Skynet Breakage Atlas
Adversarial multi-model failure rates — subscription controls vs Skynet fleet.
trials: 126
generated: 2026-07-15 14:41 UTC
P0
By model
| Model | Role | N | Flag rate | Empty | Error | p50 ms |
| Claude Max (sonnet) | control | 18 | 0.056 | 0.056 | 0.0 | 5526 |
| Codex gpt-5.4-mini | control | 18 | 0.167 | 0.0 | 0.167 | 7186 |
| Grok SuperGrok Heavy | control | 18 | 0.0 | 0.0 | 0.0 | 2703 |
| Skynet deepseek-v4-pro | fleet | 18 | 0.389 | 0.389 | 0.0 | 1912 |
| Skynet glm-5.2 | fleet | 18 | 1.0 | 1.0 | 0.0 | 8074 |
| Skynet kimi-k2.7-code | fleet | 18 | 0.167 | 0.167 | 0.0 | 1800 |
| Skynet minimax-m3 | fleet | 18 | 0.333 | 0.333 | 0.0 | 1379 |
By cell (model × family)
| Model | Family | N | Flag rate | Empty | Hard err | p50 ms |
| control-claude | cascade_seed | 3 | 0.0 | 0 | 0 | 23172 |
| control-claude | context_boundary | 3 | 0.0 | 0 | 0 | 5093 |
| control-claude | cost_threshold | 3 | 0.0 | 0 | 0 | 5526 |
| control-claude | empty_probe | 3 | 0.333 | 1 | 0 | 3387 |
| control-claude | format_json | 3 | 0.0 | 0 | 0 | 3684 |
| control-claude | multi_intent | 3 | 0.0 | 0 | 0 | 6418 |
| control-codex | cascade_seed | 3 | 1.0 | 0 | 3 | 120031 |
| control-codex | context_boundary | 3 | 0.0 | 0 | 0 | 4232 |
| control-codex | cost_threshold | 3 | 0.0 | 0 | 0 | 5321 |
| control-codex | empty_probe | 3 | 0.0 | 0 | 0 | 4305 |
| control-codex | format_json | 3 | 0.0 | 0 | 0 | 6881 |
| control-codex | multi_intent | 3 | 0.0 | 0 | 0 | 10241 |
| control-grok | cascade_seed | 3 | 0.0 | 0 | 0 | 12104 |
| control-grok | context_boundary | 3 | 0.0 | 0 | 0 | 3296 |
| control-grok | cost_threshold | 3 | 0.0 | 0 | 0 | 2703 |
| control-grok | empty_probe | 3 | 0.0 | 0 | 0 | 1663 |
| control-grok | format_json | 3 | 0.0 | 0 | 0 | 1824 |
| control-grok | multi_intent | 3 | 0.0 | 0 | 0 | 2628 |
| skynet-deepseek | cascade_seed | 3 | 1.0 | 3 | 0 | 2517 |
| skynet-deepseek | context_boundary | 3 | 0.0 | 0 | 0 | 1272 |
| skynet-deepseek | cost_threshold | 3 | 0.0 | 0 | 0 | 1302 |
| skynet-deepseek | empty_probe | 3 | 1.0 | 3 | 0 | 1483 |
| skynet-deepseek | format_json | 3 | 0.333 | 1 | 0 | 2430 |
| skynet-deepseek | multi_intent | 3 | 0.0 | 0 | 0 | 2661 |
| skynet-glm | cascade_seed | 3 | 1.0 | 3 | 0 | 9461 |
| skynet-glm | context_boundary | 3 | 1.0 | 3 | 0 | 4495 |
| skynet-glm | cost_threshold | 3 | 1.0 | 3 | 0 | 4101 |
| skynet-glm | empty_probe | 3 | 1.0 | 3 | 0 | 6547 |
| skynet-glm | format_json | 3 | 1.0 | 3 | 0 | 8172 |
| skynet-glm | multi_intent | 3 | 1.0 | 3 | 0 | 10625 |
| skynet-kimi | cascade_seed | 3 | 1.0 | 3 | 0 | 3116 |
| skynet-kimi | context_boundary | 3 | 0.0 | 0 | 0 | 1313 |
| skynet-kimi | cost_threshold | 3 | 0.0 | 0 | 0 | 1315 |
| skynet-kimi | empty_probe | 3 | 0.0 | 0 | 0 | 1648 |
| skynet-kimi | format_json | 3 | 0.0 | 0 | 0 | 1828 |
| skynet-kimi | multi_intent | 3 | 0.0 | 0 | 0 | 4076 |
| skynet-minimax | cascade_seed | 3 | 1.0 | 3 | 0 | 6269 |
| skynet-minimax | context_boundary | 3 | 0.333 | 1 | 0 | 1001 |
| skynet-minimax | cost_threshold | 3 | 0.0 | 0 | 0 | 1237 |
| skynet-minimax | empty_probe | 3 | 0.0 | 0 | 0 | 1375 |
| skynet-minimax | format_json | 3 | 0.0 | 0 | 0 | 1098 |
| skynet-minimax | multi_intent | 3 | 0.667 | 2 | 0 | 6155 |