Define the service boundary and ship target

This lab builds support-answer service comparing retriever_v1 vs reranking retriever_v2. The human in the loop is experiment owner measuring groundedness lift with SRM checks. Scope is intentionally narrower than “make AI reliable”: you will prove one oracle — after 20k sessions, v2 groundedness +2.1pp, p95 latency +80ms within budget — and one invariant — assignment sticky by user_id; analysis gated on sample ratio mismatch (SRM) p>0.001 fail. Record non-goals in your notes so a later change cannot silently expand authority. The incident mnemonic for the chapter is AB-SRM-FAIL-27; design as if that ticket is already written and you are filling evidence.

Turn the oracle into a table: input fixture, expected observable, prohibited side effect, owner, latency/cost ceiling. Separate model taste from software correctness — transport, auth, parsing, and termination must be deterministic even when generated text varies. Primary metric family: groundedness_lift_pp and srm_pvalue. Averages without a denominator or revision label do not gate release. Fake external dependencies in unit tests; live calls wait until fakes pass.

Before implementation, encode a failing check that would have caught reassignment every request → users flicker; metrics uninterpretable. That failure is the pedagogical north star for later pages: contracts reject it, happy path never performs it, validation asserts it, failure-handling contains it, observability detects it, security-ops prevents privilege tricks around it, and mastery replays it in a drill. Endpoint under study: POST /v1/support/answer.

Capacity note for planners: estimate peak demand on POST /v1/support/answer and the cost ceiling for a failed retry storm. Write the abort conditions — unbounded spend, cross-tenant leakage, or inability to roll back — before you enjoy the first green test. Prefer synthetic fixtures shaped like production over anonymized production dumps you cannot share in class. When you are tempted to widen scope, re-read the oracle (after 20k sessions, v2 groundedness +2.1pp, p95 latency +80ms within budget) and cut features that do not serve it. The teaching outcome is judgment under constraints: experiment owner measuring groundedness lift with SRM checks gets a trustworthy control, not a kitchen-sink framework. Keep the language of release decisions: promote, hold, or roll back — never “see if it gets better.”

Decide what will live in version control on day one: fixtures, contract markdown, and a failing test name. Write the cost ceiling as a hard number with currency and period. If the lab involves clusters, name the non-prod context you will use and forbid prod kubecontexts in scripts. Capture the baseline metric once before changing code so later gains are comparative. Refuse tools that hide the request path behind magic macros until the oracle is green on fakes. Your README section for this page should be five lines or fewer and still falsifiable. In this chapter the product is support-answer service comparing retriever_v1 vs reranking retriever_v2, the human stakeholder is experiment owner measuring groundedness lift with SRM checks, and the incident id you design against is AB-SRM-FAIL-27. Re-state the oracle in your notes — after 20k sessions, v2 groundedness +2.1pp, p95 latency +80ms within budget — and keep the invariant visible: assignment sticky by user_id; analysis gated on sample ratio mismatch (SRM) p>0.001 fail. Track groundedness_lift_pp and srm_pvalue as the scoreboard. Surface under change control: POST /v1/support/answer. If you only have forty minutes, finish the fixture for reassignment every request → users flicker; metrics uninterpretable before polishing UI. Promotion language stays ternary: promote, hold, or roll back based on evidence, not hope.