fix(devtools): run the corpus and the required check at the pool width

.agentctl/project.toml

@@ -36,11 +36,11 @@ verification-graph = ["devtools/verify.py", "pyproject.toml"]
 
 [operations.verify_affected]
 description = "Run Polylogue's affected-test verification plan"
-exec = ["env", "POLYLOGUE_PYTEST_WORKERS=2", "devtools", "verify"]
+exec = ["devtools", "verify"]
 pool = "pytest"
 result = "pytest"
 cache = "tree+environment"
-timeout_seconds = 3600
+timeout_seconds = 7200
 
 [operations.verify_quick]
 description = "Run Polylogue's static fast verification gates"

devtools/pytest_slot.py

@@ -319,7 +319,7 @@ def _write_launch(
         "argv": list(argv),
         "working_directory": cwd,
         "environment": dict(env),
-        "timeout_seconds": 3600,
+        "timeout_seconds": 7200,
         "result_kind": "exit",
         "log_path": str(log_path),
     }

devtools/run_tests.py

@@ -27,10 +27,11 @@
 import subprocess
 import sys
 import time
+from collections.abc import Mapping
 from pathlib import Path
 from typing import Any, cast
 
-from devtools.agent_env import agent_worker_cap, inside_agent_job
+from devtools.agent_env import HARNESS_RUN_ENV, agent_worker_cap, inside_agent_job
 from devtools.checkout_guard import (
     CheckoutImportMismatchError,
     assert_polylogue_matches_checkout,
@@ -370,16 +371,23 @@ def build_pytest_cmd(selection: list[str]) -> list[str]:
         *collection_args,
         # A focused run traces into the one checkout datafile and writes back,
         # so the graph is advanced by every managed run. It never selects: the
-        # caller already named what to run.
-        "--testmon",
-        f"--testmon-env={TESTMON_ENVIRONMENT}",
-        "--testmon-noselect",
+        # caller already named what to run. A run spawned from inside another
+        # managed run (a test exercising the harness) must not touch that
+        # datafile: its session would reset the outer run's pending graph.
+        *_testmon_args(os.environ),
         *selection,
         *worker_args,
         *_xdist_distribution_args(selection, worker_args),
     ]
 
 
+def _testmon_args(env: Mapping[str, str]) -> tuple[str, ...]:
+    """testmon flags for a focused run; none when nested in a managed run."""
+    if env.get(HARNESS_RUN_ENV):
+        return ("-p", "no:testmon")
+    return ("--testmon", f"--testmon-env={TESTMON_ENVIRONMENT}", "--testmon-noselect")
+
+
 def _selection_targets_benchmarks(selection: list[str]) -> bool:
     """Keep benchmark collection available only when the caller asks for it."""
     return any("tests/benchmarks" in argument for argument in selection)

devtools/verify.py

@@ -76,9 +76,12 @@
 PYTEST_SUMMARY_PATH = PYTEST_REPORT_DIR / "current-pytest-summary.json"
 PYTEST_OUTPUT_PATH = PYTEST_REPORT_DIR / "current-pytest-output.log"
 PYTEST_JUNIT_REPORT_DIR = PYTEST_REPORT_DIR / "junit"
-#: SQLite archive construction makes the corpus IO-bound. Two workers provide
-#: overlap without multiplying cache churn or exhausting the pytest cgroup.
-CORPUS_MAX_WORKERS = 2
+#: One fixed width for the corpus and the runner's affected tier, sized to the
+#: pytest pool's 12 GiB cgroup ceiling (eight workers peak near 10 GB) rather
+#: than host cores or free RAM. Measured 2026-09-03 uncontended: 47 minutes for
+#: 20,860 tests at eight workers; at two the same run takes about seven hours
+#: and the required check cannot finish inside its slot timeout.
+CORPUS_MAX_WORKERS = 8
 _AGENTCTL_OPERATION_ARGV = {"verify_affected": (), "verify_quick": ("--quick",), "verify_all": ("--all",)}
 _PROJECT_DESCRIPTOR = ".agentctl/project.toml"
 # These tests read the AgentCTL descriptor directly. They are the bounded
@@ -293,6 +296,24 @@ def _read_json(path: Path) -> dict[str, Any] | None:
 MAX_RERUN_NODEIDS = 300
 
 
+def _report_nodeid_to_selector(nodeid: str) -> str:
+    """Strip xdist's ``@<group>`` suffix so a report node id selects again.
+
+    ``--dist=loadgroup`` reports ``path::test[param]@group``; pytest cannot
+    collect that literal, so a rerun built from it errors before running.
+    A parametrization id may itself contain ``@``, so only a suffix after the
+    closing bracket (or after the bare test name) is removed.
+    """
+    head, sep, tail = nodeid.rpartition("@")
+    if not sep or "::" not in head:
+        return nodeid
+    if "[" in tail or "]" in tail or "/" in tail or "::" in tail:
+        return nodeid
+    if head.endswith("]") or "[" not in head.rsplit("::", 1)[-1]:
+        return head
+    return nodeid
+
+
 def _rerun_failed_once(command: Sequence[str], *, env: Mapping[str, str], artifacts: Any) -> dict[str, Any] | None:
     """Rerun exactly the failed tests once, alone and unselected.
 
@@ -306,7 +327,7 @@ def _rerun_failed_once(command: Sequence[str], *, env: Mapping[str, str], artifa
     if not isinstance(report, Mapping):
         return None
     failed = [
-        str(test["nodeid"])
+        _report_nodeid_to_selector(str(test["nodeid"]))
         for test in report.get("tests", [])
         if isinstance(test, Mapping) and test.get("outcome") in {"failed", "error"} and test.get("nodeid")
     ]

devtools/verify_testmon_selection.py

@@ -10,7 +10,7 @@
 
 from devtools.pytest_invocation import MANAGED_PLUGIN_ARGS
 from devtools.toolchain import venv_python
-from devtools.verify import _pytest_worker_args
+from devtools.verify import CORPUS_MAX_WORKERS, _pytest_worker_args
 
 
 def main(_argv: list[str] | None = None) -> int:
@@ -27,8 +27,8 @@ def main(_argv: list[str] | None = None) -> int:
     finally:
         if configured_workers is not None:
             os.environ["POLYLOGUE_PYTEST_WORKERS"] = configured_workers
-    if default_worker_args != ["--dist=loadgroup", "-n", "2"]:
-        print("testmon-selection: managed verification does not default to two workers")
+    if default_worker_args != ["--dist=loadgroup", "-n", str(CORPUS_MAX_WORKERS)]:
+        print(f"testmon-selection: managed verification does not default to {CORPUS_MAX_WORKERS} workers")
         return 1
     with tempfile.TemporaryDirectory(prefix="polylogue-testmon-gate-") as temporary:
         root = Path(temporary)
@@ -93,7 +93,7 @@ def main(_argv: list[str] | None = None) -> int:
         if not selected or not total or selected * 100 >= total * 5:
             print(f"testmon-selection: selected {selected} of {total}, expected under 5%\n{output}")
             return 1
-    print(f"testmon-selection: selected {selected} of {total}; workers=2")
+    print(f"testmon-selection: selected {selected} of {total}; workers={CORPUS_MAX_WORKERS}")
     return 0
 
 

tests/unit/devtools/test_pytest_slot.py

@@ -155,7 +155,7 @@ def test_outside_a_task_the_run_is_queued(tmp_path: Path, monkeypatch: pytest.Mo
     assert launch["operation"] == "test"
     assert launch["pool"] == "pytest"
     assert launch["result_kind"] == "exit"
-    assert launch["timeout_seconds"] == 3600
+    assert launch["timeout_seconds"] == 7200
     assert [call["argv"][0] for call in _calls(record)] == ["add", "wait", "status"]
 
 

tests/unit/devtools/test_run_tests.py

@@ -625,3 +625,14 @@ def test_absent_paths_are_resolved_against_the_checkout_not_the_caller_cwd(
     ]
 
     assert run_tests.absent_selection_paths(selection, root=checkout) == ["tests/unit/test_deleted.py"]
+
+
+def test_nested_managed_run_never_traces_into_the_checkout_datafile(monkeypatch: pytest.MonkeyPatch) -> None:
+    """Anti-vacuity: tracing unconditionally lets a test that spawns
+    `devtools test` reset the outer corpus run's graph (2026-09-05: a full
+    corpus left a one-test datafile)."""
+    from devtools.agent_env import HARNESS_RUN_ENV
+    from devtools.run_tests import _testmon_args
+
+    assert "--testmon" in _testmon_args({})
+    assert tuple(_testmon_args({HARNESS_RUN_ENV: "run-1"})) == ("-p", "no:testmon")

tests/unit/devtools/test_verify.py

@@ -914,3 +914,15 @@ def fake_run(command: list[str], **kwargs: Any) -> subprocess.CompletedProcess[s
 
     assert exit_code == 0
     assert recorded["capture_output"] is True
+
+
+def test_rerun_selector_strips_the_xdist_group_suffix() -> None:
+    """Anti-vacuity: passing the report node id through unchanged makes the
+    rerun error with "not found" for every grouped test, which is what wiped
+    the 2026-09-05 corpus rerun."""
+    from devtools.verify import _report_nodeid_to_selector
+
+    assert _report_nodeid_to_selector("tests/a.py::test_x@web-reader") == "tests/a.py::test_x"
+    assert _report_nodeid_to_selector("tests/a.py::T::test_x[p]@grp") == "tests/a.py::T::test_x[p]"
+    assert _report_nodeid_to_selector("tests/a.py::test_x[a@b]") == "tests/a.py::test_x[a@b]"
+    assert _report_nodeid_to_selector("tests/a.py::test_x") == "tests/a.py::test_x"