Delete the test tiers: a run is the whole suite or its filter

.claude/agents/orchestrator.md

@@ -71,6 +71,7 @@ A mutation loop, guest or metal, starts on a clean worktree at the head under re
 clean: `git apply --check`, `git apply`, the tests by name, `git apply -R`. None runs while an
 agent edits that worktree. A queue script passes only flags `src/testargs.rs` declares: any other
 word becomes the run's filter, and a one-test run reports as a pass.
+A run is the whole suite, or a positional filter that reaches any enabled test.
 
 ## The bench
 

.github/workflows/nightly.yml

@@ -1,15 +1,13 @@
 name: nightly
 
 # Everything that boots a guest, the host gate again to write the cache the
-# merge queue restores, and portability: on main every night, the weekly tier
-# on Sundays, and on any branch by `gh workflow run nightly.yml --ref <branch>`.
+# merge queue restores, and portability.
 # Every job's logic is `cargo run -- --ci <job>` (src/ci.rs); this file says
 # where each one runs.
 
 on:
   schedule:
-    - cron: '0 3 * * 1-6'
-    - cron: '0 3 * * 0'
+    - cron: '0 3 * * *'
   workflow_dispatch:
 
 # Never cancelled: `build` may be an hour into a bootstrap.

CLAUDE.md

@@ -69,7 +69,7 @@ The bar is not yet the tree. The standing failures are declared rather than remo
 
 ## Build & test
 
-The testing rules live where they are enforced: known reds in `src/redlist.rs`, tiers in `src/tiers.rs`, the PR gate and the nightly in `.github/workflows/`. Operationally:
+The testing rules live where they are enforced: known reds in `src/redlist.rs`, the PR gate and the nightly in `.github/workflows/`. Operationally:
 
 - `cargo run` builds everything (toolchain, kernel, bootloader, userland, image) and launches QEMU; `--build-only` skips the launch. `cargo test` runs the QEMU harness; `cargo run -- --ci host` runs every host suite, as the PR gate's required `host` check does.
 - **Agents never run QEMU.** An agent verifies with host tests and builds the image at most; the orchestrator runs every guest test, one suite at a time.

issues/boot-media/a-disk-whose-port-went-away-panics-the-boot-at-roots-hold.md

@@ -6,7 +6,7 @@ opened: 2026-09-27
 
 # A disk whose port went away panics the boot at ROOT's hold
 
-`usb_transport_break` (nightly tier) fails its boot with:
+`usb_transport_break` fails its boot with:
 
     PANIC: panicked at src/rootfs.rs:177:19:
     boot: the partition ROOT was read from, <guid>, cannot be held: Unusable

issues/boot-media/kernel-log-file-reds-beside-other-guests-and-is-green-alone.md

@@ -40,9 +40,7 @@ session against an unchanged tree, not a re-classification of its `Sched`.
 
 Not the diff it was seen from: that branch's kernel change to this path is the
 deletion of an unreachable `if rflags & TF != 0` branch in
-`arch::LogCommitGuard::close`, which removes instructions and adds none, and
-`kernel_log_file` is Nightly for `Why::Cost` rather than for flakiness
-(`src/tiers.rs`).
+`arch::LogCommitGuard::close`, which removes instructions and adds none.
 
 ## Promoted 2026-08-25
 

issues/boot-media/screen-diag-boot-leaves-no-i8042-line-on-the-panel.md

@@ -6,7 +6,7 @@ opened: 2026-09-27
 
 # `screen_diag_boot` finds no `i8042:` line on the panel
 
-`screen_diag_boot` (nightly tier) fails with `"i8042:" is not on screen five
+`screen_diag_boot` fails with `"i8042:" is not on screen five
 seconds after the boot finished`. The decoded panel ends at `Boot: complete`,
 and no i8042 line is painted in the five seconds after it. The i8042 health
 line (`report_health`, `to_screen`) never reaches the panel on the diag image.

issues/build/a-metal-only-row-cannot-be-disabled.md

@@ -7,7 +7,7 @@ opened: 2026-09-29
 # A metal-only row cannot be disabled, so its red has no legal home
 
 `tests/toyos.rs`'s `check_redlist` accepts a `src/redlist.rs` row only for a
-name `schedule` holds: the shared boot's discovered binaries, `SCREEN_TESTS`
+name `registered` holds: the shared boot's discovered binaries, `SCREEN_TESTS`
 and `MACHINE_TESTS`. A `METAL` row listed in `METAL_ONLY` has no QEMU
 registration and is in none of them, yet the metal branch filters `METAL`
 rows through the same `keep` that reads the redlist. A metal-only red can
@@ -17,13 +17,13 @@ row before any boot.
 ## Measured
 
 `hda_tone` and `hda_client_stall` red on the T14 run of `main` at `7e151819`.
-Only the code reading is verified: `check_redlist` tests `schedule.contains`, and
+Only the code reading is verified: `check_redlist` tests `registered.contains`, and
 the metal branch filters `METAL` rows through the same `keep`. Running
 `--known-red hda_tone` and `--list` against a tree with such a row is owed and
 not in this tree's record.
 
 ## Exit condition
 
-`check_redlist` accepts a name `METAL` registers as well as one `schedule`
+`check_redlist` accepts a name `METAL` registers as well as one `registered`
 does, and `cargo run -- --known-red hda_tone` can answer YES from a row the
 suite accepts.

issues/build/a-settled-thread-join-taking-the-table-lock-again-is-gated-by-nothing.md

@@ -15,11 +15,11 @@ Whether `collect` is where the kernel takes `PROCESS_TABLE` is
 A settled join then takes the lock on every wake of its wait.
 
 **Exit**: taking `PROCESS_TABLE` outside `ask_join`'s `collect` reds a host
-test or a fast-tier test. Owner: orchestrator.
+test. Owner: orchestrator.
 
 The syscall's answer-keeping is the same gap: `sys_thread_join`
 (`kernel/src/syscall/proc.rs:154`) holds one `Join` across every ask, but
 `a_join_asked_after_it_collected_keeps_its_answer` only replicates that shape
 rather than calling the syscall, so a `Join::ask` built fresh per ask
-(`join-per-ask`) stays host-green and reds only on the nightly guest
+(`join-per-ask`) stays host-green and reds only on the guest
 `fpu_isolation` (`thread_join failed`, `left: 18446744073709551614`).

issues/build/a-shards-boot-width-does-not-price-its-tests.md

@@ -11,8 +11,7 @@ reference and multiplies every liveness ceiling by the result; every shard print
 it (`host: fastest boot N ms against the reference 1320 ms — liveness ceilings
 paid at Wx width`). The proposal was to spend the same factor on the *duration
 profile*: divide each shard's measured prices by its width in
-`35383398^:src/durations.rs`'s merge, so `src/tiers.rs`'s ceiling compares like with like
-across shards of different speed, with timer-anchored names exempt because a
+`35383398^:src/durations.rs`'s merge, with timer-anchored names exempt because a
 fixed wait does not shrink on a fast host.
 
 **Measured over six hosted twelve-shard runs and refused.** Dividing by the
@@ -60,7 +59,7 @@ reverse regression is 1.248 with `var(ln width)` 7.86x `var(factor)`: even read
 charitably, ln(width) is the shared cause buried under eight times its own
 variance of noise, and dividing by it adds that noise to every price.
 
-**Within-name slope of ln(price) on ln(width), by `src/tiers.rs` relegation
+**Within-name slope of ln(price) on ln(width), by
 class** — the prediction was ~1 for ordinary and `Why::Cost` names and ~0 for
 `Why::TimerAnchored`:
 

issues/build/idle-stack-guard-price-nearly-doubled-since-its-return.md

@@ -12,6 +12,4 @@ p10-p90 shard spread that explains ordinary variance. `dump_nmi_probe` grew
 +29% over the same window (6,284 -> 8,098 ms). Either the boots these ride got
 genuinely slower — a real kernel-cost regression nobody measured on purpose —
 or the shard fleet's pricing shifted; the two hypotheses separate on a bisect
-of the boot's own timestamps across the window's landings. Noticed while
-relegating the straddler batch; the relegation hides the symptom from the
-per-PR gate, which is exactly why the growth is recorded here.
+of the boot's own timestamps across the window's landings.

issues/build/lan-talk-host-closes-judges-a-fixed-ten-second-window.md

@@ -22,4 +22,4 @@ the listener's first accept, bounded by a ceiling that fails loudly, and only
 then pace the offers into the closed connection.
 
 **Exit condition**: the test waits for the accept event instead of for a fixed
-span, and stays green under a loaded fast tier.
+span, and stays green under a loaded suite.

issues/build/log-reserve-window-negative-times-out-beside-other-guests.md

@@ -24,4 +24,4 @@ hypothesis, and one red at 18x its price is not a rate.
 Exit: a rate — the same suite run repeatedly with and without a second
 worktree's build on the host — that says whether this is contention the harness
 should schedule around or a defect in the guest's own boot, and the name is
-either re-tiered or fixed at the cause.
+fixed at the cause.

issues/build/parallel-tests-red-under-other-suites.md

@@ -64,9 +64,7 @@ changes.
   `qemu::budget(20 s)`, the phase's. Still `Sched::Parallel`. **The duration
   profile's share of this is closed**: nothing is retyped against a clock any
   more — `/system/bin/terminal` prints `terminal: ready` and `shell_echoes` waits on
-  that (`tests/toyos.rs`'s `SURFACE_UP`) — and the four-minute lane holder that
-  the profile used to seat a second desktop beside, `desktop_window_child`, is
-  `Tier::Nightly` and so never in a pull request's parallel phase.
+  that (`tests/toyos.rs`'s `SURFACE_UP`).
 - **`desktop_locale_detect`** — retired 2026-09-04, green 5 of 5 beside a full
   fast tier.
 - **`netd_connection_caps`** — retired 2026-09-04, green 5 of 5 beside a full
@@ -500,6 +498,6 @@ mechanism for it.
   driver nor this test, and the test was green in its previous fast-tier run.
 
   Owner: the i8042 tally. **Exit condition**: the verdict revising itself in a
-  parallel run — a loaded full fast tier in which `i8042_undecoded_bytes`'
+  parallel run — a loaded full suite in which `i8042_undecoded_bytes`'
   first mute line names nothing and its second names the sequence, or the
   retirement's clause narrowed to the conditions under which it holds.

issues/build/process-stats-exits-101-beside-other-guests.md

@@ -21,5 +21,5 @@ harness naming a hypothesis, and one red is not a rate.
 
 Exit: a rate — the same suite run repeatedly with and without a second
 worktree's build on the host — that says whether this is contention the harness
-should schedule around or a defect the guest has, and the name is either
-re-tiered or fixed at the cause.
+should schedule around or a defect the guest has, and the name is
+fixed at the cause.

issues/build/qemu-drops-console-output-the-harness-is-slow-to-read.md

@@ -51,4 +51,4 @@ stdio console.
 The console chardev the harness reads cannot refuse a write — for example a
 file chardev the harness follows, or a `virtserialport`, which QEMU throttles
 instead of dropping. Shown by `log_stream_stalled_reader` green in 20 of 20
-runs beside a full nightly.
+runs beside a full suite.

issues/build/smp-ap-hole-and-log-reserve-window-red-under-a-loaded-host.md

@@ -9,9 +9,7 @@ opened: 2026-09-16
 Both `Sched::Parallel`, both first sightings, both in one `cargo test` fast-tier
 run on `wt/toyos-usbhang` sharing the host's twelve guest slots throughout with
 two other worktrees' suites (`[host-slots]` names `pid 49668` and `pid 2726`
-holding slots across the whole run). The branch's own diff at the time —
-`tests/toyos.rs`'s tier table, `src/tiers.rs`'s `RELEGATED` and
-`tests/test-durations`, all for `usb_reset_records_the_phase_it_cut` — touches
+holding slots across the whole run). The branch's own diff at the time touches
 neither SMP bring-up, `spawn_init`, nor `rootfs.rs`, so it is filed rather than
 chased. Same shape as
 `issues/build/parallel-tests-red-under-other-suites.md`, filed separately

issues/build/the-console-loses-typed-keystrokes-under-host-load.md

@@ -17,6 +17,6 @@ of 6 green off the one red.
 No layer (QEMU input, i8042/USB, translator, console) has been measured yet.
 
 **Exit condition.** The layer that drops the characters is identified and
-fixed, and the test then passes 10 of 10 consecutive runs of the Fast tier
+fixed, and the test then passes 10 of 10 consecutive runs
 under the host load of the original failure (other suites running in
 parallel); then the redlist row goes.

issues/build/the-shard-split-prices-a-boot-and-not-the-image-behind-it.md

@@ -7,9 +7,9 @@ opened: 2026-08-15
 # The shard split prices a task's boot and not the image it builds first
 
 `Shard::keep` partitions the twelve shards by `tests/test-durations`, and that
-profile records what a *test* took — the number the ten-second Fast ceiling is
-also read against. A machine test that boots a config the shared image does not
-cover builds that image first, and nothing prices that build at all.
+profile records what a *test* took. A machine test that boots a config the
+shared image does not cover builds that image first, and nothing prices that
+build at all.
 
 **Measured on run `31896922288`** (`main` at `e064a96`, twelve KVM shards, the
 Fast tier), reading each shard's job log for the wall clock between one `PASS`
@@ -68,11 +68,8 @@ where it landed.
 
 Two directions, and they are not exclusive:
 
-- **Price a task by build + boot.** The profile cannot simply absorb it: the
-  same file is what the `durations` verdict reads against the 10,000 ms Fast
-  ceiling, so a task priced at 203 s would red the tier gate it has nothing to
-  do with. It wants a second profile — per *config*, not per test — that only
-  `Shard::keep` reads.
+- **Price a task by build + boot.** It wants a second profile — per *config*,
+  not per test — that only `Shard::keep` reads.
 - **Make the second image cheap.** 145 s to add `netd` and `sshd` to an image
   is a full recompile of those programs, not a relink.
 
@@ -93,9 +90,7 @@ the number this entry says nothing prices is measured on every run, deliberately
 removed from the price, and then dropped on the floor. What is missing is two
 smaller things: nothing attributes that duration to the *config* that caused it
 rather than to the worker thread that happened to ask first, and nothing writes
-it anywhere. `tests/test-durations` cannot be where it goes — the same file is
-read against `FAST_CEILING_MS` (`src/tiers.rs:84`, 10,000 ms), which is this
-entry's own reason for wanting a second profile.
+it anywhere.
 
 **What no work on this host can supply is the seed and the verdict.** The cost
 being priced is a *cold-checkout* rebuild on a hosted shard: `actions/checkout`

issues/build/the-tooling-is-a-review-prompt-and-three-workflows.md

@@ -10,7 +10,7 @@ The rules a prompt can read off a branch move into `.claude/agents/reviewer.md`,
 and the gates that held them go.
 
 - A test is green and fast or it is deleted in the same pull request and filed;
-  then `src/redlist.rs`, `src/tiers.rs`, `35383398^:src/durations.rs` and
+  then `src/redlist.rs`, `35383398^:src/durations.rs` and
   `tests/test-durations` have no subject and go.
 - The toolchain is content-addressed by the four trees that produce it, one
   directory per hash, never mutated; then the sysroot claim, `src/buildlock.rs`

issues/build/usb-transport-break-flushedstick-can-break-after-the-reboot.md

@@ -27,5 +27,5 @@ re-run. The logd branch, its tree as committed at `ebf2f76a`, measured the same
 
 The case's premise is arranged rather than raced: a write after a verified
 flush is made to go out before the job ends the boot (or the case waits on the
-break before it starts the job), and ten consecutive `--nightly
-usb_transport_break` runs are green.
+break before it starts the job), and ten consecutive
+`usb_transport_break` runs are green.

issues/build/which-pass-drain-irqs-is-entered-from-is-gated-only-by-a-nightly-guest.md

@@ -14,5 +14,5 @@ compiles. PR #564's round-2 review handed `pass_block`'s drain
 `Entered::Pass { depth: 0 }`: every host test and the fast tier stayed green,
 and only the nightly `blocked_dump` went red.
 
-**Exit**: a mutation of either call site's `Entered` reds a host test or a
-fast-tier test. Owner: orchestrator.
+**Exit**: a mutation of either call site's `Entered` reds a host test.
+Owner: orchestrator.

issues/filesystem/blockd-survives-its-death-reds-on-a-replacement-that-reorders-acknowledged-writes.md

@@ -6,7 +6,7 @@ opened: 2026-09-27
 
 # `blockd_survives_its_death` reds on a replacement that rewrites acknowledged writes out of order
 
-The nightly-tier test kills blockd with writes in flight and judges QEMU's
+The test kills blockd with writes in flight and judges QEMU's
 block trace. It has gone red on one signature, the same sectors each time:
 
 ```

issues/filesystem/home-budget-refusal-retried-is-red-on-every-nightly.md

@@ -6,7 +6,7 @@ opened: 2026-09-27
 
 # `home_budget_refusal_retried` is red on every nightly
 
-`home_budget_refusal_retried` (nightly tier) is red, and red alone, on the
+`home_budget_refusal_retried` is red, and red alone, on the
 nightlies of three trees, each time with the same two shapes:
 - main at e8d7c9c0 (run 36228604597);
 - PR #524's branch at 8c5be843 (run 36273557690);

issues/hardware/tearing-is-what-gop-cannot-give-back.md

@@ -63,7 +63,7 @@ pixels — which is the "reference capture" half of a pixel-identity check, with
 no new mechanism at all.
 
 **A per-client damage oracle exists and is green.** `desktop_typing_damage`
-(`tests/toyos.rs:6670`, registered at `:823`, Nightly) reads the compositor's
+(`tests/toyos.rs:6670`, registered at `:823`) reads the compositor's
 own `damage_px_max` while a client is driven and refuses a frame over 2% of the
 screen. Measured on the dev host 2026-08-31: `eight lines typed, 16
 appearances; biggest frame 9472 of 2073600 px over 2 intervals`, PASS. It bounds

issues/isolation/the-supervisor-is-host-tested-and-owns-the-stop.md

@@ -10,9 +10,8 @@ Held by the orchestrator. Every stage waits on PR #536 (`wt/toyos-fsd`).
 
 ## Stages
 
-Every guest test this track names is registered at `Tier::Fast` or
-`Tier::Nightly` with no `src/redlist.rs` row. A deleted or disabled test covers
-nothing.
+Every guest test this track names is registered with no `src/redlist.rs` row.
+A deleted or disabled test covers nothing.
 
 1. **The rename**, one mechanical PR, first after #536. Before stage 1 is
    briefed, the exit's search below runs once over the `rust/` fork's delta as

issues/kernel/a-claims-own-refusals-are-read-by-nothing.md

@@ -11,7 +11,7 @@ The refusals `kernel/src/pcidev/mod.rs` raises that are read back are
 `iommu_virtio_platform`'s no-unit arm, `Refusal::NoInterrupt` by
 `virtio_net_no_msix`, `Refusal::CapsTruncated` by `pci_claim_caps_truncated`,
 the domain by `userdev_dma_fault`, and `SYS_DEVICE_REG_READ`'s bound by netd's
-own `config_space_is_bounded`. These are reached by no test in any tier:
+own `config_space_is_bounded`. These are reached by no test:
 
 - `ClaimError::Ambiguous`, a config naming a device this machine has two of;
 - `ClaimError::KernelDriven`, a claim on a function one of this kernel's own
@@ -35,8 +35,8 @@ own `config_space_is_bounded`. These are reached by no test in any tier:
   no BAR at all, so what is unread is the hand-over that succeeds.
 
 A one-field mutation of any of them — an index bound compared `<=` rather than
-`<`, a window overlap accepted, a grant total never summed — leaves every arm in
-every tier green.
+`<`, a window overlap accepted, a grant total never summed — leaves every arm
+green.
 
 Owned by whoever next adds a boot config whose own test binary holds a claimable
 function: netd holds this machine's only claim, and a test that makes it fail

issues/kernel/a-log-rings-owner-is-named-only-when-logd-reads-its-registration.md

@@ -14,7 +14,7 @@ Until `logd` reads the frame the owner word is 0, and `push` then keeps
 nothing for anyone. A child that starts flooding in that window can take
 every slot, the owner's included.
 
-`log_ring_keeps_the_owners_slots` (fast tier) is red this way beside the
+`log_ring_keeps_the_owners_slots` is red this way beside the
 other `log_` guests and green alone. Its `/log` holds `===READY===`,
 `===TEST_START test_rs_log_flood===` and 1917 flood lines: exactly the ring's
 1919 shared slots, with no slot left for test-runner's `===TEST_END`.

issues/kernel/desktop-window-child-freeze.md

@@ -134,10 +134,7 @@ machine that is still running passes somewhere. Judge the next occurrence by the
 signature exactly as before; a green run of this test proves what it always
 proved, which is nothing.
 
-Two things a reader looking for the next sighting needs. The test is
-`Tier::Nightly` (`src/tiers.rs`), so a plain `cargo test` does not run it at all
-— `cargo test --test toyos-build -- --nightly desktop_window_child` does. And
-the one instrument that could name what a stopped CPU is doing has still never
+The one instrument that could name what a stopped CPU is doing has still never
 been fired at one: `sched::dump`'s NMI probe separates a CPU spinning with `IF`
 clear from one halted with its kick undelivered from one wedged below the
 interrupt layer. Take `info registers -a` over QMP before pressing Ctrl+Alt+D,

issues/kernel/no-kernel-address-is-drawn-per-boot.md

@@ -14,7 +14,7 @@ with `RANDOMIZE_BASE` and `RANDOMIZE_MEMORY`
 (`debian.master/config/annotations:10674,10677`): at most 9 bits of text slot,
 and a direct map placed at PUD granularity.
 
-**Exit**: on every proving machine, over 64 boots in a `Tier::Weekly` test,
+**Exit**: on every proving machine, over 64 boots in a test,
 each claimed bit is set 12 to 52 times, the guard of
 `issues/kernel/the-kernel-has-no-stack-protector.md` and the first spawn's
 bases included. **Mutation**: a fixed base. **Oracle**: Linux's bits.