refactor(test): adopt native pytest-testmon lifecycle

.github/workflows/nightly-scale.yml

@@ -1,6 +1,6 @@
 name: Nightly Scale
 
-# Runs the large-tier scale fixtures (issue #1183) outside the normal
+# Runs the explicit large benchmark nodes outside the normal
 # CI critical path, captures benchmark results as structured JSON, and
 # compares against the committed baseline to detect regressions (#1220).
 #
@@ -41,16 +41,19 @@ jobs:
           python-version: "3.14"
       - run: uv sync --extra dev --frozen
 
-      - name: Run large-tier scale tests
+      - name: Run large benchmark campaign
         env:
           POLYLOGUE_FORCE_PLAIN: "1"
           HYPOTHESIS_PROFILE: ci
         run: |
-          uv run pytest -q --tb=short --ignore=tests/integration \
-            -m "scale_large" -p no:randomly -n 0 \
+          uv run pytest -q --tb=short -p no:randomly -n 0 \
+            --benchmark-enable \
             --benchmark-json=nightly-results.json \
             --benchmark-group-by=group \
-            tests/benchmarks
+            'tests/benchmarks/test_daemon_convergence.py::test_convergence_scale_tier[xxl-mega-session]' \
+            tests/benchmarks/test_daemon_convergence.py::test_convergence_huge_session_memory_bounded \
+            tests/benchmarks/test_scale_tiers.py::test_bench_50k_fts_search_returns_results \
+            tests/benchmarks/test_scale_tiers.py::test_bench_50k_list_sessions
 
       - name: Upload benchmark results
         uses: actions/upload-artifact@v7

CLAUDE.md

@@ -336,12 +336,13 @@ testmon-affected set.
 
 - `mypy --strict` (via `devtools verify`) is the primary net for type/identifier
   refactors — trust it. Config in `pyproject.toml`, no exclude list.
-- Seed testmon on a fresh checkout / after harness or dependency changes:
-  `devtools verify --seed-testmon --skip-slow`. A linked worktree auto-bootstraps
-  its testmon cache from the main checkout's valid seed instead (`devtools/testmon_bootstrap.py`),
-  so this is only needed when the main checkout itself has no seed yet.
-- Reserve `devtools verify --all` (full non-integration run) for
-  harness/dependency changes or a final pre-PR diagnostic.
+- Plain `devtools verify` owns the native pytest-testmon lifecycle. It repairs
+  invalid local state, optionally copies a matching main-checkout database,
+  and automatically runs the complete correctness corpus when no valid native
+  environment exists. Never ask an operator or agent to seed or repair it.
+- Reserve `devtools verify --all` (complete unit/property/fuzz/integration
+  correctness corpus; performance benchmarks excluded) for harness/dependency
+  changes or a final pre-PR diagnostic.
 - `devtools verify --quick` = format + lint + mypy + `render all --check`
   (no tests); it runs on `git push` via the pre-push hook. It is a fast gate,
   not a substitute for the default baseline before a PR.
@@ -522,7 +523,7 @@ isolated XDG paths + archive root.
   without a managed run identity or an explicit basetemp root is forced to
   `/realm/tmp/polylogue-pytest` (NVMe). Managed `devtools test` and
   `devtools verify` runs may use bounded `/dev/shm` tmpfs only after the
-  runtime policy admits the requested demand; full-suite and seed-testmon
+  runtime policy admits the requested demand; full/bootstrap native
   runs default to NVMe because their aggregate fixture tree can exceed the
   supervised tmpfs ceiling.
   `POLYLOGUE_PYTEST_BASETEMP_MIN_FREE_MB` overrides required headroom; an
@@ -565,7 +566,7 @@ Core loop:
   changing docs, CLI help, or schema. **Gotcha:** `render all --check` can print
   per-surface `sync OK` yet still exit 1 — grep the output for `out of sync`,
   don't trust the tail line.
-- `devtools verify [--quick|--all|--lab|--seed-testmon]` — see
+- `devtools verify [--quick|--all|--lab]` — see
   [Verification](#verification--testmon-inner-loop-never-blanket-run).
 - `devtools test <sel>` — focused pytest through the managed harness.
 - `devtools lab …` — executable schema/provider/pipeline/lane checks.

CONTRIBUTING.md

@@ -55,9 +55,8 @@ targeting `master`.
 2. Create a branch from `origin/master`.
 3. Work on the branch. Git hooks enforce format and lint on commit, and
    run `devtools verify --quick` on push.
-4. Run `devtools verify` before creating the PR. The default pytest step uses
-   pytest-testmon affected-test selection; run `devtools verify --seed-testmon`
-   first if the dependency database is not seeded.
+4. Run `devtools verify` before creating the PR. It automatically builds or
+   repairs native pytest-testmon state, then uses affected-test selection.
 5. Open a pull request. The template has required sections — fill them
    all in. The PR title becomes the squash-merge subject on `master`.
 6. CI must pass. Fix failures on the branch, do not merge with red CI.
@@ -295,27 +294,29 @@ local pytest selection is accelerated by pytest-testmon.
 
 ```bash
 devtools verify            # static/generated gates + pytest-testmon affected tests
-devtools verify --seed-testmon --skip-slow  # seed/update affected-test DB
-devtools verify --all      # explicit full non-integration pytest diagnostic
+devtools verify --all      # complete correctness corpus; benchmarks excluded
 devtools verify --quick    # format + lint + mypy + render all --check (skip tests)
 devtools verify --lab      # explicit lab checks beyond the quick/default loop
 ```
 
 The quick gate runs on `git push` via the active pre-push hook. It's a fast
-check, not a substitute for the default baseline. The default command fails
-fast when `.cache/testmon/testmondata` and `.cache/testmon/seed.json` are
-missing; do not rely on silent full-suite fallback.
+check, not a substitute for the default baseline. The default command safely
+repairs missing or invalid native testmon state and automatically builds a new
+environment when collection semantics change.
 
 `devtools verify` does not replay a prior verify result. It always runs the
-static gates and then invokes pytest-testmon for affected-test selection from
-the current source, dependency, and Python-version state. The default pytest
-step combines marker filters with `--testmon-forceselect` so scale-tier
-deselection does not silently expand the run back to the whole suite; affected
-testmon runs are single-process by default to avoid xdist collection skew.
-Polylogue does not maintain a parallel changed-file router for helper/config
-paths; use `devtools verify --seed-testmon` when you intentionally want to
-refresh the dependency database and `devtools verify --all` for an explicit full
-diagnostic.
+static gates. With a valid native environment, it invokes pytest-testmon for
+affected-test selection from the current source, dependency, and Python-version
+state. When it bootstraps a missing or invalid native environment, it runs the
+complete correctness corpus. The pytest step covers unit, property, fuzz, and
+integration tests while excluding the separately operated `tests/benchmarks`
+performance surface. It uses
+`--testmon-forceselect` for affected selection, with one parallel `not
+load_sensitive` lane and one serial `load_sensitive` lane over the same native
+environment. `tui` is a category marker and remains parallel unless a test is
+also explicitly `load_sensitive`. Use
+`devtools verify --all` for the complete correctness corpus; there is no
+manual seed or repair command.
 
 Add `devtools release build-package` or `nix flake check` when touching packaging or
 Nix expressions. See [TESTING.md](TESTING.md) and [docs/devtools.md](docs/devtools.md)