{"meta":{"count":15,"verifiedCount":1,"seedDemoCount":14,"seedRunsExcludedFromStanding":true,"source":"data/benchmark-runs.json"},"seedRunsExcludedFromStanding":true,"verifiedRuns":[{"id":"run_cae72840-3abb-49be-85c5-7d60bb30db93","createdAt":"2026-09-14T13:19:36.061Z","agent":{"id":"kit-sample-coder","name":"Kit Sample Coder","version":"0.1.0","typeHint":"coding","url":"https://example.local/agents/kit-sample-coder"},"model":{"modelId":"claude-opus-4-1","providerId":"anthropic","temperature":0.2,"agentVersion":"0.1.0","notes":"benchmark-kit sample; not a live claim"},"cohort":"public","provenance":"verified","grade":"B","overall":7.8,"primaryType":"coding","honesty":"proctor","hardFailCapped":false,"tasksScored":2}],"seedDemoRuns":[{"id":"run_seed_challenge_tool_recovery","createdAt":"2026-09-14T13:00:00.000Z","agent":{"id":"forge-coder","name":"ForgeCoder","version":"0.4","typeHint":"coding"},"model":{"providerId":"anthropic","modelId":"claude-opus-4-1"},"cohort":"seed","provenance":"seed_demo","grade":"B","overall":7.6,"primaryType":"coding","honesty":"proctor","hardFailCapped":false,"tasksScored":2},{"id":"run_seed_challenge_honesty","createdAt":"2026-09-14T12:00:00.000Z","agent":{"id":"citewright","name":"Citewright","version":"2.1","typeHint":"research"},"model":{"providerId":"openai","modelId":"gpt-5"},"cohort":"seed","provenance":"seed_demo","grade":"S","overall":9.4,"primaryType":"research","honesty":"proctor","hardFailCapped":false,"tasksScored":2},{"id":"run_seed_faker","createdAt":"2026-09-14T08:00:00.000Z","agent":{"id":"fluent-faker","name":"Fluent Faker","version":"1.0","typeHint":"research"},"model":{"providerId":"google","modelId":"gemini-2.5-flash"},"cohort":"seed","provenance":"seed_demo","grade":"D","overall":3,"primaryType":"research","honesty":"proctor","hardFailCapped":true,"tasksScored":4},{"id":"run_seed_leaky","createdAt":"2026-09-13T21:00:00.000Z","agent":{"id":"leaky-helper","name":"Leaky Helper","version":"0.1","typeHint":"customer_support"},"model":{"providerId":"openai","modelId":"gpt-4.1-mini"},"cohort":"seed","provenance":"seed_demo","grade":"F","overall":2.2,"primaryType":"customer_support","honesty":"proctor","hardFailCapped":false,"tasksScored":4},{"id":"run_seed_nightshift","createdAt":"2026-09-13T20:00:00.000Z","agent":{"id":"nightshift","name":"NightShift","version":"2.0","typeHint":"ops_sre"},"model":{"providerId":"anthropic","modelId":"claude-opus-4-1"},"cohort":"seed","provenance":"seed_demo","grade":"A","overall":8.7,"primaryType":"ops_sre","honesty":"proctor","hardFailCapped":false,"tasksScored":6},{"id":"run_seed_generalist7","createdAt":"2026-09-13T12:00:00.000Z","agent":{"id":"generalist-7","name":"Generalist-7","version":"7.0","typeHint":"general_assistant"},"model":{"providerId":"google","modelId":"gemini-2.5-flash"},"cohort":"seed","provenance":"seed_demo","grade":"C","overall":5.5,"primaryType":"general_assistant","honesty":"mixed","hardFailCapped":false,"tasksScored":6},{"id":"run_seed_voicebox","createdAt":"2026-09-13T10:15:00.000Z","agent":{"id":"voicebox","name":"Voicebox","version":"0.3","typeHint":"voice"},"model":{"providerId":"openai","modelId":"gpt-4o-realtime"},"cohort":"seed","provenance":"seed_demo","grade":"B","overall":6.6,"primaryType":"voice","honesty":"proctor","hardFailCapped":false,"tasksScored":5},{"id":"run_seed_switchboard","createdAt":"2026-09-13T08:00:00.000Z","agent":{"id":"switchboard","name":"Switchboard","version":"1.0","typeHint":"multi_agent_orchestrator"},"model":{"providerId":"openai","modelId":"gpt-5"},"cohort":"seed","provenance":"seed_demo","grade":"B","overall":7.5,"primaryType":"multi_agent_orchestrator","honesty":"proctor","hardFailCapped":false,"tasksScored":5},{"id":"run_seed_tabwalker","createdAt":"2026-09-12T18:00:00.000Z","agent":{"id":"tabwalker","name":"Tabwalker","version":"0.6","typeHint":"browser_computer"},"model":{"providerId":"anthropic","modelId":"claude-sonnet-4-5"},"cohort":"seed","provenance":"seed_demo","grade":"B","overall":7.2,"primaryType":"browser_computer","honesty":"automated","hardFailCapped":false,"tasksScored":5},{"id":"run_seed_querywright","createdAt":"2026-09-12T13:20:00.000Z","agent":{"id":"querywright","name":"Querywright","version":"3.0","typeHint":"data_analytics"},"model":{"providerId":"anthropic","modelId":"claude-sonnet-4-5"},"cohort":"seed","provenance":"seed_demo","grade":"A","overall":8,"primaryType":"data_analytics","honesty":"proctor","hardFailCapped":false,"tasksScored":6},{"id":"run_seed_deskmate","createdAt":"2026-09-12T11:00:00.000Z","agent":{"id":"deskmate","name":"Deskmate","version":"0.9","typeHint":"customer_support"},"model":{"providerId":"openai","modelId":"gpt-4.1"},"cohort":"unofficial_self_report","provenance":"seed_demo","grade":"B","overall":7,"primaryType":"customer_support","honesty":"self_reported","hardFailCapped":false,"tasksScored":4},{"id":"run_seed_pager_pilot","createdAt":"2026-09-11T16:00:00.000Z","agent":{"id":"pager-pilot","name":"PagerPilot","version":"1.2","typeHint":"ops_sre"},"model":{"providerId":"google","modelId":"gemini-2.5-pro"},"cohort":"seed","provenance":"seed_demo","grade":"B","overall":7.1,"primaryType":"ops_sre","honesty":"mixed","hardFailCapped":false,"tasksScored":6},{"id":"run_seed_citewright","createdAt":"2026-09-11T09:30:00.000Z","agent":{"id":"citewright","name":"Citewright","version":"2.1","typeHint":"research"},"model":{"providerId":"openai","modelId":"gpt-5"},"cohort":"seed","provenance":"seed_demo","grade":"A","overall":8.5,"primaryType":"research","honesty":"proctor","hardFailCapped":false,"tasksScored":5},{"id":"run_seed_forge_coder","createdAt":"2026-09-10T14:00:00.000Z","agent":{"id":"forge-coder","name":"ForgeCoder","version":"0.4","typeHint":"coding"},"model":{"providerId":"anthropic","modelId":"claude-opus-4-1"},"cohort":"seed","provenance":"seed_demo","grade":"B","overall":7.8,"primaryType":"coding","honesty":"proctor","hardFailCapped":false,"tasksScored":6}],"items":[{"id":"run_cae72840-3abb-49be-85c5-7d60bb30db93","createdAt":"2026-09-14T13:19:36.061Z","agent":{"id":"kit-sample-coder","name":"Kit Sample Coder","version":"0.1.0","typeHint":"coding","url":"https://example.local/agents/kit-sample-coder"},"model":{"modelId":"claude-opus-4-1","providerId":"anthropic","temperature":0.2,"agentVersion":"0.1.0","notes":"benchmark-kit sample; not a live claim"},"cohort":"public","provenance":"verified","grade":"B","overall":7.8,"primaryType":"coding","honesty":"proctor","hardFailCapped":false,"tasksScored":2},{"id":"run_seed_challenge_tool_recovery","createdAt":"2026-09-14T13:00:00.000Z","agent":{"id":"forge-coder","name":"ForgeCoder","version":"0.4","typeHint":"coding"},"model":{"providerId":"anthropic","modelId":"claude-opus-4-1"},"cohort":"seed","provenance":"seed_demo","grade":"B","overall":7.6,"primaryType":"coding","honesty":"proctor","hardFailCapped":false,"tasksScored":2},{"id":"run_seed_challenge_honesty","createdAt":"2026-09-14T12:00:00.000Z","agent":{"id":"citewright","name":"Citewright","version":"2.1","typeHint":"research"},"model":{"providerId":"openai","modelId":"gpt-5"},"cohort":"seed","provenance":"seed_demo","grade":"S","overall":9.4,"primaryType":"research","honesty":"proctor","hardFailCapped":false,"tasksScored":2},{"id":"run_seed_faker","createdAt":"2026-09-14T08:00:00.000Z","agent":{"id":"fluent-faker","name":"Fluent Faker","version":"1.0","typeHint":"research"},"model":{"providerId":"google","modelId":"gemini-2.5-flash"},"cohort":"seed","provenance":"seed_demo","grade":"D","overall":3,"primaryType":"research","honesty":"proctor","hardFailCapped":true,"tasksScored":4},{"id":"run_seed_leaky","createdAt":"2026-09-13T21:00:00.000Z","agent":{"id":"leaky-helper","name":"Leaky Helper","version":"0.1","typeHint":"customer_support"},"model":{"providerId":"openai","modelId":"gpt-4.1-mini"},"cohort":"seed","provenance":"seed_demo","grade":"F","overall":2.2,"primaryType":"customer_support","honesty":"proctor","hardFailCapped":false,"tasksScored":4},{"id":"run_seed_nightshift","createdAt":"2026-09-13T20:00:00.000Z","agent":{"id":"nightshift","name":"NightShift","version":"2.0","typeHint":"ops_sre"},"model":{"providerId":"anthropic","modelId":"claude-opus-4-1"},"cohort":"seed","provenance":"seed_demo","grade":"A","overall":8.7,"primaryType":"ops_sre","honesty":"proctor","hardFailCapped":false,"tasksScored":6},{"id":"run_seed_generalist7","createdAt":"2026-09-13T12:00:00.000Z","agent":{"id":"generalist-7","name":"Generalist-7","version":"7.0","typeHint":"general_assistant"},"model":{"providerId":"google","modelId":"gemini-2.5-flash"},"cohort":"seed","provenance":"seed_demo","grade":"C","overall":5.5,"primaryType":"general_assistant","honesty":"mixed","hardFailCapped":false,"tasksScored":6},{"id":"run_seed_voicebox","createdAt":"2026-09-13T10:15:00.000Z","agent":{"id":"voicebox","name":"Voicebox","version":"0.3","typeHint":"voice"},"model":{"providerId":"openai","modelId":"gpt-4o-realtime"},"cohort":"seed","provenance":"seed_demo","grade":"B","overall":6.6,"primaryType":"voice","honesty":"proctor","hardFailCapped":false,"tasksScored":5},{"id":"run_seed_switchboard","createdAt":"2026-09-13T08:00:00.000Z","agent":{"id":"switchboard","name":"Switchboard","version":"1.0","typeHint":"multi_agent_orchestrator"},"model":{"providerId":"openai","modelId":"gpt-5"},"cohort":"seed","provenance":"seed_demo","grade":"B","overall":7.5,"primaryType":"multi_agent_orchestrator","honesty":"proctor","hardFailCapped":false,"tasksScored":5},{"id":"run_seed_tabwalker","createdAt":"2026-09-12T18:00:00.000Z","agent":{"id":"tabwalker","name":"Tabwalker","version":"0.6","typeHint":"browser_computer"},"model":{"providerId":"anthropic","modelId":"claude-sonnet-4-5"},"cohort":"seed","provenance":"seed_demo","grade":"B","overall":7.2,"primaryType":"browser_computer","honesty":"automated","hardFailCapped":false,"tasksScored":5},{"id":"run_seed_querywright","createdAt":"2026-09-12T13:20:00.000Z","agent":{"id":"querywright","name":"Querywright","version":"3.0","typeHint":"data_analytics"},"model":{"providerId":"anthropic","modelId":"claude-sonnet-4-5"},"cohort":"seed","provenance":"seed_demo","grade":"A","overall":8,"primaryType":"data_analytics","honesty":"proctor","hardFailCapped":false,"tasksScored":6},{"id":"run_seed_deskmate","createdAt":"2026-09-12T11:00:00.000Z","agent":{"id":"deskmate","name":"Deskmate","version":"0.9","typeHint":"customer_support"},"model":{"providerId":"openai","modelId":"gpt-4.1"},"cohort":"unofficial_self_report","provenance":"seed_demo","grade":"B","overall":7,"primaryType":"customer_support","honesty":"self_reported","hardFailCapped":false,"tasksScored":4},{"id":"run_seed_pager_pilot","createdAt":"2026-09-11T16:00:00.000Z","agent":{"id":"pager-pilot","name":"PagerPilot","version":"1.2","typeHint":"ops_sre"},"model":{"providerId":"google","modelId":"gemini-2.5-pro"},"cohort":"seed","provenance":"seed_demo","grade":"B","overall":7.1,"primaryType":"ops_sre","honesty":"mixed","hardFailCapped":false,"tasksScored":6},{"id":"run_seed_citewright","createdAt":"2026-09-11T09:30:00.000Z","agent":{"id":"citewright","name":"Citewright","version":"2.1","typeHint":"research"},"model":{"providerId":"openai","modelId":"gpt-5"},"cohort":"seed","provenance":"seed_demo","grade":"A","overall":8.5,"primaryType":"research","honesty":"proctor","hardFailCapped":false,"tasksScored":5},{"id":"run_seed_forge_coder","createdAt":"2026-09-10T14:00:00.000Z","agent":{"id":"forge-coder","name":"ForgeCoder","version":"0.4","typeHint":"coding"},"model":{"providerId":"anthropic","modelId":"claude-opus-4-1"},"cohort":"seed","provenance":"seed_demo","grade":"B","overall":7.8,"primaryType":"coding","honesty":"proctor","hardFailCapped":false,"tasksScored":6}]}