# coedit — 106.5s exact, anchored on tests/test_coedit.py (76.4s).
#
# Seconds are per FILE and EXACT, from docs/probes/ci-durations-4core.txt — every
# phase of every test on the 4-core public runner, run 32754826551. The older
# tables in docs/probes/ were taken on the 2-core box a PRIVATE repository gets,
# AND at `--durations-min=0.05`, and both halves of that were wrong for this cut:
# the leg order inverted when the cores doubled, and the 50ms floor hid 37s of
# the suite on exactly the leg that had 1103 tests. Modelling that hidden cost
# was tried and the fit did not hold — see docs/probes/ci-speed.md § 0b. The
# threshold is zero here, so there is nothing left to model.
#
# The cut is what a longest-first pass makes of those numbers, and it lands at
# 106.4-106.5s a leg: a 0.1s spread, against 93.0-119.8s on the cut it replaces.
# `tests/test_editor.py` is 101.6s of the 106.4s it sits in, so this is 5s off
# the floor and there is nothing further to win by moving files.
#
# **Balance is not what to spend the next hour on.** Runs of the SAME lists put a
# leg 44s apart — `views` measured 76.57s and 120.37s across four — so no re-cut
# is visible in one run, and four were not enough to separate this cut from the
# one it replaced (120.4s against 122.8s, in either direction depending on which
# runs you take). What would actually move the gate is the floor, and the floor
# is one file.
#
# Real sockets, a real uvicorn on a real port, and literal `sleep(0.5)` budgets.
# This leg exists so those stay serial, in file order, in one process, on a
# machine nothing else is running AT THE SAME TIME — the condition every one of
# those margins was measured under. Ballast is fine and always was: the other
# files here run before or after it in the same process, never beside it.
#
# **What must never happen is this leg being internally parallelised**, and that
# is also why `pytest-xdist` cannot help the gate even where it helps a leg: the
# critical path is bounded by a leg that is not allowed to split, so making the
# other four faster buys total machine seconds and zero wall clock. Measured, in
# docs/probes/ci-speed.md § 0b.
tests/test_coedit.py
tests/test_hill.py
tests/test_deck.py
tests/test_gitdoor.py
tests/test_cli.py
tests/test_headers.py
tests/test_product.py
tests/test_identity.py
tests/test_socket_offer.py
