G7.6: never let bookkeeping kill the monitor

An unwritable state path raised and took the whole canary down. That is the
worst possible trade for a monitoring tool: it reports nothing at all, and
reports it silently. State is an optimisation for transition detection; the
probing is the point.

Found by fat-fingering an env var, which is exactly how it would happen in
production.

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

api/tests/test_invariants.py

@@ -601,3 +601,11 @@ def test_g35_did_not_disable_validation_to_register():
     one-time ordering problem."""
     for f in (ROOT / "scripts").glob("*.py"):
         assert "skip_validation" not in f.read_text()
+
+
+def test_g76_canary_survives_an_unwritable_state_path():
+    """A monitoring tool that dies of a config problem reports nothing at all,
+    and reports it silently. State is an optimisation; probing is the point."""
+    src = (ROOT / "scripts" / "canary.py").read_text()
+    save = src[src.index("def save_state") : src.index("def send_alert")]
+    assert "except OSError" in save

scripts/canary.py

@@ -156,8 +156,20 @@ def load_state() -> dict:
 
 
 def save_state(results: list[Result]) -> None:
-    with open(STATE_PATH, "w") as f:
-        json.dump({r.name: r.status for r in results}, f, indent=2)
+    """Never let bookkeeping kill the monitor.
+
+    State is an optimisation — it lets the next run tell "still broken" from
+    "just broke". The probing is the valuable part. An unwritable path used to
+    raise here and take the whole canary down, which is the worst possible
+    trade: a monitoring tool that dies of a config problem reports nothing at
+    all, and reports it silently.
+    """
+    try:
+        with open(STATE_PATH, "w") as f:
+            json.dump({r.name: r.status for r in results}, f, indent=2)
+    except OSError as exc:
+        print(f"!! could not save state to {STATE_PATH}: {exc}")
+        print("   (probes still ran; transition detection is degraded this run)")
 
 
 def send_alert(subject: str, lines: list[str]) -> bool: