Choose an answer to see why it is right or wrong.
JavaScript is off, so every explanation is shown at once.
01 Where do the decision guides (database, infrastructure, model selection) sit relative to the technology radar and the domain standards, per ADR-0021?
Above the radar and above the domain standards: the guides say which family for which workload, the radar says what is adopted, and the domain standards say how to build with the choice. Correct. ADR-0021's three-layer model is selection (which) above radar (what) above domain standards (how); the guides consume the radar's ring for each option and defer build detail to the domain standards. Below the radar: the radar decides which family for which workload, and the guides just restate the radar's rings in more detail. Incorrect. The guides sit above the radar and explicitly do not restate its rings - they consume each option's ring and add the which-for-which mapping the radar does not provide. Equivalent to the domain standards - naming, reliability, and saas are the same layer as database selection and infrastructure selection. Incorrect. Domain standards are a separate, lower layer that says how to build once a choice is made (e.g. naming's schema rules); selection guides are the which layer above them. 02 A team wants to add full-text search to a Cogitave product. According to database selection's decision tree, what is the correct step-1 answer, and why does it not fall through to the Postgres default?
Cogitave Core's fused query layer - step 1 (access pattern) routes full-text/semantic/graph traversal straight to Core before the relational default is ever reached. Correct. Step 1 of the five-step tree branches on access pattern; full-text/semantic/graph short-circuits to Core's query layer under the Cogitave override, bypassing the Postgres-first default entirely. Postgres with pgvector - Postgres is always the first thing to try before considering any override. Incorrect. For Cogitave products, the override makes Core's query layer the default for full-text/vector/graph; standing up pgvector as a parallel stack alongside Core would itself be a deviation. A standalone Elasticsearch cluster - full-text search always requires a dedicated search engine outside the operational store. Incorrect. A standalone Elasticsearch cluster is explicitly a deviation requiring a measured recall/scale need and an ADR; it is not the default for a Cogitave product. 03 A team is deploying a new agentic flow for a task that has no prior eval evidence and is correctness-critical. What does model selection's default posture say to do?
Start at the most capable Claude tier to establish the correctness bar and a golden output set, then optimize down tier-by-tier only once an eval proves a cheaper tier holds quality. Correct. The default for a new or correctness-critical task is the most capable tier - correctness over speed - and moving down is a deviation gated on eval evidence, never a starting assumption. Start at the fastest, cheapest tier to control cost, and escalate only if it fails outright in production. Incorrect. That is the inverse of the guide's directive; starting cheap for a new, correctness-critical task with no eval evidence is exactly the deviation the guide forbids. Pick whichever tier the team is most familiar with, since model choice is a matter of developer preference. Incorrect. Model selection is a directive guidance system with a stated default and an eval-gated deviation rule - it is not a preference call. 04 A team wants to deviate from a decision guide's default (for example, choosing a VM autoscaling group instead of a container on managed Kubernetes). What must the deviation record, and where?
A measured need - a benchmark or production signal tied to a reliability SLO/error budget, not a hunch - recorded in an ADR with the chosen option and a migration path; the default itself needs no such record. Correct. Every decision guide's deviation rule requires a measured ceiling and an ADR (or, for model selection, eval evidence) before deviating; landing on the default requires nothing recorded. Nothing - any team can choose a different option at any time as long as it still works in production. Incorrect. A decision guide prescribes a default precisely so deviations are not ad hoc; an unrecorded deviation is exactly what the guidance system exists to prevent. A comment in the pull request description explaining the preference, with no ADR required. Incorrect. The deviation rule requires an ADR (recording the workload, the measured ceiling, the chosen option's trade-offs, and the migration path) - an informal PR comment does not meet that bar.