# F30 — operator outcome-test grandfather list (seeded 2026-05-21).
#
# Every entry is a CLI subcommand implementation file or an MCP tool
# (anchored under server.py with the synthetic suffix ``/@tool:<name>``)
# that lacks an end-to-end outcome test. The Plan B-parity post-mortem
# identified this gap: 5233 unit + contract + BDD tests passed while the
# production-path LoCoMo benchmark returned 5% — every layer's fakes
# hid composition failures, and no test exercised the subprocess path
# against a real ingested fact.
#
# Baseline shrinks only. Adding an outcome test for any entry below
# REMOVES it from the file in the same commit. New subcommands /
# new MCP tools MUST land with their outcome test — the gate hard-fails
# net-new entries.
#
# Migration plan (target: empty file within 2 weeks of F30 landing):
#   wave 1 (Plan B-parity surfaces): ingest_chat → prep round-trip
#     covers ``ingest-chat`` and the ``prep`` MCP tool simultaneously
#   wave 2 (synthesis surfaces): brief, research, usage_guide, contradict
#   wave 3 (retrieval/operations): search, embed, probe-config, warm
#   wave 4 (long-tail diagnostic + admin): benchmark, soak, eval, setup,
#     onboard, worker, store, summarise, classify, wikilinks, etc.
#
