experiment-os: 'run' self-registers the declared cohort (fresh-DB bootstrap)

The scheduler ran 'experiments run/adjudicate' but nothing registered the five
experiments/*.json into the experiments table, so on a fresh DB the harness
replayed over an empty set and the ledger stayed empty forever (owner hit this
on first deploy). registry.register_cohort() registers every cohort JSON
idempotently (skips already-registered, tolerates a malformed file); 'run'
calls it at startup so a fresh deployment self-bootstraps with no manual
register step. +test.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

research/experiments/cli.py

@@ -39,9 +39,14 @@ def main(argv=None) -> int:
         return 0
 
     if cmd == "run":
-        from research.experiments import replay
+        from research.experiments import registry, replay
         store = _store()
         try:
+            # self-bootstrap: register the declared cohort (idempotent) so a fresh
+            # DB replays the experiments instead of silently running over nothing.
+            newly = registry.register_cohort(store.con)
+            if newly:
+                print(f"registered cohort: {', '.join(newly)}")
             n = replay.run(store)
             print(f"experiments run: {n} new run rows")
         finally:

research/experiments/registry.py

@@ -15,6 +15,7 @@ from datetime import datetime, timezone
 from pathlib import Path
 
 _SCHEMA_SQL = Path(__file__).resolve().parents[2] / "storage" / "experiments_schema.sql"
+_COHORT_DIR = Path(__file__).resolve().parents[2] / "experiments"
 
 # Every experiment JSON must carry these; fidelity must carry these floors.
 _REQUIRED = ("id", "hypothesis", "population", "intervention", "bars", "fidelity")
@@ -75,6 +76,28 @@ def register(con, json_path: str | Path) -> str:
     return exp_id
 
 
+def register_cohort(con, cohort_dir: str | Path | None = None) -> list[str]:
+    """Register every ``experiments/*.json`` not already registered (idempotent).
+
+    The declared cohort must be in the ``experiments`` table before ``run`` can
+    replay anything — otherwise the harness silently does nothing on a fresh DB.
+    ``run`` calls this at startup so a fresh deployment self-bootstraps with no
+    manual ``register`` step. Already-registered ids are skipped (immutable);
+    a malformed file is logged and skipped, never fatal to the run."""
+    apply_schema(con)
+    d = Path(cohort_dir) if cohort_dir else _COHORT_DIR
+    newly: list[str] = []
+    for f in sorted(d.glob("*.json")):
+        try:
+            newly.append(register(con, f))
+        except ValueError as exc:
+            # already registered (the common nightly case) or malformed — skip it
+            if "already registered" not in str(exc):
+                import logging
+                logging.getLogger(__name__).warning("skip cohort file %s: %s", f.name, exc)
+    return newly
+
+
 def _row_to_dict(row) -> dict:
     return {
         "id": row[0],

tests/unit/experiments/test_registry.py

@@ -47,3 +47,19 @@ def test_list_running_only(store, tmp_path):
     store.con.execute("UPDATE experiments SET status='failed' WHERE id='b_v1'")
     running = registry.list_experiments(store.con, running_only=True)
     assert [e["id"] for e in running] == ["a_v1"]
+
+
+def test_register_cohort_bootstraps_and_is_idempotent(store):
+    """run's self-bootstrap: register_cohort registers every experiments/*.json
+    from the real cohort dir, and a second call adds nothing (idempotent)."""
+    from research.experiments import registry
+    first = registry.register_cohort(store.con)
+    assert len(first) == 5                       # the shipped first cohort
+    assert "pm1_v1" in first and "time_stop_10m_v1" in first
+    # every one landed as a running experiment
+    running = {e["id"] for e in registry.list_experiments(store.con, running_only=True)}
+    assert set(first) <= running
+    # idempotent: re-running registers nothing new, raises nothing
+    second = registry.register_cohort(store.con)
+    assert second == []
+    assert len(registry.list_experiments(store.con)) == 5