# kpr-fix-02 — seeding order was gating dispatch as if it were a dependency,
# so a false linear chain overrode the real graph and exactly one task was
# ever in flight across a full day.
#
# Gated in the PR that makes them pass, per docs/ci/test-gating-policy.md, and in
# their own fragment so they cannot collide with a sibling's append to core.txt
# (tsg-policy-03).
#
# test_dependency_gate_precedence.py pins the rule at the promoter, including the
# half that is invisible in the happy path: with no junction rows the scalar must
# STILL hold a task behind a manual gate, and only a test says so.
#
# test_dep_supersede.py holds the shared rule and, more importantly, the two
# things that would fail SILENTLY if it drifted: the manual-gate carve-out (a
# gate is a HOLD, not seeding order, and releasing one ships work a human
# deliberately held) and the agreement between the SQL and Python renderings of
# the same rule — the dispatcher uses the SQL one and nothing else compares them.
#
# The other three cover the sites a released task would still stall at, which is
# why they join the allowlist now: the HTTP done-guard a worker session uses to
# report its own completion, the draft-ready guard on its PR, and the reports
# that otherwise describe a merge/dispatch policy the board does not have.
tests/kanban/test_dependency_gate_precedence.py
tests/kanban/test_dep_supersede.py
tests/kanban/test_dep_supersede_consumers.py
tests/kanban/test_cli_list_shows_blockers.py
tests/ci/test_pr_watcher_draft_ready.py
