Record successful Tmax path discovery for all 27 retry tasks

Verify all 27 rerun payload hashes against the 14,564 already-published tasks. Update current discovery status and README coverage; preserve underlying run evidence and unchanged task payloads.

tmax_v1/README.md

@@ -30,7 +30,7 @@ Intended benchmark transfer: **Terminal-Bench 2.0 and OpenThoughts-TBLite 2.0**,
 - **1. Build and setup: 14,564 changed, 2 archived.** Changed all kept tasks to build with Apptainer: 13,943 tasks share two dependency images and keep their task-specific files and setup separate; startup times were measured to make sure the shared images do not slow startup. The other 621 tasks keep their own images. Changed 3 tasks so the services their tests need start reliably. Archived 1 task that needs Docker inside the container and 1 task whose setup cannot remove executable permissions.
 - **2. Reference solution (reward 1): not applicable.** TMax has no reference solutions.
 - **3. No-op baseline (reward 0): 12 changed, 34 archived.** Changed 12 tasks whose tests failed for reasons unrelated to the agent's work, such as missing test inputs or services the tests start themselves. Archived 22 tasks whose verifier did not finish on the unchanged task: in 11, the planted bug (endless loop, deadlock, cycle without a visited check, diverging integrator) runs inside the tests without a time limit, so the no-op hangs until the 120 s verifier timeout; in 7, the tests run a slow reference program or dependency build before checking the agent's work, so a correct solution may also exceed the timeout; 4 crash or hang for other reasons (a background database holding the test output, unsafe code run inside the test process, unbounded service launches). Archived 6 tasks whose unchanged state already scores 1, 4 tasks whose tests contain no test functions, 1 task whose reference program is empty and 1 task whose tests kill their own test process. Manually approved 132 no-op failures where the tests' setup fails because the agent's output or service is missing, which is the expected result when the agent does nothing.
-- **4. Rule-based checks: 14,564 changed, 0 archived.** Changed all 14,564 kept tasks so their instruction ends with the time limit and the request not to cheat. Flagged 1,095 tasks for relative paths. Automatic repair cleared warnings for 349 tasks. 746 tasks still have warnings. Inspected 50: all sampled warnings were harmless (commands run from a directory the instruction names, example archive or path-traversal names, package patterns). Accepted the remaining path warnings after this review.
+- **4. Rule-based checks: 14,564 changed, 0 archived.** Changed all 14,564 kept tasks so their instruction ends with the time limit and the request not to cheat. Flagged 1,095 tasks for relative paths. Completed path discovery for all 1,095 flagged tasks. Automatic repair cleared warnings for 349 tasks. 746 tasks still have warnings. Inspected 50: all sampled warnings were harmless (commands run from a directory the instruction names, example archive or path-traversal names, package patterns). Accepted the remaining path warnings after this review.
 
 Git-history cleanup, timestamp normalization and repository solution-leak checking have not been completed.
 

tmax_v1/metadata/path-discovery-audit-20261027.json

@@ -1,51 +1,49 @@
 {
-  "date": "2026-10-27",
-  "reported_missing_count": 28,
-  "confirmed_empty_discovery_records": 27,
-  "tasks": [
-    "task_000004_aeddda76",
-    "task_000024_d94c710d",
-    "task_000127_7c28cce2",
-    "task_000237_a4f6b704",
-    "task_000239_b8b7e3b5",
-    "task_000348_6d9e8578",
-    "task_002481_9f9d35ab",
-    "task_002579_198b9870",
-    "task_002660_c466c2d8",
-    "task_002701_b8e0731d",
-    "task_002959_ed369f20",
-    "task_003027_5859e181",
-    "task_003163_04d087d7",
-    "task_003235_a325a044",
-    "task_004096_4ffdffa7",
-    "task_004160_2a05c6f2",
-    "task_004371_7147f89d",
-    "task_004470_b3d3970e",
-    "task_005487_63a31840",
-    "task_005692_8df67b9a",
-    "task_006664_c7216ff8",
-    "task_007466_85eb29a3",
-    "task_007820_5d98cd5b",
-    "task_008022_d4fe9be0",
-    "task_008109_fabda92c",
-    "task_008171_71cbcef5",
-    "task_008244_2bcbb52d"
-  ],
-  "flagged_kept_tasks": 1095,
-  "flagged_tasks_without_any_record": [],
-  "failed_discovery_kept_tasks": 27,
-  "image_preparation_statuses": {
-    "cached": 3,
-    "built": 78
+  "status": "completed",
+  "tasks": 27,
+  "setup_passed": 27,
+  "discovery_completed": 27,
+  "job_id": "5253845",
+  "job_elapsed_seconds": 315,
+  "stage_wall_seconds": 130.8,
+  "task_payload_hashes": {
+    "task_000004_aeddda76": "c279c0ff50a049f7b029ee355b717146a610fee55aad3e37e5b250769ddb162c",
+    "task_000024_d94c710d": "68ef747d3d32a693be0215c79243166a1509571d6018f8c0c28a56f3195535a1",
+    "task_000127_7c28cce2": "4c69932b621776ec5fb9103e20dc48c53702209825e0cbde64d0e8af90bf8853",
+    "task_000237_a4f6b704": "dd831c67f11e240b82fcf2810460dded2083e21e4293cad66d2b7fd66b5afe7d",
+    "task_000239_b8b7e3b5": "ecb04ef6f8899a362396204b1d2691d368d6b3b79390bf34d2a6f376966bfc2e",
+    "task_000348_6d9e8578": "263d955428bcbb7c5b3a388b01b4ed11403ea9d0747e95d032342bce09a8686f",
+    "task_002481_9f9d35ab": "653e20eec2ee78f1188838018b33f72bf7c206f88ba60ca178f0349662ee150b",
+    "task_002579_198b9870": "9c7b9ca4561fbe1d00445ff784ae80f2ff11df674870a43b40572c3108a23989",
+    "task_002660_c466c2d8": "06454c000fa23f5941e270f31f2381566c6d537284cf4ba067bc5610eaa0dcc6",
+    "task_002701_b8e0731d": "876a5ba138b2aac1cf18b5b9863d30705715b315a3798b06c1081dcbde718f44",
+    "task_002959_ed369f20": "04cf1a03ef3548d1e002ed3214a37c9914fdfb0e34af1d47b9340098c4a251e6",
+    "task_003027_5859e181": "d1d107b07ac03ee16a27942a9d917f4f5d84648209a5771e1bbcfd0c72760762",
+    "task_003163_04d087d7": "721260c9c967320ae8846b81959f2ab5fb067f36214aa179d8af8cb67c2fd569",
+    "task_003235_a325a044": "4bd7239a58888f8cf7937ed6859f2a931cafc7ceb21bb92c1387e0f33d228c3b",
+    "task_004096_4ffdffa7": "bed88baef9f1ad441cb55c8b39d71704df0a82d09d1737c1a941bfd4ab8ec470",
+    "task_004160_2a05c6f2": "e4b88fd4a838d57802a63cea16eba570ac112bd272cbb8357ccc6d3218299c2e",
+    "task_004371_7147f89d": "e845b169c49aa9b11f40f2b6bf61661e589da8354897a22d5301bad388edd862",
+    "task_004470_b3d3970e": "1c5e4539b3d675b8dd3ca73898448f16347c4203d8015c80ccd3e33c49fb16c4",
+    "task_005487_63a31840": "267d0301bfb1af160ee50178947bd611fd5429cdb0849af427b445bae8d330f3",
+    "task_005692_8df67b9a": "aa527bb708a80481404e7e1cc3fea0143f389f45c65b99f4039d05dc19271171",
+    "task_006664_c7216ff8": "e3d7ddd4ea70a40b4b551fa4c11269d422ee047ab58b43af3c9cd920e52ed0a7",
+    "task_007466_85eb29a3": "da1e76922458b09be5901f47336deffd89f828a8c5364b731e20e6bcc7dd923a",
+    "task_007820_5d98cd5b": "0b7b0016efe46e7cad6319610856c9252692558e109d25396bafd2cc683118eb",
+    "task_008022_d4fe9be0": "e72f91f49045717d76c8481c4cb2dc5aadc242501b1728724dc2ff6cc1f6a548",
+    "task_008109_fabda92c": "84f37809ba8e807eb11490be6e3e1d0b42b91c7d36ee3e9b6fe306d06668b019",
+    "task_008171_71cbcef5": "beb04a0330290725f0d698c69b97012d147c24e5eb131751b0ef974c0e919a07",
+    "task_008244_2bcbb52d": "beabcf252612f5b517fe54aa2c2504cf070ceec830f0147235fbecb6a1de81cd"
   },
-  "findings": [
-    "All 27 errors have empty agent_baseline, no observations, and no saved reason.",
-    "All 1095 flagged kept tasks have a resolver record; 27 are error records. No 28th absent record identified.",
-    "Current discovery wrapper discards build_task environments/error details, retaining only status and instruction_paths; this is consistent with the archived empty records. Frozen implementation verification interrupted by SSH timeout.",
-    "Archived setup directory for task_000004_aeddda76 contains directories only, no per-task diagnostic file.",
-    "Image preparation records contain 78 built and 3 cached entries, zero failed entries. Shared cache write warnings explicitly say local image remains usable; no demonstrated causal link to the 27 discovery errors."
-  ],
-  "root_cause": "Underlying per-task preparation/inspection exception not recoverable from the records inspected; do not classify as task fault or cache fault without additional evidence.",
-  "accepted_static_warnings": 746,
-  "next_step": "Retry discovery for the exact 27 tasks with verified target-cluster images, prebuilt-image enforcement, and persisted full preparation diagnostics."
+  "published_payloads_verified": true,
+  "path_findings_by_status": {
+    "unsupported": 5,
+    "missing": 20,
+    "unique": 22
+  },
+  "task_payloads_changed": false,
+  "published_tasks": 14564,
+  "flagged_tasks_with_completed_discovery": 1095,
+  "report": "metadata/path-discovery-stage1-20261027.json",
+  "note": "Completed discovery records filesystem findings; it does not assert that every path has a unique match or remove accepted static warnings."
 }

tmax_v1/metadata/release-review-20261027.json

@@ -139,7 +139,7 @@
   ],
   "task_payloads_unchanged": true,
   "path_discovery": {
-    "status": "incomplete, accepted by review",
+    "status": "completed",
     "tasks": [
       "task_000004_aeddda76",
       "task_000024_d94c710d",
@@ -170,8 +170,11 @@
       "task_008244_2bcbb52d"
     ],
     "count": 27,
-    "acceptance": "Explicit user acceptance; omit this operational detail from README. Original errors remain preserved.",
-    "original_records": "path-discovery-audit-20261027.json"
+    "setup_passed": 27,
+    "discovery_completed": 27,
+    "job_id": "5253845",
+    "published_payloads_verified": true,
+    "report": "path-discovery-stage1-20261027.json"
   },
   "static_path_warnings": {
     "count": 746,
@@ -181,7 +184,7 @@
   },
   "historical_validation_report": "validation-run-20261022.json",
   "historical_report_note": "The original report is preserved as historical evidence; current release counts, image coverage and acceptance decisions are superseded by this amendment and release-status.json.",
-  "stages_passed_semantics": "Acceptance after recorded review; manually accepted no-op results and incomplete path scans are not newly measured automatic passes.",
+  "stages_passed_semantics": "Acceptance after recorded review; manually accepted no-op results are distinguished from automatic passes. Path-discovery completion is supported by the saved stage-1 reports.",
   "images": {
     "bundles": 623,
     "tasks": 14564,

tmax_v1/release-status.json

@@ -24,9 +24,13 @@
   "archived_tasks": 36,
   "review_metadata": "tmax_v1/metadata/release-review-20261027.json",
   "path_discovery": {
-    "status": "incomplete, accepted by review",
+    "status": "completed",
     "tasks": 27,
-    "acceptance": "User approved keeping these tasks without rerunning discovery; no successful scan is asserted."
+    "setup_passed": 27,
+    "discovery_completed": 27,
+    "flagged_tasks_with_completed_discovery": 1095,
+    "job_id": "5253845",
+    "report": "tmax_v1/metadata/path-discovery-stage1-20261027.json"
   },
   "accepted_static_path_warnings": 746
 }