Datasets:
platform stringclasses 4
values | fixture_version stringclasses 1
value | source_sha256 stringclasses 2
values | generated_at stringclasses 4
values | node stringclasses 3
values | arch stringclasses 2
values | experiment stringclasses 4
values | title stringclasses 4
values | change stringclasses 4
values | variant stringclasses 2
values | scenario stringclasses 4
values | client stringclasses 1
value | timeout_ms int64 100 2k | max_attempts int64 1 5 | retry stringclasses 1
value | outcome stringclasses 4
values | attempts int64 1 5 | elapsed_ms listlengths 1 5 | transport_errors int64 0 0 | findings listlengths 0 3 | messages stringclasses 7
values | trace_sha256 stringclasses 7
values | suite_pass bool 1
class |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
local | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T05:46:52.225Z | v24.22.0 | arm64 | transient-recovery | Give a transient failure a bounded recovery window | One attempt → three attempts. The same tool fails twice, then succeeds. | before | transient-error | official-sdk-reference-client | 2,000 | 1 | tool_error_only | tool_error | 1 | [
5
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | 14806ab6057f6c8ee28a864697ebe38a634032f184ac1abc3288a21a48e1ceaa | true |
local | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T05:46:52.225Z | v24.22.0 | arm64 | transient-recovery | Give a transient failure a bounded recovery window | One attempt → three attempts. The same tool fails twice, then succeeds. | after | transient-error | official-sdk-reference-client | 2,000 | 3 | tool_error_only | success | 3 | [
3,
2,
1
] | 0 | [
"tool_loop",
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | ce922a08104db02be5625a86a53afad9a5739c94b55b0d0c6da3ee3776b00fbc | true |
local | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T05:46:52.225Z | v24.22.0 | arm64 | permanent-failure | Stop wasting retries on a permanently failing tool | Five attempts → two attempts. The tool remains broken; the client stops earlier. | before | tool-error | official-sdk-reference-client | 2,000 | 5 | tool_error_only | tool_error | 5 | [
3,
3,
0,
0,
0
] | 0 | [
"tool_loop",
"retry_storm",
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | 6fe85ae2dbc3d2d2a2ca15b4513e471174d53e4c5fc202f3ae69865140dfda2e | true |
local | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T05:46:52.225Z | v24.22.0 | arm64 | permanent-failure | Stop wasting retries on a permanently failing tool | Five attempts → two attempts. The tool remains broken; the client stops earlier. | after | tool-error | official-sdk-reference-client | 2,000 | 2 | tool_error_only | tool_error | 2 | [
3,
2
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | d08501886fa2fe3eb0d29916f39e6fbc6b1fb418b29cca0dfd2f5a71ede6e81a | true |
local | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T05:46:52.225Z | v24.22.0 | arm64 | deadline-fit | Fit the deadline to the declared tool latency | 100 ms deadline → 2000 ms deadline for a tool with a 400 ms injected delay. | before | delay | official-sdk-reference-client | 100 | 1 | tool_error_only | timeout | 1 | [
102
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Error: reference c... | ca8d0b945d71b0c1814ae234d73f877fdb46b3ebc3a440e95d6c2114ecafbe2a | true |
local | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T05:46:52.225Z | v24.22.0 | arm64 | deadline-fit | Fit the deadline to the declared tool latency | 100 ms deadline → 2000 ms deadline for a tool with a 400 ms injected delay. | after | delay | official-sdk-reference-client | 2,000 | 1 | tool_error_only | success | 1 | [
404
] | 0 | [] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Probe response: pr... | 45543eb60cd9bf030322aa827426f549c995b208427aa6168c0bc8c37390cbb9 | true |
local | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T05:46:52.225Z | v24.22.0 | arm64 | malformed-control | Keep protocol failures terminal | One-attempt budget → five-attempt budget; protocol errors must still stop after one call. | before | malformed | official-sdk-reference-client | 2,000 | 1 | tool_error_only | protocol_error | 1 | [
7
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Error: reference c... | b01d8e246ec581b6d332074ed83420c64eddb36eaff010f3f056bf4dc52fec9d | true |
local | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T05:46:52.225Z | v24.22.0 | arm64 | malformed-control | Keep protocol failures terminal | One-attempt budget → five-attempt budget; protocol errors must still stop after one call. | after | malformed | official-sdk-reference-client | 2,000 | 5 | tool_error_only | protocol_error | 1 | [
3
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Error: reference c... | b01d8e246ec581b6d332074ed83420c64eddb36eaff010f3f056bf4dc52fec9d | true |
aws-lambda | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:45:57.573Z | v24.21.0 | arm64 | transient-recovery | Give a transient failure a bounded recovery window | One attempt → three attempts. The same tool fails twice, then succeeds. | before | transient-error | official-sdk-reference-client | 2,000 | 1 | tool_error_only | tool_error | 1 | [
137
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | 14806ab6057f6c8ee28a864697ebe38a634032f184ac1abc3288a21a48e1ceaa | true |
aws-lambda | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:45:57.573Z | v24.21.0 | arm64 | transient-recovery | Give a transient failure a bounded recovery window | One attempt → three attempts. The same tool fails twice, then succeeds. | after | transient-error | official-sdk-reference-client | 2,000 | 3 | tool_error_only | success | 3 | [
159,
20,
39
] | 0 | [
"tool_loop",
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | ce922a08104db02be5625a86a53afad9a5739c94b55b0d0c6da3ee3776b00fbc | true |
aws-lambda | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:45:57.573Z | v24.21.0 | arm64 | permanent-failure | Stop wasting retries on a permanently failing tool | Five attempts → two attempts. The tool remains broken; the client stops earlier. | before | tool-error | official-sdk-reference-client | 2,000 | 5 | tool_error_only | tool_error | 5 | [
65,
55,
39,
1,
59
] | 0 | [
"tool_loop",
"retry_storm",
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | 6fe85ae2dbc3d2d2a2ca15b4513e471174d53e4c5fc202f3ae69865140dfda2e | true |
aws-lambda | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:45:57.573Z | v24.21.0 | arm64 | permanent-failure | Stop wasting retries on a permanently failing tool | Five attempts → two attempts. The tool remains broken; the client stops earlier. | after | tool-error | official-sdk-reference-client | 2,000 | 2 | tool_error_only | tool_error | 2 | [
60,
20
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | d08501886fa2fe3eb0d29916f39e6fbc6b1fb418b29cca0dfd2f5a71ede6e81a | true |
aws-lambda | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:45:57.573Z | v24.21.0 | arm64 | deadline-fit | Fit the deadline to the declared tool latency | 100 ms deadline → 2000 ms deadline for a tool with a 400 ms injected delay. | before | delay | official-sdk-reference-client | 100 | 1 | tool_error_only | timeout | 1 | [
101
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Error: reference c... | ca8d0b945d71b0c1814ae234d73f877fdb46b3ebc3a440e95d6c2114ecafbe2a | true |
aws-lambda | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:45:57.573Z | v24.21.0 | arm64 | deadline-fit | Fit the deadline to the declared tool latency | 100 ms deadline → 2000 ms deadline for a tool with a 400 ms injected delay. | after | delay | official-sdk-reference-client | 2,000 | 1 | tool_error_only | success | 1 | [
446
] | 0 | [] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Probe response: pr... | 45543eb60cd9bf030322aa827426f549c995b208427aa6168c0bc8c37390cbb9 | true |
aws-lambda | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:45:57.573Z | v24.21.0 | arm64 | malformed-control | Keep protocol failures terminal | One-attempt budget → five-attempt budget; protocol errors must still stop after one call. | before | malformed | official-sdk-reference-client | 2,000 | 1 | tool_error_only | protocol_error | 1 | [
83
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Error: reference c... | b01d8e246ec581b6d332074ed83420c64eddb36eaff010f3f056bf4dc52fec9d | true |
aws-lambda | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:45:57.573Z | v24.21.0 | arm64 | malformed-control | Keep protocol failures terminal | One-attempt budget → five-attempt budget; protocol errors must still stop after one call. | after | malformed | official-sdk-reference-client | 2,000 | 5 | tool_error_only | protocol_error | 1 | [
60
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Error: reference c... | b01d8e246ec581b6d332074ed83420c64eddb36eaff010f3f056bf4dc52fec9d | true |
google-cloud-run-job | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:30:42.000Z | v24.22.0 | x64 | transient-recovery | Give a transient failure a bounded recovery window | One attempt → three attempts. The same tool fails twice, then succeeds. | before | transient-error | official-sdk-reference-client | 2,000 | 1 | tool_error_only | tool_error | 1 | [
12
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | 14806ab6057f6c8ee28a864697ebe38a634032f184ac1abc3288a21a48e1ceaa | true |
google-cloud-run-job | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:30:42.000Z | v24.22.0 | x64 | transient-recovery | Give a transient failure a bounded recovery window | One attempt → three attempts. The same tool fails twice, then succeeds. | after | transient-error | official-sdk-reference-client | 2,000 | 3 | tool_error_only | success | 3 | [
10,
1,
1
] | 0 | [
"tool_loop",
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | ce922a08104db02be5625a86a53afad9a5739c94b55b0d0c6da3ee3776b00fbc | true |
google-cloud-run-job | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:30:42.000Z | v24.22.0 | x64 | permanent-failure | Stop wasting retries on a permanently failing tool | Five attempts → two attempts. The tool remains broken; the client stops earlier. | before | tool-error | official-sdk-reference-client | 2,000 | 5 | tool_error_only | tool_error | 5 | [
9,
1,
1,
1,
1
] | 0 | [
"tool_loop",
"retry_storm",
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | 6fe85ae2dbc3d2d2a2ca15b4513e471174d53e4c5fc202f3ae69865140dfda2e | true |
google-cloud-run-job | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:30:42.000Z | v24.22.0 | x64 | permanent-failure | Stop wasting retries on a permanently failing tool | Five attempts → two attempts. The tool remains broken; the client stops earlier. | after | tool-error | official-sdk-reference-client | 2,000 | 2 | tool_error_only | tool_error | 2 | [
7,
1
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | d08501886fa2fe3eb0d29916f39e6fbc6b1fb418b29cca0dfd2f5a71ede6e81a | true |
google-cloud-run-job | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:30:42.000Z | v24.22.0 | x64 | deadline-fit | Fit the deadline to the declared tool latency | 100 ms deadline → 2000 ms deadline for a tool with a 400 ms injected delay. | before | delay | official-sdk-reference-client | 100 | 1 | tool_error_only | timeout | 1 | [
101
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Error: reference c... | ca8d0b945d71b0c1814ae234d73f877fdb46b3ebc3a440e95d6c2114ecafbe2a | true |
google-cloud-run-job | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:30:42.000Z | v24.22.0 | x64 | deadline-fit | Fit the deadline to the declared tool latency | 100 ms deadline → 2000 ms deadline for a tool with a 400 ms injected delay. | after | delay | official-sdk-reference-client | 2,000 | 1 | tool_error_only | success | 1 | [
413
] | 0 | [] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Probe response: pr... | 45543eb60cd9bf030322aa827426f549c995b208427aa6168c0bc8c37390cbb9 | true |
google-cloud-run-job | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:30:42.000Z | v24.22.0 | x64 | malformed-control | Keep protocol failures terminal | One-attempt budget → five-attempt budget; protocol errors must still stop after one call. | before | malformed | official-sdk-reference-client | 2,000 | 1 | tool_error_only | protocol_error | 1 | [
12
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Error: reference c... | b01d8e246ec581b6d332074ed83420c64eddb36eaff010f3f056bf4dc52fec9d | true |
google-cloud-run-job | 2026-10-14.1 | e35d7973d8b581c1fe3a1c9c70923c8a089e59b05861e1cf2c2110a7c6a1ae01 | 2026-10-15T06:30:42.000Z | v24.22.0 | x64 | malformed-control | Keep protocol failures terminal | One-attempt budget → five-attempt budget; protocol errors must still stop after one call. | after | malformed | official-sdk-reference-client | 2,000 | 5 | tool_error_only | protocol_error | 1 | [
7
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Error: reference c... | b01d8e246ec581b6d332074ed83420c64eddb36eaff010f3f056bf4dc52fec9d | true |
turing-rk1-arm64 | 2026-10-14.1 | 2b5904b9cae9665a4c73414ca289d50b6940cc8a00b388c22b4c22335d1ff95f | 2026-10-26T20:51:38.219Z | v24.22.1 | arm64 | transient-recovery | Give a transient failure a bounded recovery window | One attempt → three attempts. The same tool fails twice, then succeeds. | before | transient-error | official-sdk-reference-client | 2,000 | 1 | tool_error_only | tool_error | 1 | [
10
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | 14806ab6057f6c8ee28a864697ebe38a634032f184ac1abc3288a21a48e1ceaa | true |
turing-rk1-arm64 | 2026-10-14.1 | 2b5904b9cae9665a4c73414ca289d50b6940cc8a00b388c22b4c22335d1ff95f | 2026-10-26T20:51:38.219Z | v24.22.1 | arm64 | transient-recovery | Give a transient failure a bounded recovery window | One attempt → three attempts. The same tool fails twice, then succeeds. | after | transient-error | official-sdk-reference-client | 2,000 | 3 | tool_error_only | success | 3 | [
7,
2,
2
] | 0 | [
"tool_loop",
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | ce922a08104db02be5625a86a53afad9a5739c94b55b0d0c6da3ee3776b00fbc | true |
turing-rk1-arm64 | 2026-10-14.1 | 2b5904b9cae9665a4c73414ca289d50b6940cc8a00b388c22b4c22335d1ff95f | 2026-10-26T20:51:38.219Z | v24.22.1 | arm64 | permanent-failure | Stop wasting retries on a permanently failing tool | Five attempts → two attempts. The tool remains broken; the client stops earlier. | before | tool-error | official-sdk-reference-client | 2,000 | 5 | tool_error_only | tool_error | 5 | [
7,
2,
2,
2,
2
] | 0 | [
"tool_loop",
"retry_storm",
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | 6fe85ae2dbc3d2d2a2ca15b4513e471174d53e4c5fc202f3ae69865140dfda2e | true |
turing-rk1-arm64 | 2026-10-14.1 | 2b5904b9cae9665a4c73414ca289d50b6940cc8a00b388c22b4c22335d1ff95f | 2026-10-26T20:51:38.219Z | v24.22.1 | arm64 | permanent-failure | Stop wasting retries on a permanently failing tool | Five attempts → two attempts. The tool remains broken; the client stops earlier. | after | tool-error | official-sdk-reference-client | 2,000 | 2 | tool_error_only | tool_error | 2 | [
9,
2
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Injected temporary... | d08501886fa2fe3eb0d29916f39e6fbc6b1fb418b29cca0dfd2f5a71ede6e81a | true |
turing-rk1-arm64 | 2026-10-14.1 | 2b5904b9cae9665a4c73414ca289d50b6940cc8a00b388c22b4c22335d1ff95f | 2026-10-26T20:51:38.219Z | v24.22.1 | arm64 | deadline-fit | Fit the deadline to the declared tool latency | 100 ms deadline → 2000 ms deadline for a tool with a 400 ms injected delay. | before | delay | official-sdk-reference-client | 100 | 1 | tool_error_only | timeout | 1 | [
103
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Error: reference c... | ca8d0b945d71b0c1814ae234d73f877fdb46b3ebc3a440e95d6c2114ecafbe2a | true |
turing-rk1-arm64 | 2026-10-14.1 | 2b5904b9cae9665a4c73414ca289d50b6940cc8a00b388c22b4c22335d1ff95f | 2026-10-26T20:51:38.219Z | v24.22.1 | arm64 | deadline-fit | Fit the deadline to the declared tool latency | 100 ms deadline → 2000 ms deadline for a tool with a 400 ms injected delay. | after | delay | official-sdk-reference-client | 2,000 | 1 | tool_error_only | success | 1 | [
408
] | 0 | [] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Probe response: pr... | 45543eb60cd9bf030322aa827426f549c995b208427aa6168c0bc8c37390cbb9 | true |
turing-rk1-arm64 | 2026-10-14.1 | 2b5904b9cae9665a4c73414ca289d50b6940cc8a00b388c22b4c22335d1ff95f | 2026-10-26T20:51:38.219Z | v24.22.1 | arm64 | malformed-control | Keep protocol failures terminal | One-attempt budget → five-attempt budget; protocol errors must still stop after one call. | before | malformed | official-sdk-reference-client | 2,000 | 1 | tool_error_only | protocol_error | 1 | [
21
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Error: reference c... | b01d8e246ec581b6d332074ed83420c64eddb36eaff010f3f056bf4dc52fec9d | true |
turing-rk1-arm64 | 2026-10-14.1 | 2b5904b9cae9665a4c73414ca289d50b6940cc8a00b388c22b4c22335d1ff95f | 2026-10-26T20:51:38.219Z | v24.22.1 | arm64 | malformed-control | Keep protocol failures terminal | One-attempt budget → five-attempt budget; protocol errors must still stop after one call. | after | malformed | official-sdk-reference-client | 2,000 | 5 | tool_error_only | protocol_error | 1 | [
10
] | 0 | [
"unhandled_tool_error"
] | [{"role":"user","content":"Synthetic reference-client probe. No language model participated."},{"role":"assistant","content":null,"tool_calls":[{"id":"probe-1","type":"function","function":{"name":"lab_probe","arguments":"{\"message\":\"probe\"}"}}]},{"role":"tool","tool_call_id":"probe-1","content":"Error: reference c... | b01d8e246ec581b6d332074ed83420c64eddb36eaff010f3f056bf4dc52fec9d | true |
MCP Failure Lab traces
32 deterministic runs of the MCP Failure Lab reference client against a deliberately faulty MCP server, on four platforms (local container, AWS Lambda, Google Cloud Run job, and a Turing RK1 arm64 module in the Cloud Lab Works lab).
Each row is one experiment under one retry/deadline policy (before or after a policy change), with the
outcome, per-attempt timings, the OpenAI-message-shaped trace and the findings from Agent Trace Clinic.
Synthetic. No language model participated. The "assistant" turns come from a scripted reference client. Timings are observed elapsed milliseconds on free-tier infrastructure, not benchmarks, and not a ranking of clouds.
Experiments
| experiment | what changes |
|---|---|
| transient-recovery | 1 attempt → 3 attempts; the tool fails twice, then succeeds (tool_error → success) |
| permanent-failure | 5 attempts → 2 attempts; the tool stays broken, the client stops earlier |
| deadline-fit | 100 ms → 2000 ms deadline for a tool with 400 ms injected delay (timeout → success) |
| malformed-control | 1 → 5 attempt budget; a malformed result is a protocol error and must still stop after one call |
Fields
platform, experiment, variant (before/after), scenario, timeout_ms, max_attempts, retry,
outcome, attempts, elapsed_ms (list), transport_errors, findings (Clinic finding ids),
messages (JSON string, OpenAI chat format with tool_calls), trace_sha256, source_sha256
(hash of the lab source that produced the run), fixture_version, generated_at.
Use
from datasets import load_dataset
ds = load_dataset("cloudlabworks/mcp-failure-lab-traces", split="train")
print(ds.filter(lambda r: r["outcome"] != "success")[0])
Intended uses: regression fixtures for MCP clients, examples of tool-failure traces for agent evals, and a seed for tool-call reliability fine-tuning data.
Provenance
The Turing RK1 (Rockchip RK3588, 8 cores, 32 GB, Ubuntu 24.04, Node 24 arm64, run natively, not in a container) rows come from repo
revision fe185bd, so their source_sha256 differs from the cloud rows; all four experiments reach the same outcomes.
Generated by node runner.mjs (fixture 2026-10-14.1); flattened by
scripts/build-hf-dataset.mjs. data/runs.jsonl sha256 602b5488fae1f5b67145458d84822ae068baee77ffc6bdae85faef52b988088c.
Live replay: agent-reliability-lab Space.
Built by Cloud Lab Works.
- Downloads last month
- 20