chore(quest): plan checks at moq's cost, from the gate and local check audit

quest/m0/README.md

@@ -3,3 +3,7 @@
 ## Goal
 
 Urgent fixes to anything a consumer already depends on.
+
+## Required
+
+- [Checks at moq's cost](/quest/m0/checks/README.md) - `just ci` selects by the diff, builds warm, and runs the differential optimized, with a nightly full `just gate`

quest/m0/checks/README.md

@@ -0,0 +1,26 @@
+# Checks at moq's cost
+
+## Goal
+
+The land gate and agents' local checks cost what moq's do: `just ci` runs only what a
+change touches, builds warm, and runs the differential fixtures optimized, while a nightly
+`just gate` runs the full suite. Measured against the 2026-11-01 audit's baseline: about
+33 CPU-hours a day in the gate (65 CPU-minutes a run) and 100 to 130 locally.
+
+## Plan
+
+Decided with the maintainer on 2026-11-01, from an audit of the gate's and agents' checks
+(nixos2 and the host's systemd accounting): copy how moq works. `just ci` before every push
+stays the rule, and costs less once it selects by the diff; CONTRIBUTING.md and the skills
+keep their rules. All of it is m0, since every agent pays the cost now.
+
+Each child reports its CPU saved, cold and warm, against `origin/main`.
+
+## Required
+
+- [The task fixture outside the gate](/quest/m0/checks/task-fixture.md) - the `task` differential passes on every host, not only in the gate's container
+- [A cheaper differential](/quest/m0/checks/differential.md) - fixtures' wasm builds at opt-level 1 under single-threaded Node, and `stable_names` reuses `lang`'s translation
+- [Checks by the diff](/quest/m0/checks/dispatch.md) - `just ci` runs only the projects a change touches, as moq's `sh/dispatch.sh` does, with a nightly `just gate`
+- [A warm gate](/quest/m0/checks/cache.md) - the land gate builds through mbx, as moq's does
+- [Stable names on a small fixture](/quest/m0/checks/stable-names.md) - `stable_names` runs on a small fixture with generic instances and trait impls
+- [Fixtures that merge](/quest/m0/checks/fixture-layout.md) - `lang` splits by module, and `reject`'s snapshots key by item, not line

quest/m0/checks/cache.md

@@ -0,0 +1,16 @@
+# [S] A warm gate
+
+## Goal
+
+The land gate's `just ci` builds through mbx, as moq's does, so a run rebuilds only what
+changed: measured on nixos2, root clippy falls from 399 to 83 CPU-seconds warm, and uniffi's
+test build from 291 to 20.
+
+## Plan
+
+Decided with the maintainer on 2026-11-01: mbx in a per-repository cache mount, as moq's
+gate has. PR code can write to it, a risk moq already accepts; mbx keys each compilation on
+its compiler, target, and inputs. Local worktrees share it too, where the dev shell already
+can.
+
+Public API: none.

quest/m0/checks/differential.md

@@ -0,0 +1,18 @@
+# [S] A cheaper differential
+
+## Goal
+
+The ts differential fixtures run their native wasm32 side built at opt-level 1, under
+`node --single-threaded`, and `stable_names` reuses the `lang` test's translation as its
+base, with every case still compared.
+
+## Plan
+
+Measured by the 2026-11-01 checks audit on nixos2: `lang`'s differential spends about 590 of
+its 872 CPU-seconds running a 212 MB debug wasm in V8. At opt-level 1 it takes 164, with 0
+differences across 709k cases. Overflow checks and debug assertions stay on, so panics and
+overflow still match. Apply it to every differential fixture. `stable_names`'s base config
+equals `lang`'s roots, so its first translation repeats the `lang` test's (about 150
+CPU-seconds).
+
+Public API: none.

quest/m0/checks/dispatch.md

@@ -0,0 +1,18 @@
+# [M] Checks by the diff
+
+## Goal
+
+`just ci` runs only what a change touches, against the base the land gate passes as
+`QUEST_LAND_BASE` or `origin/main` locally, as moq's `sh/dispatch.sh` does, and `just gate`
+runs the full suite nightly on `main`. A quest-only change runs `quest check` and the lints.
+
+## Plan
+
+Decided with the maintainer on 2026-11-01: copy moq's dispatch. Since the gate started,
+147 of 201 PRs touched only ts and quest files, 25 only quest files, 2 uniffi, and 16 ffi;
+uniffi has never failed in the gate. Shared inputs (the flake, toolchains, the root
+`Cargo.*` and justfile, `.github`) select everything. Add `just ci-select` to show the
+selection, and a nightly job running `just gate`, which the nightly lacks today. Update
+CONTRIBUTING.md's description of the gate to say what moq's does.
+
+Public API: none.

quest/m0/checks/fixture-layout.md

@@ -0,0 +1,16 @@
+# [M] Fixtures that merge
+
+## Goal
+
+`lang` splits into fixtures by its modules, sharing one wasm target dir, and `reject`'s
+snapshots key each refusal by its item rather than its line number, so two PRs adding cases
+rarely conflict and a targeted run builds only its module.
+
+## Plan
+
+Decided with the maintainer on 2026-11-01. Of 71 merge-conflict ejections in four days, 43
+were in `lang` (31.7k lines, 108k lines of TS) and 28 in `reject`, and gate failure #374 was
+`reject`'s line numbers shifting inside a batch. Each conflict costs a fixer run, a local
+`just ci`, and another gate run.
+
+Public API: none.

quest/m0/checks/stable-names.md

@@ -0,0 +1,15 @@
+# [S] Stable names on a small fixture
+
+## Goal
+
+`stable_names` checks that output names stay stable on a small fixture of its own that still
+has generic instances and trait impls, instead of translating all of `lang`.
+
+## Plan
+
+Decided with the maintainer on 2026-11-01: it caught 3 failures nothing else did, so it
+stays in the gate, but on a fixture that costs a fraction of `lang`'s (about 300
+CPU-seconds saved after [the reuse](/quest/m0/checks/differential.md)). Cover each naming
+rule it relies on in `lang` today.
+
+Public API: none.

quest/m0/checks/task-fixture.md

@@ -0,0 +1,16 @@
+# [S] The task fixture outside the gate
+
+## Goal
+
+The ts `task` differential passes on every host `just ci` runs on, as it does in the land
+gate's container, so an agent's local `just ci` on `main` is green.
+
+## Plan
+
+Found by the 2026-11-01 checks audit: on `main` at `8e6a7219`, `task` fails on nixos2 and in
+the fixer's container, alone or in parallel, with "35417 cases, 3059 panics, 628
+differences", and passes in the gate. It made the fixer hand back #387, #393, #394, and #397.
+Reproduce it, find what differs between the hosts (Node's version, threads, timers, the
+environment), and fix the cause, never by loosening the comparison.
+
+Public API: none.