# Conformance items that are RED today, as `driver:item`.
#
# The suite went live against adapters that already had defects. Gating on
# "everything green" would have meant gating on nothing, because a job that is
# red on day one gets switched off within a week. So this file is the debt
# ledger: a failure NOT listed here fails the build, and a line here whose item
# starts PASSING also fails the build, so the ledger cleans itself up instead of
# outliving the defect it documents.
#
# Same shape as platform's scripts/lint_org_scoping_baseline.txt.
#
# Shrink this file. Never add a line to make a build go green — a new red item
# means the change under test broke something, which is the whole point.
#
# Recorded 2026-08-15 against decimalai-python HEAD. Each line names a defect
# already reported in the framework audit; the fix belongs in the adapter, not
# here.

# ── Recorded 2026-08-15, second pass ─────────────────────────────────────────
# The block above was recorded from a run in which only 4 of 11 frameworks had
# their libraries installed — the other 7 were skipped and their failures were
# never seen. That is the same "a partial matrix reads as coverage" mistake the
# CI job's REQUIRE_ALL flag exists to prevent, made while writing the ledger.
#
# These are the rest, from a full matrix with every framework present, verified
# deterministic across 4 consecutive runs and identical with serial children.
# Every one is a defect the acceptance grading already reported.

# ── EMPTY, 2026-08-16 ────────────────────────────────────────────────────────
# The last three lines (crewai:C3, crewai:C4, crewai:C9) came off when the
# defects behind them were fixed. One correction is worth leaving behind,
# because the line that recorded them said something the suite never reported:
# it claimed "a concurrent lane's trace can carry another lane's prompt". It
# could not, and it never did. C9's message on this column was eight copies of
# "<lane>'s trace does not carry its OWN prompt" — the foreign-prompt branch
# never fired — because CrewAI's traces arrived with llm_calls == [] and C9
# grades text taken only from llm_calls. An empty haystack fails the own-prompt
# clause and makes the foreign-prompt clause vacuous. So C9 red here was a
# SYMPTOM of C3/C4, not an isolation defect, and closing C3 closed it.
#
# Nothing is listed now. That is the intended end state, not an omission: a
# failure not named here fails the build.
