Do not attribute the harness's own filesystem actions to the agent

`mkdir -p /workspace/scratch`, which our setup solver runs before every agent, classifies
as a write inside the directory. The scorer's `find` and `test -d` classify as reads.
Counted as the agent's, every run would report a write and the Phase 2 headline number
would be 100% for a reason that has nothing to do with any agent. It would have looked
like a spectacular result.

Inspect nests each tool execution in a span of type "tool", so the agent's own actions
are exactly the sandbox events inside one. Analysis now filters on that.

scripts/verify_snapshot.py drives the real task in a real container with a scripted model
that writes a note, and checks the note comes back. Without it, a broken snapshot would
look identical to a true null result: agents simply not writing anything. That is the
failure this pilot exists to avoid, so it is worth a script.

Claude-Session: https://claude.ai/code/session_01Cq98H7sNoSJdL3W98f18bu

justfile

@@ -39,3 +39,7 @@ pilot:
 # Rebuild the CSV and hand-read file from an existing log, no re-run.
 report log_dir="logs/pilot":
     uv run python scripts/report.py {{log_dir}}
+
+# Proves the scorer captures a file an agent leaves in scratch. Free, needs Docker.
+verify-snapshot:
+    uv run python scripts/verify_snapshot.py