Harness step 2: live suites as tables

Orientation

Step 2 of mo-harness-in-mo, after mo-harness-step-1-executor. The Python under toolchain/harness/ is being cut to what its role needs before most of it moves to Mo. Six hand-written case lists become one runner and case tables; tests that cannot fail are replaced by ones that can. Review findings: audit/evidence/2026-09-19/fable-overnight-review/README.md (E3, and the Tests notes under Workspace HTTP). Worker: a fresh Claude Opus session, bypass permissions, own worktree and Herdr tab. The lead owns wiki, audit, acceptance.

Write scope

toolchain/harness/executor/** and toolchain/harness/provider/** (any file type, so the stale READMEs can be fixed). Not evidence/, attempts/ or sources/ directories: history. No Mo source, compiler, mo-wiki/, audit/, HANDOFF.md. You may not run anything against the machine mo-executor-r01, Docker or /opt: another worker may be using the host, and the live reruns are the lead’s. No push. Never use tr.

Parts

  1. One live runner, case tables. recovery/live.py, workspace_http/live.py, workspace_http/local.py, test_workspace_live.py, selftest.py and application/controls.py keep every case they have today, by name, with the same command lines and the same printed summary shapes (the lead’s reruns and records quote them: {"count": 17, "passed": 17, ...}, {"controls": 22, ...}, {"groups": 22, ...}, {"controls": 23, ...}). Shared setup, run, cleanup and reporting move into one module; each suite becomes a table of cases plus the few functions that are really its own. Step 1 changed two injection strings in recovery/live.py; keep them.
  2. E3. test_foreign_state_retained, test_missing_root_is_not_runtime_absence, test_dispatched_missing_storage_requires_proof and test_conflicting_local_manifest_refused_before_transport assert the pre-seeded default. Rewrite each so it fails when recover()’s body is broken (show it: replace read_json with a function that raises, keep the output). Do not rewrite recover() itself; that moves to Mo in step 5.
  3. The HTTP test double. WorkspaceDouble.call ignores args. Give the local suite real coverage of path containment, exact_edit matching (missing, multiple, empty) and truncation, without the machine, and stop tests mutating production globals where a parameter will do.
  4. Stale references. provider/README.md, provider/auth/README.md, provider/bridge/README.md and provider/auth/attempt.sh name runners step 1 deleted; point them at executor/guarded.py. recovery/readiness_probes.py fails to import: fix or delete it, saying which and why.

Numbers and done when

All local suites pass under toolchain/bench/step36/guard.py with real summary lines and exit codes (before: executor 96, application 25, local.py 22). Counted Python lines (non-blank, non-comment) before and after, per directory, tests and other code separately, with the command. A list of every live case name before and after, proving none was lost. Small commits as yourself with a Co-Authored-By line naming your model. Write your final report to toolchain/harness/executor/STEP-2-REPORT.md and commit it: commits, decisions the brief did not cover, what the lead must rerun on the machine.

Result

Accepted 19 Sep 2026, 10:37 AM ET (decision-log); full suite 244 of 244. The worker’s report is committed with its work; the lead’s runs are in audit/evidence/2026-09-19/fable-lead-verification/.