Frame the local Ollama router experiment

This lab builds Ollama-backed local router with policy-controlled cloud fallback. The human in the loop is legal analyst summarizing NDAs that must not leave the laptop by default. Scope is intentionally narrower than “make AI reliable”: you will prove one oracle — local llama3.2:3b answers fixture brief at ≥ 25 tok/s; fallback path logs reason=local_timeout — and one invariant — cloud fallback requires explicit policy flag + redaction pass; local p95 ≤ 6s for 512 tokens. Record non-goals in your notes so a later change cannot silently expand authority. The incident mnemonic for the chapter is LOCAL-FALLBACK-PII-03; 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: local_share ≥ 0.8 and cloud_fallback_with_pii == 0. 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 fallback silently ships full NDA text to cloud when Ollama is down. 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 http://127.0.0.1:11434/api/generate.

Capacity note for planners: estimate peak demand on POST http://127.0.0.1:11434/api/generate 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 (local llama3.2:3b answers fixture brief at ≥ 25 tok/s; fallback path logs reason=local_timeout) and cut features that do not serve it. The teaching outcome is judgment under constraints: legal analyst summarizing NDAs that must not leave the laptop by default 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 Ollama-backed local router with policy-controlled cloud fallback, the human stakeholder is legal analyst summarizing NDAs that must not leave the laptop by default, and the incident id you design against is LOCAL-FALLBACK-PII-03. Re-state the oracle in your notes — local llama3.2:3b answers fixture brief at ≥ 25 tok/s; fallback path logs reason=local_timeout — and keep the invariant visible: cloud fallback requires explicit policy flag + redaction pass; local p95 ≤ 6s for 512 tokens. Track local_share ≥ 0.8 and cloud_fallback_with_pii == 0 as the scoreboard. Surface under change control: POST http://127.0.0.1:11434/api/generate.