay conversation | Joint | 5.1, 5.2, 1.7 | 1.6.2, 4.1, 5.2 | journey |
| C10 | Gaming free vouchers / allergen safety | AWS | 4.1, 5.2 | 3.1, 3.4 | journey |
| C11 | Who saw whose data? (aggregate-first privacy) | AWS | — | 3.2, 3.3, 2.3 | capability-map |
| C12 | The ops center: spikes, drift, a hallucination | Joint | 4.6, 5.5 | 4.1, 4.2, 4.3, 5.1 | swot |
| C13 | Designing the platform (requirements + WAF) | AWS | (arch. judgment) | 1.1, 2.2, 2.3 | BMC + value-chain |
Distribution: 2 Claude-primary · 4 AWS-primary · 7 Joint. The joint cases are the narrative spine; the single-cert cases carry each exam’s non-overlapping surface.
4. Matrix A — Claude task statements → cases (all 27)
Section titled “4. Matrix A — Claude task statements → cases (all 27)”| Claude task | Covered by |
|---|---|
| 1.1 Agentic loop / stop_reason | C1, C2 |
| 1.2 Coordinator–subagent orchestration | C3 |
| 1.3 Subagent invocation / context passing / Task | C3 |
| 1.4 Enforcement & handoff (prerequisite gates) | C1 |
| 1.5 Hooks (PostToolUse, interception, normalization) | C1 |
| 1.6 Task decomposition (chaining vs dynamic) | C3, C7 |
| 1.7 Session state / resume / fork | C9 |
| 2.1 Tool interfaces & descriptions | C2 |
| 2.2 Structured error responses | C4 |
| 2.3 Tool distribution / tool_choice | C2 |
| 2.4 MCP server integration / resources | C6, C8 |
| 2.5 Built-in tools (Read/Write/Grep/Glob…) | C8 |
| 3.1 CLAUDE.md hierarchy | C8 |
| 3.2 Slash commands & skills | C8 |
| 3.3 Path-specific rules | C8 |
| 3.4 Plan mode vs direct execution | C8 |
| 3.5 Iterative refinement | C8 |
| 3.6 CI/CD integration (-p, json schema) | C8 |
| 4.1 Explicit criteria / false positives | C5, C10 |
| 4.2 Few-shot prompting | C2, C5 |
| 4.3 Structured output (tool_use/JSON) | C5 |
| 4.4 Validation / retry / feedback | C5 |
| 4.5 Batch processing | C7 |
| 4.6 Multi-instance / multi-pass review | C8, C12 |
| 5.1 Conversation context preservation | C9 |
| 5.2 Escalation & ambiguity resolution | C9, C10 |
| 5.3 Error propagation (multi-agent) | C4 |
| 5.4 Context in large codebase exploration | C8 |
| 5.5 Human review & confidence calibration | C12 |
| 5.6 Provenance & uncertainty in synthesis | C3, C4 |
All 27 covered. Concentration risk: Claude D3 (20% of exam) lives almost entirely in C8 — see watchlist item W2.
5. Matrix B — AWS task statements → cases (all tasks)
Section titled “5. Matrix B — AWS task statements → cases (all tasks)”| AWS task | Covered by |
|---|---|
| 1.1 Analyze requirements / design (WAF GenAI Lens) | C13 |
| 1.2 Select & configure FMs (resilience, customization) | C4 (resilience), C13 (selection), C6 (1.2.4 fine-tuned taste model) |
| 1.3 Data validation & processing pipelines | C4, C5, C6 |
| 1.4 Vector store solutions | C6 |
| 1.5 Retrieval mechanisms (chunk/embed/hybrid/rerank) | C6 |
| 1.6 Prompt engineering & governance | C9, C10, C5 |
| 2.1 Agentic AI & tool integrations | C1, C3 |
| 2.2 Model deployment strategies | C7, C13 |
| 2.3 Enterprise integration architectures | C13, C11, C8 (2.3.5) |
| 2.4 FM API integrations (sync/async/stream/resilience) | C2, C4, C7 |
| 2.5 App integration patterns & dev tools (Q Developer) | C8, C3, C12 |
| 3.1 Input/output safety controls | C10, C5 |
| 3.2 Data security & privacy | C11, C9 |
| 3.3 AI governance & compliance | C11, C12 |
| 3.4 Responsible AI | C10, C12 |
| 4.1 Cost optimization | C12, C7 |
| 4.2 Performance optimization | C12, C6 |
| 4.3 Monitoring systems | C12 |
| 5.1 Evaluation systems | C12, C5, C6 |
| 5.2 Troubleshooting | C9, C2, C6, C12 |
All AWS tasks covered. Weakest: 1.2.4 FM customization/fine-tuning deployment — see watchlist W1.
6. Gaps & forced-fit watchlist
Section titled “6. Gaps & forced-fit watchlist”W1 — AWS 1.2.4 (FM fine-tuning / Model Registry / LoRA) has a thin home. HungerSync’s delay/demand models are classic ML, not FMs. Cleanest fix: give the taste classifier a small fine-tuned FM and anchor 1.2.4 in C6. Decision needed: add that to the world, or accept light coverage of a low-weight skill.
W2 — Claude D3 (20%) concentrates in C8. A fifth of the Claude exam in one case
is fragile. Recommend splitting C8 into C8a (CLAUDE.md / rules / skills / plan
mode — config & workflows) and C8b (CI/CD review of settlement code — -p,
JSON schema, multi-pass review, independent-reviewer). C8b also naturally pulls in
the deterministic-finance theme from C1.
W3 — Don’t force the single-cert cases. C6/C10/C11 (AWS-primary) have near-zero Claude content because Claude explicitly excludes RAG internals, security infra, and has no safety domain. C2/C8 (Claude-primary) have thin AWS content (AWS dev tooling is just Q Developer + the CI/CD gateway). Leave the “—” cells empty; that honesty is the point.
W4 — Edge/IoT (your PiAware + weather) has no exam home yet. Both exams are light on edge; AWS 2.3.4 (Outposts/Wavelength) is the only near-fit. Park edge as a world-flavor anchor in C4’s ingestion story, and reserve it as a richer vein for the later GCP/Azure playthroughs rather than forcing it into AIP-C01.
W5 — AWS streaming (2.4.2) is an AWS-only sliver. Claude excludes streaming; AWS tests it. The live-ordering token stream to the passenger UI in C2/C9 is its home — mark it AWS-sidebar-only so we don’t imply Claude-exam relevance.
W6 — Multimodal (AWS 1.3.2) is optional enrichment. Menu photos / a snapshot of an allergy card could justify it in C5/C6. Low weight; treat as opt-in, not required.
7. Decisions needed before locking the map
Section titled “7. Decisions needed before locking the map”- Split C8? (Recommended: yes → C8a + C8b, making 14 cases.)
- W1: add a fine-tuned taste FM to cover AWS 1.2.4, or accept light coverage?
- Case count target — 13–14 full cases is comprehensive; a tighter 9–10 would merge some joint cases. Which granularity do you want?
- Ordering principle for the eventual novel — chronological ground-stop arc, or difficulty ramp? (Doesn’t affect the cases, only the later thread through them.)