# Pinned census of #[ignore]d tests, one line per test unit (Codeberg #191c).
#
# An #[ignore]d test is run by nothing. It compiles, it is registered, and
# it is never executed: Codeberg #189 found a suite that had been in that
# state for a month while it was broken, and nothing anywhere said so.
# This file makes the size of that bucket a reviewed number instead of a
# side effect. scripts/check-ignored-counts.py compares it against the
# real census and fails `just standard` on any difference.
#
# Format:  <expected count>   <cargo selector for the unit>
# Units not listed here are expected to have ZERO ignored tests, so a new
# test binary is covered without an entry. Lines starting with # and blank
# lines are ignored.
#
# Raising a number here is a one-line diff on purpose. That is the whole
# mechanism: adding a legitimately-ignored test stays cheap, but the
# decision lands in a diff a reviewer can ask about. Say in the commit
# message WHY the test cannot be routed into a tier and WHERE it does get
# run instead.
#
# Baseline measured 2026-08-05 (Codeberg #191): 35 workspace-wide.

3    -p leviculum-core --lib
3    -p leviculum-core --doc

# measure_send_lock_costs is a timing harness for the lock-budget doc page,
# not a behaviour test: it prints microseconds and asserts nothing. Run by
# hand with `--ignored --nocapture` when that page is re-measured.
1    -p leviculum-lxmf --lib

# Batch B2 of #196 landed the build epoch and took the last two staged red
# tests (U5, U17) green, so this drops from 4 to 2. Both remaining stay for
# good — the U16 timing harness, and the case marked CANNOT BE ARRANGED (it
# needs a core-side inbound Resource cancel).
2    -p leviculum-lxmf --test direct_delivery_attempts

# Four of these need hardware or a live service: two RNode cases wanting a
# board on /dev/ttyACM0, two i2p cases wanting a running i2pd SAM bridge.
#
# 4 -> 5 (2026-09-21): append_cost, added by 20261fbb as the evidence behind
# docs/src/concepts/core-lock-budget.md "A durable store append is inside the
# budget too". Like measure_send_lock_costs above it, it prints microseconds
# and asserts nothing, so a tier that ran it would gate nothing; and its
# default directory in an unattended run is the platform tmpfs, where fsync
# never reaches a device and the figure is 60x cheaper than the answer. Run by
# hand with LEVICULUM_APPEND_BENCH_DIR pointed at a real disk, whenever that
# doc section is re-measured or the store's write/fsync/rename/fsync path
# changes. Nothing in the release path re-measures it.
5    -p leviculum-std --lib
1    -p leviculum-std --doc
9    -p leviculum-std --test mvr

# 20 in rnsd_interop, of which the status_parity_tests::* are NOT
# unobserved: `just standard` runs them by name through
# scripts/run-status-parity.sh (Codeberg #191a). They keep the #[ignore]
# because they must run serially, apart from the parallel suite.
#
# 15 -> 20 (2026-08-30): the five rpc_interop_tests::* that drive a client
# out of LEVICULUM_RNS_ROOT — an RNS 1.5.2 checkout that is deliberately not
# vendored, so a fresh clone cannot run them. They run by hand against
# /home/lew/coding/Reticulum in every batch touching the RPC/compat surface,
# and before releases.
#
# 20 -> 21 (2026-09-14): local_client_announce_latency_ab, the A/B the
# removal of LOCAL_CLIENT_ANNOUNCE_DELAY_MS owed. Each arm spawns a Python
# rnsd and the pair costs about 30 s, and what it pins is one stack against
# the other rather than a behaviour of ours alone. Run by hand with
# `--ignored --nocapture` whenever local-client announce scheduling is
# touched on either side; nothing in the release path re-measures it.
#
# 21 -> 22 (2026-09-19): served_program_count_agrees_across_status_clients,
# added by 898723d4 as the acceptance for the Serving-line fix. It is in the
# observed group, not the unobserved one: run-status-parity.sh runs it by
# name in `just standard`, and its EXPECTED moved to 4 in the same commit
# as this line. 898723d4 moved neither pin.
#
# 22 -> 23 (2026-09-23): python_rncp_sender_fails_on_the_same_dropped_identify,
# the reference arm of #131, added by 82425837. It cannot be routed into a
# tier: it measures the REFERENCE rather than guarding our code -- it asserts
# that Python's `rncp` fails the same dropped identify our sender now recovers
# from, so a tier running it would gate us on a limitation of the reference --
# and it needs 150 s of a Python sender that never concludes. It is run instead
# by name, `cargo test -p leviculum-std --test rnsd_interop
# python_rncp_sender_fails_on_the_same_dropped_identify -- --ignored`, whenever
# the identify-loss fix direction is revisited.
23   -p leviculum-std --test rnsd_interop

# measure_propagation_stamp_at_cost_13 is a proof-of-work timing sample, not
# an assertion: its runtime is a coin-flip distribution and only means
# anything in --release on the host it runs on, by hand. Run manually before
# a release and whenever the stamp code changes (see the test's header).
1    -p lnmsg --test stamp_timing
