From 2ad14a915dc2c47f12abe413a87f5f896ec5d47c Mon Sep 17 00:00:00 2001 From: twentyOne2x Date: Wed, 15 Jul 2026 04:04:33 +0200 Subject: [PATCH] Fix live benchmark preflight embedding --- .../blinded-protocol.json | 6 +++--- scripts/run_leo_m3taversal_oos_handler_suite.py | 2 +- tests/test_run_leo_m3taversal_oos_handler_suite.py | 2 ++ 3 files changed, 6 insertions(+), 4 deletions(-) diff --git a/docs/reports/leo-oos-reasoning-benchmark-20260715/blinded-protocol.json b/docs/reports/leo-oos-reasoning-benchmark-20260715/blinded-protocol.json index 812ae40..a464d50 100644 --- a/docs/reports/leo-oos-reasoning-benchmark-20260715/blinded-protocol.json +++ b/docs/reports/leo-oos-reasoning-benchmark-20260715/blinded-protocol.json @@ -34,10 +34,10 @@ "seed_commitment_sha256": "5c3e3f42b37bddae727aaedc1fab882935bdcb5be656635ec6ecabadf1c50cd4", "seed_not_embedded": true }, - "created_at_utc": "2026-07-15T02:02:01.785948+00:00", + "created_at_utc": "2026-07-15T02:04:20.353700+00:00", "frozen_before_live_execution": true, "generator_version": "blinded-family-generator-v2", - "protocol_hash_sha256": "9567ad33671c786f8a1b63b4608ef18c60107002f3d5a47271cd8c687128445d", + "protocol_hash_sha256": "e0a1fe6e8b878bdfb895648875cedfe6ed74eb9c26f32049316b8cfe54df27c7", "protocol_id": "leo-m3taversal-oos-5c3e3f42b37bddae", "schema": "livingip.leoM3taversalOosProtocol.v1", "scorer_version": "invariant-reasoning-live-receipts-and-factual-ablation-v2", @@ -49,7 +49,7 @@ "db_context_plugin_sha256": "b546368c398ab21fe6b6912763fbcd736ead566e1286cf303ae3f4173f2c33f7", "execution_manifest_sha256": "07f62d80634bdcc34db544317304bdb0f3b8d0ea16818bbf9430f6b15c271c47", "generic_handler_sha256": "3321a783aaf7cf34b70d53147f123a81757480e5c9c09bb854b9723eb129da97", - "handler_runner_sha256": "e0cdb1018e074ed3e00d167ee63e57eb67813f3ea0e24347901c39302f63680d", + "handler_runner_sha256": "7db5344a8452548ee901a067e509de16944aae32a775cea88f1fc1b4405f6bde", "kb_tool_sha256": "d0f5c5a38d1c65b39ebf688e53adbd7e202639a583ac00814a193aea7dd5fab1", "protocol_module_sha256": "68a747922d17bac16a67051c4a2bb6b4949c06956f17289d95c90d883778eea2", "readonly_guard_sha256": "bce1c44420c5e374fdb0edeb6bba2f5a4c2096aaeee94ea9153c26e761f6b630", diff --git a/scripts/run_leo_m3taversal_oos_handler_suite.py b/scripts/run_leo_m3taversal_oos_handler_suite.py index e2c4ffa..412312b 100755 --- a/scripts/run_leo_m3taversal_oos_handler_suite.py +++ b/scripts/run_leo_m3taversal_oos_handler_suite.py @@ -113,7 +113,7 @@ def build_guarded_remote_script( + "\nOOS_GUARD_SOURCE_SHA256 = " + json.dumps(hashlib.sha256(guard_source.encode()).hexdigest()) + "\nOOS_PROMPT_LEAKAGE_SCAN = " - + json.dumps(leakage_scan) + + repr(leakage_scan) + "\n\n" + function_marker, ) diff --git a/tests/test_run_leo_m3taversal_oos_handler_suite.py b/tests/test_run_leo_m3taversal_oos_handler_suite.py index a485ae9..fc8b1b9 100644 --- a/tests/test_run_leo_m3taversal_oos_handler_suite.py +++ b/tests/test_run_leo_m3taversal_oos_handler_suite.py @@ -33,6 +33,8 @@ def test_guarded_remote_script_compiles_and_installs_preexecution_gate(grounding leakage_scan={"pass": True, "exact_prompt_matches": []}, ) compile(script, "", "exec") + assert "OOS_PROMPT_LEAKAGE_SCAN = {'pass': True" in script + assert 'OOS_PROMPT_LEAKAGE_SCAN = {"pass": true' not in script assert "install_read_only_tool_surface" in script assert "trace_payload_sha256" in script assert 'report["preexecution_safety_gate"]' in script