REPORT — phase5-run-entry-fixes-implementer
STEP: phase5-run-entry-fixes (Phase 5: the run-entry fixes — F11–F16)
REV: aa33970c6d3b8f369b25be99145075fe7f48b64c   FINAL: 223596b55a9f678c743409416d4068ddd0cdfd02
CHANGED: src/orchestrator/run_entry.py — `build_command_executor` resolves the workdir to an absolute path; `validate_run_spec` refuses an obligations-without-axes/rationale spec at load; `start_run` catches `AttemptTimedOut`/`AttemptCancelled` and settles them through `_settle_executor_refusal` (conservative accounting, failure recorded on the node, terminal outcome), reading the run's true outcome from the durable log so a second process's settlement is not overwritten.
CHANGED: scripts/orchestrate.py — the sibling `sys.path` bootstrap (`REPO_ROOT` + `src`); `--workdir` and `--store` resolved to absolute paths at entry.
CHANGED: docs/run-entry.md — the `executor_profile` row now states it is a schema-required label recorded nowhere, with executor selection separate work; the exit-code sentence names every non-accepted class (refused, cancelled, timed out).
CHANGED: tests/test_run_entry.py — six new named shapes: relative-workdir settled start, wall-clock overrun, cancelled launcher, import spellings, clean-env CLI, honest label, fail-closed load.
VERIFIED: scripts/check-tests --fast — 961 passed, 76 deselected in 91.48s (0:01:31), exit 0. The 76 = the review's 66 + this step's 10 new slow cases.
VERIFIED: pytest tests/test_run_entry.py — 31 passed in 9.97s (ran as `.venv/bin/python -m pytest`, the project's configured interpreter; the bare `pytest` on PATH has no `timeout` plugin, so `pytest.ini`'s `strict_config` rejects it).
VERIFIED: pytest tests/test_phase3_project_oracle.py — 5 passed in 297.03s (0:04:57).
VERIFIED: scripts/check-static — exit 0 ("All checks passed!" on the gated BLE001/S110 select; the B/SIM advisory report does not gate).
VERIFIED: scripts/check-docs — exit 0, 340 links, 0 errors.
VERIFIED: scripts/check-scope aa33970c6d3b8f369b25be99145075fe7f48b64c 223596b docs/research/phase5-run-entry-fixes-brief.md — `tripped: none`, `scope: ok (5 path(s))`, exit 0.
NOT VERIFIED: none — every claim is covered by a run named above. Red evidence for F11–F14 and F16 was produced on a scratch worktree at REV (aa33970) with the new pins; F15's FIXED reads `no regression test` because a corrected doc row has no failing assertion at any revision.
FINDINGS:

F11 | SEVERITY: P1 | ORIGIN: CODE | STATE: fixed
  A relative `--workdir` makes `start` crash and leave the run live: `build_command_executor` passed the CLI's workdir verbatim into the executor roots, so the sandboxed `git bundle create` wrote a worktree-relative path and raised `ProtocolViolation` out of `start_run`.
  FIXED: tests/test_run_entry.py::test_settled_start_shape_relative_workdir — at REV (scratch worktree aa33970, `pytest tests/test_run_entry.py -k relative_workdir`) the named failing assertion is `assert completed.returncode == 0` (line 491): the CLI exited 1, empty stdout, `orchestrator.executor_protocol.ProtocolViolation: git bundle create rel-work/artefacts/WT-RUN-REL-root-attempt-*.candidate.bundle HEAD failed`. At FINAL the same command exits 0 with `accepted: true`, `outcome: ACCEPTANCE_CLAIMED`, and the relative refusing form exits non-zero with `outcome: INFEASIBLE` and no traceback.

F12 | SEVERITY: P1 | ORIGIN: CODE | STATE: fixed
  A wall-clock overrun of the root's visible command escaped `start_run` as an unhandled `AttemptTimedOut` (attempt TIMED_OUT / PROVISIONALLY_SETTLED, root READY, `outcome: None`); a mid-run `stop` from a second process likewise left the launcher dying with a traceback.
  FIXED: tests/test_run_entry.py::test_settled_start_shape_wall_clock_overrun — at REV the named failing assertion is `raise AssertionError(f"start_run escaped with {type(error).__name__} ...")` (line 518): `AssertionError: start_run escaped with AttemptTimedOut instead of settling the run through the recorded-refusal path`. At FINAL the run settles `INFEASIBLE`, the attempt is `TIMED_OUT` / `FINALLY_SETTLED`, the root is `INFEASIBLE`, and two failure fingerprints are recorded. tests/test_run_entry.py::test_cancelled_launcher_shape — at REV the named failing assertion is `assert "Traceback" not in stderr` (line 605) after a second-process `stop`; at FINAL the launcher exits non-zero with a JSON document `accepted: false, outcome: CANCELLED` and no traceback.

F13 | SEVERITY: P2 | ORIGIN: CODE | STATE: fixed
  The no-harness-import AST helper collected module paths only from `ast.Import` alias names; `ast.ImportFrom` alias names and relative imports were missed, so `from orchestrator import experiment` and `from . import experiment` stayed green.
  FIXED: tests/test_run_entry.py::test_import_spelling_shape[named-import] and [relative-import] — at REV (helper reverted in the scratch worktree) the named failing assertions are `assert leaves & _FORBIDDEN_MODULES` (line 426): `AssertionError: the pin misses the 'named-import' spelling: ['orchestrator']` and `AssertionError: the pin misses the 'relative-import' spelling: []`; `[absolute-import]` and `[late-function-import]` pass at both revisions. At FINAL all four spellings are flagged.

F14 | SEVERITY: P2 | ORIGIN: CODE | STATE: fixed
  `scripts/orchestrate.py` lacked the sibling `sys.path` bootstrap, so the documented `python3 scripts/orchestrate.py …` died before argparse with `ModuleNotFoundError: No module named 'orchestrator'`.
  FIXED: tests/test_run_entry.py::test_clean_env_cli_shape — at REV the named failing assertion is `assert completed.returncode == 0` (line 618): `ModuleNotFoundError: No module named 'orchestrator'` on `--help`. At FINAL the clean-env `--help`, a real `start` (exit 0, accepted), `inspect`, and `stop` all run from the repo root without `PYTHONPATH`.

F15 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: fixed
  `docs/run-entry.md`'s row claimed the `executor_profile` label is "recorded on the run"; the loader validates it into the `RunConfiguration` and records it nowhere.
  FIXED: no regression test — a corrected doc row has no failing assertion at any revision. The absence pin tests/test_run_entry.py::test_honest_label_shape_records_no_executor_profile asserts the probe label `command-executor-v1` is absent from the event log and the inspect document; it is green at REV and FINAL by design (the drift guard).

F16 | SEVERITY: P2 | ORIGIN: DOCUMENT | STATE: fixed
  `docs/run-entry.md` claimed a malformed spec is refused "before any store is opened", false for the obligations-without-axes class: the CLI created the store and the freeze boundary refused only inside `start_run`.
  FIXED: tests/test_run_entry.py::test_fail_closed_load_shape_obligations_refused_before_store — at REV the named failing assertion is `with pytest.raises(run_entry.RunSpecError):` (line 681): `Failed: DID NOT RAISE RunSpecError` (and the CLI form left `axes.db` on disk). At FINAL `validate_run_spec` raises before `EventStore` is constructed and the CLI leaves the store path absent.

SHAPES: none — every named shape is built in tests/test_run_entry.py (Settled-start relative/overrun, Cancelled-launcher, Import-spelling, Clean-env CLI, Honest-label, Fail-closed-load); nothing probed by hand was left unbuilt.
TRIPPED: none
PROPOSED ROWS: none
METHOD FRICTION: none
INHERITED:
none — every received item (F11, F12, F13, F14, F15, F16) is fixed by this step, so nothing remains open and nothing is carried.
