Parent: #128
Goal
Add the target-environment operator path for the first real multi-model decision: one Local LLM Server target, 2+ discovered models, one frozen use case/benchmark protocol, Campaign execution, terminal Results and same-case drill-down through the built browser.
Owner
tests/real_runtime/ only, plus focused deterministic tests for harness construction. Do not duplicate campaign/recommendation policy in the harness.
Acceptance
- requires Local LLM Server and
local-llm-identity-v1;
- discovers at least two requested models from one target;
- launches one frozen campaign with one run per candidate;
- waits for a terminal campaign state and retains campaign/run IDs;
- browser executes Find best setup -> Models -> Review -> Campaign Results;
- Results exposes policy-backed recommendation or explicit no-rank outcome;
- at least one same-case comparison is opened;
- screenshots/trace/video are retained;
- owned Performance Lab listener/process is cleaned on every exit path.
Evidence boundary
Hosted CI validates harness behavior only. Final acceptance is REAL_ENVIRONMENT in VALUE-02D.
Integration gate
Development may proceed on an isolated branch, but do not merge to dev until VALUE-01D #120 passes.
Parent: #128
Goal
Add the target-environment operator path for the first real multi-model decision: one Local LLM Server target, 2+ discovered models, one frozen use case/benchmark protocol, Campaign execution, terminal Results and same-case drill-down through the built browser.
Owner
tests/real_runtime/only, plus focused deterministic tests for harness construction. Do not duplicate campaign/recommendation policy in the harness.Acceptance
local-llm-identity-v1;Evidence boundary
Hosted CI validates harness behavior only. Final acceptance is REAL_ENVIRONMENT in VALUE-02D.
Integration gate
Development may proceed on an isolated branch, but do not merge to
devuntil VALUE-01D #120 passes.