{
  "benchmark": {
    "goal_prompt": "Build the requested web application autonomously in this clean task workspace.\nRead the complete product requirements from prd/mvp.txt and do not deviate from them.\nIf an assets directory exists, use its files where the requirements call for them. Do not modify the assets directory.\nCreate the complete runnable application under output/app and nowhere else.\nUse only files present in this workspace; do not inspect paths outside it.\nDo not request or depend on API keys or other external secrets.\n\nViBench runtime contract:\n1. Create output/app/setup-environment.sh. It runs in a Debian container with Python 3.12 and Node.js 22 installed. It must install all dependencies, create database schema or seed data when the application needs them, be safe to run more than once, and exit without starting the server.\n2. Create output/app/start-server.sh. It must change to its own directory with cd \"$(dirname \"$0\")\", start the application in the foreground without redirecting stdin, stdout, or stderr, listen on the APPLICATION_PORT environment variable, and accept requests for any host name.\n3. The runtime provides POSTGRES_DATABASE_URL and APPLICATION_PORT. The application must work in a fresh environment with an empty database after setup-environment.sh, then start-server.sh.\n4. Make both scripts executable. Do not start a long-running server as your final action.\n\nAdd stable, descriptive data-testid attributes to interactive controls and meaningful display elements. Build a clean, modern interface in addition to meeting the functional requirements.\nVerify your work using only checks you create or standard project tooling.\nFinish with a concise summary of the implementation and verification performed.\n",
    "goal_prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840",
    "stage": "mvp",
    "suite": "standard",
    "upstream_repository": "https://github.com/ViBench/vibench-public",
    "upstream_revision": "5baa6892bad7dcf6ec0bee4f44e4e46445271f19"
  },
  "cells": [
    {
      "cell_id": "barber--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "barber--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 397.143912,
          "finished_at": "2026-08-28T10:59:03.686177Z",
          "started_at": "2026-08-28T10:52:26.551361Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "e138b3afee082067d2682b09d870729d923f23ac3851b1c21431761bc7669364",
          "output_sha256": "3296b460ddf7d2e0bb622c1a3fbd4f634da760968b1c12c4ca21634ea2bf5271",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": null,
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 273408,
          "input_tokens": 358265,
          "output_tokens": 18067,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "barber",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "barber--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "barber",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "barber--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "barber",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "barber--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "barber",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "book_journey--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "book_journey--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 383.208075,
          "finished_at": "2026-08-28T11:17:53.529449Z",
          "started_at": "2026-08-28T11:11:30.330144Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "d0fc65e5f946aa00bc42b8ebc947e0cb61747ec6a6cf9d602bcc6adfbafc3279",
          "output_sha256": "61c078b6bc5a87606a80fa855b2096ebc8674cc0fc2e782867084fedad07c6ee",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": null,
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 202496,
          "input_tokens": 240915,
          "output_tokens": 18554,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "book_journey",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "book_journey--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "book_journey",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "book_journey--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "book_journey",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "book_journey--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "book_journey",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "canary--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "canary--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": "vibench-native-canary-http-v1",
          "full_points": 10,
          "score": 10,
          "status": "passed",
          "test_plans_evaluated": 1
        },
        "execution": {
          "duration_seconds": 133.262931,
          "finished_at": "2026-08-28T10:19:58.601390Z",
          "started_at": "2026-08-28T10:17:45.340777Z"
        },
        "hashes": {
          "evaluation_sha256": "59555fad264283c4780c4aedc7adce8da2fd7fe49d693cf0da3c9ff6bc24c8cb",
          "input_sha256": "ec0d31019fb8a477f9e28ba6227f0970efd0f2eb9347693fbcdf4035e2deec5f",
          "output_sha256": "b5cd27149c75c9fadcbab4cf4d6e02cd1c19e148de928bd6ee2135a643a6fcce",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": null,
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 113920,
          "input_tokens": 139697,
          "output_tokens": 5925,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "engineering_calibration_only"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "canary",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "canary--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "canary",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "canary--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "canary",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "canary--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "canary",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "collabrative_kaban--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": {
          "dispatch_attempts": 2,
          "voided_attempts": [
            {
              "attempt_number": 1,
              "classification": "voided_pre_invocation_infrastructure_failure",
              "no_provider_invocation": true,
              "reason_code": "harness_import_failed_before_runner_entry",
              "reason_summary": "The harness adapter import failed before the native runner started."
            }
          ],
          "voided_pre_invocation_attempts": 1
        },
        "cell_id": "collabrative_kaban--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 727.826429,
          "finished_at": "2026-08-28T11:39:25.481226Z",
          "started_at": "2026-08-28T11:27:19.636624Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "296dba2e78533dee1df0a0f1326cabfa08dab63aa158828bb40b14cafa58157c",
          "output_sha256": "f2dd45f0e4af7fede35d0b7c509a9f2163c739a63b405af7c1dbc26182a3ca36",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": {
          "authorization_hash_bound": true,
          "classification": "verified_os_clock_adjustment",
          "receipt_present": false
        },
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 724736,
          "input_tokens": 798585,
          "output_tokens": 33162,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "collabrative_kaban",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "collabrative_kaban--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "collabrative_kaban",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "collabrative_kaban--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "collabrative_kaban",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "collabrative_kaban--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "collabrative_kaban",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "creative_community--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "creative_community--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 637.082785,
          "finished_at": "2026-08-28T11:55:18.613879Z",
          "started_at": "2026-08-28T11:44:39.529130Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "919176f3f2c5a004e1c4f6bbd7ba286c4fef9cbf756ec9bd68b950d7809cf016",
          "output_sha256": "d437f07f04f3130ac611f21116c07f2fe122e22f9b1b6887190f9a7a168aa8c9",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": {
          "authorization_hash_bound": true,
          "classification": "verified_os_clock_adjustment",
          "receipt_present": false
        },
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 452096,
          "input_tokens": 504222,
          "output_tokens": 31269,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "creative_community",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "creative_community--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "creative_community",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "creative_community--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "creative_community",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "creative_community--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "creative_community",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "energy_audit--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "energy_audit",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "energy_audit--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "energy_audit",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "energy_audit--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "energy_audit",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "energy_audit--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "energy_audit",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "family_friendly_venue--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "family_friendly_venue",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "family_friendly_venue--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "family_friendly_venue",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "family_friendly_venue--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "family_friendly_venue",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "family_friendly_venue--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "family_friendly_venue",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "family_social--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "family_social",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "family_social--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "family_social",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "family_social--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "family_social",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "family_social--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "family_social",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "fleet_management--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "fleet_management--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 373.553038,
          "finished_at": "2026-08-28T10:35:44.218449Z",
          "started_at": "2026-08-28T10:29:30.672054Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "11a4b6ffccdde57776c4f57066f656a900fd3bb781033817e96a13af648c8d5b",
          "output_sha256": "fea5acdea258b4b473d859453c1001e43caca3524596d09bd34b482969b03402",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": null,
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 185344,
          "input_tokens": 215388,
          "output_tokens": 19499,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "fleet_management",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "fleet_management--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "fleet_management",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "fleet_management--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "fleet_management",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "fleet_management--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "fleet_management",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "furniture_freight--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "furniture_freight--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 476.714407,
          "finished_at": "2026-08-28T14:15:33.205870Z",
          "started_at": "2026-08-28T14:07:36.499950Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "b3a9227aa7afa0f3b9432d610172df2ccceabbea5c91933ebf9e97d79d03afc6",
          "output_sha256": "99a47b3122f4a09615b73cb6b20c06d6e19bcdba031747baa2247d44251f1385",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": {
          "authorization_hash_bound": false,
          "classification": "paired_clock_ledger_verified",
          "receipt_present": true
        },
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 226048,
          "input_tokens": 311834,
          "output_tokens": 24289,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "furniture_freight",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "furniture_freight--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "furniture_freight",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "furniture_freight--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "furniture_freight",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "furniture_freight--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "furniture_freight",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "hvac--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "hvac",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "hvac--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "hvac",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "hvac--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "hvac",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "hvac--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "hvac",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "language_learning--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "language_learning--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 472.920536,
          "finished_at": "2026-08-28T12:17:39.593229Z",
          "started_at": "2026-08-28T12:09:46.674854Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "f603fcc724569025f3b235306dc8d42098999e33d15da5f1083d01ca72d19d81",
          "output_sha256": "f69126b8198cd78b06b4e4a1c865d79caff72ddbcc8475967d867ad24834737a",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": {
          "authorization_hash_bound": false,
          "classification": "paired_clock_ledger_verified",
          "receipt_present": true
        },
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 600832,
          "input_tokens": 645014,
          "output_tokens": 22850,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "language_learning",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "language_learning--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "language_learning",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "language_learning--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "language_learning",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "language_learning--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "language_learning",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "logistics--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "logistics",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "logistics--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "logistics",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "logistics--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "logistics",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "logistics--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "logistics",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "mafia--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "mafia--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 462.252648,
          "finished_at": "2026-08-28T13:32:13.898939Z",
          "started_at": "2026-08-28T13:24:31.653192Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "3e0efab5953a28a2894731dbe2efeca2cb109cbc135d4aa5b93d9b85c08873f8",
          "output_sha256": "32086a1f60739341d39abc5af3d3760475110028b4dcc215f3bf878cb7db4550",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": {
          "authorization_hash_bound": false,
          "classification": "paired_clock_ledger_verified",
          "receipt_present": true
        },
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 380160,
          "input_tokens": 446747,
          "output_tokens": 19964,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "mafia",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "mafia--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "mafia",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "mafia--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "mafia",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "mafia--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "mafia",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "market_place--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "market_place--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 586.601445,
          "finished_at": "2026-08-28T14:04:18.717123Z",
          "started_at": "2026-08-28T13:54:32.133933Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "a83b6762d6c20ad64ed94a534ba6b311678f3222bec93cf9e38c41af320ac160",
          "output_sha256": "5903b34e368d546ff208f829c3332344f71234943f39074a85c19e4ee6b1ba80",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": {
          "authorization_hash_bound": false,
          "classification": "paired_clock_ledger_verified",
          "receipt_present": true
        },
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 339968,
          "input_tokens": 378507,
          "output_tokens": 24218,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "market_place",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "market_place--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "market_place",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "market_place--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "market_place",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "market_place--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "market_place",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "monopoly--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "monopoly--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 493.186402,
          "finished_at": "2026-08-28T13:41:41.794203Z",
          "started_at": "2026-08-28T13:33:28.615157Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "b5ae33ec3e552ef6704e3dcb8adf04891268947b94440cf01b24d24a70fd2ba5",
          "output_sha256": "c5b7104c8ad2c1c241f2ca28fef9244cd486c62ea94477ef12581bfdcfb22716",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": {
          "authorization_hash_bound": false,
          "classification": "paired_clock_ledger_verified",
          "receipt_present": true
        },
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 476928,
          "input_tokens": 520859,
          "output_tokens": 24052,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "monopoly",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "monopoly--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "monopoly",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "monopoly--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "monopoly",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "monopoly--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "monopoly",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "notes--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "notes--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 437.977028,
          "finished_at": "2026-08-28T13:53:18.416950Z",
          "started_at": "2026-08-28T13:46:00.446460Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "1437b437695146cedcf1a8cd4a6854d50b0c60b9b55d37d441506384f70c59f1",
          "output_sha256": "ecbf8ea5be423c152bf079b09da3c27373a0df1ca7b3c24bba6dbd6dbd9cf482",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": {
          "authorization_hash_bound": false,
          "classification": "paired_clock_ledger_verified",
          "receipt_present": true
        },
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 291584,
          "input_tokens": 350688,
          "output_tokens": 18257,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "notes",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "notes--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "notes",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "notes--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "notes",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "notes--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "notes",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "online_whiteboard--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "online_whiteboard--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 567.410231,
          "finished_at": "2026-08-28T13:21:47.598480Z",
          "started_at": "2026-08-28T13:12:20.196719Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "63a516153f7210205f4cb2a7ec4a6b49fbdd793569d45f54c776d721db64c101",
          "output_sha256": "4d2ac9fa2906a682d6db9eee263dc8dc0f759ba25c684b0db6a9260a6dbd4ffe",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": {
          "authorization_hash_bound": false,
          "classification": "paired_clock_ledger_verified",
          "receipt_present": true
        },
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 397568,
          "input_tokens": 474328,
          "output_tokens": 20893,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "online_whiteboard",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "online_whiteboard--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "online_whiteboard",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "online_whiteboard--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "online_whiteboard",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "online_whiteboard--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "online_whiteboard",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "pilot_logbook--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "pilot_logbook",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "pilot_logbook--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "pilot_logbook",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "pilot_logbook--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "pilot_logbook",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "pilot_logbook--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "pilot_logbook",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "quiz--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "quiz--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 368.016862,
          "finished_at": "2026-08-28T12:23:47.721409Z",
          "started_at": "2026-08-28T12:17:39.710881Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "a73828422d3cbb67cc6bc4c4f7e68e07f3d36f0d6bcd651aef1f5c3088d373dc",
          "output_sha256": "5d39637166e41615ca30d9f61aeba895848a18db1e7929b27f61a2ddcf36c147",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": {
          "authorization_hash_bound": false,
          "classification": "paired_clock_ledger_verified",
          "receipt_present": true
        },
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 418048,
          "input_tokens": 457609,
          "output_tokens": 18331,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "quiz",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "quiz--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "quiz",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "quiz--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "quiz",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "quiz--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "quiz",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "resume_builder--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "resume_builder--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 507.077521,
          "finished_at": "2026-08-28T10:44:11.365228Z",
          "started_at": "2026-08-28T10:35:44.296821Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "2c7b0dc5b92cdca8ec7f1a49726451396921a423177c31711df5c2dd5e0a6fed",
          "output_sha256": "835de491793cccff0b30255f85da7d5bfa8a14d2816e6d2edf6fa5f0b8ba800a",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": null,
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 352000,
          "input_tokens": 391695,
          "output_tokens": 24365,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "resume_builder",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "resume_builder--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "resume_builder",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "resume_builder--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "resume_builder",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "resume_builder--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "resume_builder",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "slack--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "slack--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 559.709888,
          "finished_at": "2026-08-28T12:08:04.290927Z",
          "started_at": "2026-08-28T11:58:44.594052Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "40c98acca0387fc353a1fba5597d10117161a86e3fddf7d93e3a54f60044aec7",
          "output_sha256": "2a1051659efd78795dad27e125050ec1c93fc75c341d8a8bbbbbd68e5b5e73b0",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": {
          "authorization_hash_bound": false,
          "classification": "paired_clock_ledger_verified",
          "receipt_present": true
        },
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 441600,
          "input_tokens": 488890,
          "output_tokens": 28041,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "slack",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "slack--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "slack",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "slack--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "slack",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "slack--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "slack",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "srm--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "srm",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "srm--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "srm",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "srm--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "srm",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "srm--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "srm",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    },
    {
      "cell_id": "wedding--codex-sol-high-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result": {
        "attempt_history": null,
        "cell_id": "wedding--codex-sol-high-native--r1",
        "evaluation": {
          "evaluator_id": null,
          "full_points": null,
          "score": null,
          "status": "not_run",
          "test_plans_evaluated": 0
        },
        "execution": {
          "duration_seconds": 905.877193,
          "finished_at": "2026-08-28T12:38:53.646818Z",
          "started_at": "2026-08-28T12:23:47.837073Z"
        },
        "hashes": {
          "evaluation_sha256": null,
          "input_sha256": "643035cc35ab3d18dc5b29c41ae395c37b8621200761c8b8d7ce674108a5fc8a",
          "output_sha256": "557c79cd936febf38efca6331d8eb9113a81bcc360b9619efa3269fa4f600165",
          "prompt_sha256": "dafbc1491d68cd5075376535132c84b61876e50fec256a61625d1abd4253b840"
        },
        "plan_outcomes": [],
        "run_status": "succeeded",
        "schema_version": 1,
        "timing_integrity": {
          "authorization_hash_bound": false,
          "classification": "paired_clock_ledger_verified",
          "receipt_present": true
        },
        "treatment": {
          "adapter": "codex",
          "agent_config_id": "codex-sol-high",
          "requested_effort": "high",
          "requested_model": "gpt-5.6-sol",
          "runner_kind": "native_cli",
          "strategy": "native_single_agent"
        },
        "usage": {
          "cached_input_tokens": 1240832,
          "input_tokens": 1320185,
          "output_tokens": 44706,
          "total_tokens": null
        },
        "validity": {
          "reasons": [
            "host_read_containment_unverified",
            "effective_model_identity_unreported",
            "evaluation_not_run"
          ],
          "score_eligible": false
        }
      },
      "result_status": "observed",
      "task_id": "wedding",
      "treatment_id": "codex-sol-high-native"
    },
    {
      "cell_id": "wedding--grok-4.6-xhigh-native--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "wedding",
      "treatment_id": "grok-4.6-xhigh-native"
    },
    {
      "cell_id": "wedding--repoprompt-agent-default--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "wedding",
      "treatment_id": "repoprompt-agent-default"
    },
    {
      "cell_id": "wedding--claude-fable-opus5-high-native-orchestration--r1",
      "plan_status": "planned",
      "repetition": 1,
      "result_status": "missing",
      "task_id": "wedding",
      "treatment_id": "claude-fable-opus5-high-native-orchestration"
    }
  ],
  "experiment": {
    "distinct_test_plans": 74,
    "id": "vibench-native-mvp-96-r1",
    "planned_cells": 96,
    "repetitions": 1,
    "scope": "Full 24-app MVP corpus",
    "title": "ViBench Native Agents · Full MVP Corpus"
  },
  "generated_at": "2026-08-28T14:16:46Z",
  "methodology": {
    "anomaly_policy": "Upstream wedding/test1 declares 55 points, but its scored steps total 43 and one step has no points field. The primary score keeps the plan unchanged. The 73-plan anomaly-excluded value is a non-primary audit diagnostic.",
    "artifact_policy": "The public bundle contains only normalized result fields, hashes, the canonical goal prompt, and links to public upstream PRDs and test plans.",
    "complete_score_policy": "The primary ViBench-compatible treatment score appears only after all 74 plans have resolved outcomes. Partial values are available-case diagnostics.",
    "missing_policy": "Every planned cell remains visible. A missing result means no publishable result record was supplied, not a score of zero.",
    "score_policy": "Only the fixed ViBench evaluator produces comparable corpus scores. The main score follows the pinned ViBench analyzer and uses all 74 plans. It first takes the mean plan score for each app. It then takes the mean score of the 24 apps. The wedding/test1 plan has an upstream points mismatch, so the site also shows a separate 73-plan audit diagnostic. That diagnostic is not the main score. A candidate build or seeding failure is zero. An evaluator error or a missing run is not zero. Calibration scores are engineering results."
  },
  "schema_version": 1,
  "score_universe": [
    {
      "declared_full_points": 22,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "003204e1c71e6879ff4dc940e67d38025d6f450d0c178e948d3e22aba61aa353",
      "step_points_sum": 22,
      "task_id": "barber"
    },
    {
      "declared_full_points": 30,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "06395372fa0f16d7a3f5c33ff7c86c3112326758b32972de76356977f53f45d7",
      "step_points_sum": 30,
      "task_id": "barber"
    },
    {
      "declared_full_points": 34,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "e39cb7e58c0578f95e52b3e4aafd932ecdf60a4f21163e5c224698e29b29c57c",
      "step_points_sum": 34,
      "task_id": "barber"
    },
    {
      "declared_full_points": 27,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "dd82119148a7b7b1105e5ba10dc3a8b0c853b88c7edf57a68a65ee1e9df179b8",
      "step_points_sum": 27,
      "task_id": "book_journey"
    },
    {
      "declared_full_points": 27,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "6d5cdd8fb1ab1c7d699b52fab547bcd0ba01f4020ec7d961d9136566444dd800",
      "step_points_sum": 27,
      "task_id": "book_journey"
    },
    {
      "declared_full_points": 42,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "480d1fcc9f05bc27afacdeaf9baf492450e9aa21b2b5a4eec4cbb7047a65fd64",
      "step_points_sum": 42,
      "task_id": "book_journey"
    },
    {
      "declared_full_points": 10,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "d4c9216b30805924797ad31a7d804ec26e27573adc25532525fde00cb929f560",
      "step_points_sum": 10,
      "task_id": "canary"
    },
    {
      "declared_full_points": 30,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "5481dd6152a6b25b753762873bea3be9a0223d061a8984e1b38e08ccd1cab432",
      "step_points_sum": 30,
      "task_id": "collabrative_kaban"
    },
    {
      "declared_full_points": 54,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "54ba13e9597ca0b124504382ab7c3e13a61db2cfc4d37ca1ceb339034f5609d9",
      "step_points_sum": 54,
      "task_id": "collabrative_kaban"
    },
    {
      "declared_full_points": 62,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "5b55ae62e1adda1159d40c4fae9f977e4ecf3d36298079fcb2a72bd5f5b12109",
      "step_points_sum": 62,
      "task_id": "collabrative_kaban"
    },
    {
      "declared_full_points": 40,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "480e5108193d6e5f9040dd7b7c3941dcef3078adfb96b2b4dee3f9dd22e8a17a",
      "step_points_sum": 40,
      "task_id": "creative_community"
    },
    {
      "declared_full_points": 64,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "29fb38d3ea2915e39b0ae24b1726711c37d7c3adedbd2b9fd48c3290e860e7f5",
      "step_points_sum": 64,
      "task_id": "creative_community"
    },
    {
      "declared_full_points": 60,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "dd54d7bfdb6809e0979439f5262ba49febd99c14f6a4b2c0c1826fa74258b623",
      "step_points_sum": 60,
      "task_id": "creative_community"
    },
    {
      "declared_full_points": 70,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "a8b40336fea74c5504cf9325570aede62dc409fef5e9a379fe420be8be546df5",
      "step_points_sum": 70,
      "task_id": "energy_audit"
    },
    {
      "declared_full_points": 68,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "d82eee7f16d1747d87cfdec12490824adcfa9e91215edaec4a1a01f2efff389f",
      "step_points_sum": 68,
      "task_id": "energy_audit"
    },
    {
      "declared_full_points": 50,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "43b566f24f600739f017cfd909ebbd207aade8130c866fee0cee5720d3afb665",
      "step_points_sum": 50,
      "task_id": "energy_audit"
    },
    {
      "declared_full_points": 38,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "86cd00d235071bb06dae3d8ee332281dd2cf1a4df818793a1b5b14c67fe94550",
      "step_points_sum": 38,
      "task_id": "family_friendly_venue"
    },
    {
      "declared_full_points": 44,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "1501413dae3241ef80e7a533a98753fee19546a9c19daa98b17fab29670f94be",
      "step_points_sum": 44,
      "task_id": "family_friendly_venue"
    },
    {
      "declared_full_points": 48,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "bcaba778c93b701523917d94fa7763eadb0c3a5fe8bea3197d1aace7fdc373b6",
      "step_points_sum": 48,
      "task_id": "family_friendly_venue"
    },
    {
      "declared_full_points": 42,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "9ec05f0985a3e02824fe77283f361c34bfc35c4ffe6e43802a13c8198b2377c3",
      "step_points_sum": 42,
      "task_id": "family_social"
    },
    {
      "declared_full_points": 70,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "7fbd00444b0174d355d76c1027aecd0e84caf5152029dfcdef38389e31ab8167",
      "step_points_sum": 70,
      "task_id": "family_social"
    },
    {
      "declared_full_points": 70,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "47f1dd845b20bf12eeaf845a0767c810c45e4d135f0c48cbf872c5be01556d59",
      "step_points_sum": 70,
      "task_id": "family_social"
    },
    {
      "declared_full_points": 34,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "c15160fdb9fba23b3718d854f1c655448a5044e4f2fb42389df820fd1ff87660",
      "step_points_sum": 34,
      "task_id": "fleet_management"
    },
    {
      "declared_full_points": 44,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "932ae1ccba455a7d2102ce1ed928e245f56e7f3f908943f06fce409dd2283166",
      "step_points_sum": 44,
      "task_id": "fleet_management"
    },
    {
      "declared_full_points": 36,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "fbb48b545a13929f2dd0a22b9133e0563fde8e84b96fb42fb41bae59f6128f96",
      "step_points_sum": 36,
      "task_id": "fleet_management"
    },
    {
      "declared_full_points": 20,
      "plan_id": "test4",
      "quality_flags": [],
      "sha256": "15f87e5cd384b920bd70b83c3e2501bc3d8718a2b071b1cee92edd5d1ab947e3",
      "step_points_sum": 20,
      "task_id": "fleet_management"
    },
    {
      "declared_full_points": 66,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "1db9c81dd113cb71fe2906c80e0b0634c4c02e224bb966ad918dc70583127ee5",
      "step_points_sum": 66,
      "task_id": "furniture_freight"
    },
    {
      "declared_full_points": 44,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "b27ebbe23e93d5153dfd1c64c1a660be76f367024ed9794c05e36924a1d5715e",
      "step_points_sum": 44,
      "task_id": "furniture_freight"
    },
    {
      "declared_full_points": 34,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "0d7edb65fd160482697bb3079137cb8df6d965170f29795cbefb7ffe4683b66c",
      "step_points_sum": 34,
      "task_id": "furniture_freight"
    },
    {
      "declared_full_points": 97,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "57818ef3f0fdf0527daff8ddd629daa016c96d5419ecebee4b70b64987583d24",
      "step_points_sum": 97,
      "task_id": "hvac"
    },
    {
      "declared_full_points": 80,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "bffc068c4f4cf7e8a4a7f9b8bb087d8dd1c0fc92ffaa11e8d2e3b0e2f4527f55",
      "step_points_sum": 80,
      "task_id": "hvac"
    },
    {
      "declared_full_points": 101,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "19b1097b69364bbf5e7edf2b3cebc354cf6af8ff8b81e7c320222ab44d7ad732",
      "step_points_sum": 101,
      "task_id": "hvac"
    },
    {
      "declared_full_points": 76,
      "plan_id": "test4",
      "quality_flags": [],
      "sha256": "e8485d4ac76e43238dab53cb2074b537a916c5d1e0f2a96bbd96b7e465475c02",
      "step_points_sum": 76,
      "task_id": "hvac"
    },
    {
      "declared_full_points": 35,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "df376b5283f9456fd2ea221a9bf0120729cef393726ed8812baf70874737fd27",
      "step_points_sum": 35,
      "task_id": "language_learning"
    },
    {
      "declared_full_points": 30,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "289785f1a3cb77876d8783ea4931aa04af53f8ecc4c0666aab903cfe7f0399a5",
      "step_points_sum": 30,
      "task_id": "language_learning"
    },
    {
      "declared_full_points": 30,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "7a35cef4cd819c595b89227ebca3c330b676ab372d6f7c5db51edb3fa33e44da",
      "step_points_sum": 30,
      "task_id": "language_learning"
    },
    {
      "declared_full_points": 30,
      "plan_id": "test4",
      "quality_flags": [],
      "sha256": "8d6f7efb92d4337ad9e5b2715100fc3addaed8145b3a4ca6b902d0f1088b2bb7",
      "step_points_sum": 30,
      "task_id": "language_learning"
    },
    {
      "declared_full_points": 50,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "7b83907bb5f174ca23020bba8b66a81e6fcc9d71099482a5c90052b93e68d636",
      "step_points_sum": 50,
      "task_id": "logistics"
    },
    {
      "declared_full_points": 40,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "79b60d6ce29bed271b539e425ea31b2e961711cebb47c209eb0c40b9b2ad62ac",
      "step_points_sum": 40,
      "task_id": "logistics"
    },
    {
      "declared_full_points": 17,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "627afbd4895464218584a443b565a9cd77e3fe9fb6a18cd104f28996e47798e0",
      "step_points_sum": 17,
      "task_id": "logistics"
    },
    {
      "declared_full_points": 16,
      "plan_id": "test4",
      "quality_flags": [],
      "sha256": "a5e821703360339809fa0ebc6126232feacdfa6e00fa244bf79d97b90dab2e05",
      "step_points_sum": 16,
      "task_id": "logistics"
    },
    {
      "declared_full_points": 52,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "c7e4b171b9a3ae93a31ca9d316b84c0335ac23a99baa290e5f9c0c5fbddb642f",
      "step_points_sum": 52,
      "task_id": "mafia"
    },
    {
      "declared_full_points": 96,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "4a45c36ca7ccde0e402edd7dd881913257fa43a18a3b33124d51a68106ae55d1",
      "step_points_sum": 96,
      "task_id": "mafia"
    },
    {
      "declared_full_points": 68,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "ba403ed51c95acd093a8f9f32892540b36f821d542f843dc6451529ef3140754",
      "step_points_sum": 68,
      "task_id": "market_place"
    },
    {
      "declared_full_points": 45,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "aae697c568ca53d09702b1b5bcf3d40452a046831cd0efb09b8924286f31f8e5",
      "step_points_sum": 45,
      "task_id": "market_place"
    },
    {
      "declared_full_points": 40,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "7bc7628dd8b3438054dc12a5f41c28ecefafb2ec0342dacebd570e49664ea12a",
      "step_points_sum": 40,
      "task_id": "market_place"
    },
    {
      "declared_full_points": 85,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "6969ab2a280225d9f3fc8eb5773925e41848f77000c7ef6404fab43950a6e361",
      "step_points_sum": 85,
      "task_id": "monopoly"
    },
    {
      "declared_full_points": 85,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "1d34d5f893c44ad7a58405b1d7dd210df9d47b9ea8c37131603753bd5f622087",
      "step_points_sum": 85,
      "task_id": "monopoly"
    },
    {
      "declared_full_points": 50,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "9b4a50346ea755edf3c170003f1232bd6ad6c1750963c886e55abfb6920c18a7",
      "step_points_sum": 50,
      "task_id": "notes"
    },
    {
      "declared_full_points": 56,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "601ae0e67e1c654d9413c7f7b7d24489b1ae7af076ec4b8c7f12c01434a2aae1",
      "step_points_sum": 56,
      "task_id": "notes"
    },
    {
      "declared_full_points": 60,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "6fcb86c4c9aab1caa5d07574a9073ed0719dd08ff8edd557a787cca8839c40b7",
      "step_points_sum": 60,
      "task_id": "notes"
    },
    {
      "declared_full_points": 50,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "98f4ab6be50ada436c8b38d829b4a1ce29e609c9c754b7ee55cee6285592712d",
      "step_points_sum": 50,
      "task_id": "online_whiteboard"
    },
    {
      "declared_full_points": 50,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "585f71ee7f6d64a934494fe450f996ac88ab656d9e0ac83c5fbd6b692b01be3d",
      "step_points_sum": 50,
      "task_id": "online_whiteboard"
    },
    {
      "declared_full_points": 56,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "bd8940d23f4807d79e4e36362547b632c9130a1bc02cc4acf6042fb3b4996a1a",
      "step_points_sum": 56,
      "task_id": "online_whiteboard"
    },
    {
      "declared_full_points": 82,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "0c324b15f995df000f7d1d1a80dd5072af7ffb27a35a6e28d25e2f51de88cc46",
      "step_points_sum": 82,
      "task_id": "pilot_logbook"
    },
    {
      "declared_full_points": 78,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "add8ba2f2a136df45c3506336f29c7362fe922ca988e04ebb45110716025a064",
      "step_points_sum": 78,
      "task_id": "pilot_logbook"
    },
    {
      "declared_full_points": 45,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "72bf3cf95f608a745e17ced01ea766a816d918bb4db40a9d9804798350605f55",
      "step_points_sum": 45,
      "task_id": "pilot_logbook"
    },
    {
      "declared_full_points": 14,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "14951fa0c8d80b32867f859065a3cb54d847b892d9330d359cf5bcee6e9a7249",
      "step_points_sum": 14,
      "task_id": "quiz"
    },
    {
      "declared_full_points": 13,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "9dae77d59dc9fc143c33582b55a784131e868c94600b914f638cdd6f5e4193e3",
      "step_points_sum": 13,
      "task_id": "quiz"
    },
    {
      "declared_full_points": 20,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "57c23e4106447aa0f668e97535d708f480adfe99deb98c92b62ca4839b632ffa",
      "step_points_sum": 20,
      "task_id": "quiz"
    },
    {
      "declared_full_points": 36,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "12bee34083f2b696cdc140c0402f2c34840613e9b4cdaa673d54bb6518af93d4",
      "step_points_sum": 36,
      "task_id": "resume_builder"
    },
    {
      "declared_full_points": 40,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "3d57523bd314da5cc1b5a698f7d7f51d8d87e81ccb560684139cdac6abcadb9d",
      "step_points_sum": 40,
      "task_id": "resume_builder"
    },
    {
      "declared_full_points": 46,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "caab4000b9d0c10b8775a0dad1e80955691925f2938162a27293ca307a53da4a",
      "step_points_sum": 46,
      "task_id": "resume_builder"
    },
    {
      "declared_full_points": 34,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "6adc999f89307611d6bc677ce3fa097748e4d64bf91bd2bedd5be51244050ef6",
      "step_points_sum": 34,
      "task_id": "slack"
    },
    {
      "declared_full_points": 46,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "2f34e4fca7b76451f817ca44ca66a879fc3d389714273dc55f7d2f56a69ddfdd",
      "step_points_sum": 46,
      "task_id": "slack"
    },
    {
      "declared_full_points": 40,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "8b58cf235824629e5b52dd332b463fa60bed708eb417d0af83443aea27273b9d",
      "step_points_sum": 40,
      "task_id": "slack"
    },
    {
      "declared_full_points": 110,
      "plan_id": "test1",
      "quality_flags": [],
      "sha256": "0f602f63e73b18f33341740578cd19186d7c9b65a398e59979b0be61d499c0a2",
      "step_points_sum": 110,
      "task_id": "srm"
    },
    {
      "declared_full_points": 78,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "433c27eb027e28a408885e3894d8faa05513e807f614aa6bb8fa65ae99f1950e",
      "step_points_sum": 78,
      "task_id": "srm"
    },
    {
      "declared_full_points": 90,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "dca9698545f70208a74dfd0206da0a5b1c11e0c0f6003a63c0f913ab495e1e65",
      "step_points_sum": 90,
      "task_id": "srm"
    },
    {
      "declared_full_points": 96,
      "plan_id": "test4",
      "quality_flags": [],
      "sha256": "f0b36c9a8a60b6e8008f22adbe8cbb890a0b5f47aa1e71b202d69f0fe324bd2c",
      "step_points_sum": 96,
      "task_id": "srm"
    },
    {
      "declared_full_points": 55,
      "plan_id": "test1",
      "quality_flags": [
        "declared_points_do_not_match_steps"
      ],
      "sha256": "0ae61a6bae6431bcba48b2004e634faefdca2cdfb15253206dfe86771457d01d",
      "step_points_sum": 43,
      "task_id": "wedding"
    },
    {
      "declared_full_points": 70,
      "plan_id": "test2",
      "quality_flags": [],
      "sha256": "2b254bd20b2342b4a473ff34304bd66f0b8278a46022792a2c39772775e7ef2c",
      "step_points_sum": 70,
      "task_id": "wedding"
    },
    {
      "declared_full_points": 61,
      "plan_id": "test3",
      "quality_flags": [],
      "sha256": "aa3890a07a0865a0e328c6cd73c2ac5929f8aa0d1efd1cdb69668019f8da87e1",
      "step_points_sum": 61,
      "task_id": "wedding"
    },
    {
      "declared_full_points": 65,
      "plan_id": "test4",
      "quality_flags": [],
      "sha256": "2fdb7ffdcc42d9b20400947695b9891758a0529b22106dbb8deabc1b19debf76",
      "step_points_sum": 65,
      "task_id": "wedding"
    }
  ],
  "summary": {
    "eligible_scored_cells": 0,
    "missing_cells": 79,
    "observed_cells": 17,
    "scored_cells": 1
  },
  "tasks": [
    {
      "id": "barber",
      "label": "Barber Scheduling",
      "prd": {
        "sha256": "c95414bb5b727a8f1278bc64d88a667aad3af7e86847f8c3d5749436f7f8b5f0",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/barber/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "003204e1c71e6879ff4dc940e67d38025d6f450d0c178e948d3e22aba61aa353",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/barber/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "06395372fa0f16d7a3f5c33ff7c86c3112326758b32972de76356977f53f45d7",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/barber/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "e39cb7e58c0578f95e52b3e4aafd932ecdf60a4f21163e5c224698e29b29c57c",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/barber/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "book_journey",
      "label": "Book Journey",
      "prd": {
        "sha256": "81c1ae3f829128aad59b943997e64fb7ecd16bb25cf29371ceba2fd12ae7a223",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/book_journey/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "dd82119148a7b7b1105e5ba10dc3a8b0c853b88c7edf57a68a65ee1e9df179b8",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/book_journey/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "6d5cdd8fb1ab1c7d699b52fab547bcd0ba01f4020ec7d961d9136566444dd800",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/book_journey/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "480d1fcc9f05bc27afacdeaf9baf492450e9aa21b2b5a4eec4cbb7047a65fd64",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/book_journey/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "canary",
      "label": "Canary",
      "prd": {
        "sha256": "9bd41a6cd5e86b39dba52d9e79db6fc920cacc33f9c8b47719b64d641ea5c628",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/canary/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "d4c9216b30805924797ad31a7d804ec26e27573adc25532525fde00cb929f560",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/canary/tests/mvp/test1.txt"
        }
      ]
    },
    {
      "id": "collabrative_kaban",
      "label": "Collaborative Kanban",
      "prd": {
        "sha256": "a696df6676ae0b0b1cf7a812781842ea5cf2ea1de1486ff4c02afd18d893bb4e",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/collabrative_kaban/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "5481dd6152a6b25b753762873bea3be9a0223d061a8984e1b38e08ccd1cab432",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/collabrative_kaban/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "54ba13e9597ca0b124504382ab7c3e13a61db2cfc4d37ca1ceb339034f5609d9",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/collabrative_kaban/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "5b55ae62e1adda1159d40c4fae9f977e4ecf3d36298079fcb2a72bd5f5b12109",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/collabrative_kaban/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "creative_community",
      "label": "Creative Community",
      "prd": {
        "sha256": "fdf0894a081e475f78dfbf1beb2c792526606aeec1a44e54ce4431f5eb47fc8b",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/creative_community/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "480e5108193d6e5f9040dd7b7c3941dcef3078adfb96b2b4dee3f9dd22e8a17a",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/creative_community/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "29fb38d3ea2915e39b0ae24b1726711c37d7c3adedbd2b9fd48c3290e860e7f5",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/creative_community/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "dd54d7bfdb6809e0979439f5262ba49febd99c14f6a4b2c0c1826fa74258b623",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/creative_community/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "energy_audit",
      "label": "Energy Audit",
      "prd": {
        "sha256": "4b38cf0ec59e7847d36e0412fede298ff07f8a7bcc7e2ed337992086da4c8eb5",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/energy_audit/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "a8b40336fea74c5504cf9325570aede62dc409fef5e9a379fe420be8be546df5",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/energy_audit/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "d82eee7f16d1747d87cfdec12490824adcfa9e91215edaec4a1a01f2efff389f",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/energy_audit/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "43b566f24f600739f017cfd909ebbd207aade8130c866fee0cee5720d3afb665",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/energy_audit/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "family_friendly_venue",
      "label": "Family-Friendly Venue",
      "prd": {
        "sha256": "12d11726347aa0a22c6a45984d2e6c4acb696d29537b16c924748d3b9764e5ff",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/family_friendly_venue/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "86cd00d235071bb06dae3d8ee332281dd2cf1a4df818793a1b5b14c67fe94550",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/family_friendly_venue/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "1501413dae3241ef80e7a533a98753fee19546a9c19daa98b17fab29670f94be",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/family_friendly_venue/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "bcaba778c93b701523917d94fa7763eadb0c3a5fe8bea3197d1aace7fdc373b6",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/family_friendly_venue/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "family_social",
      "label": "Family Social",
      "prd": {
        "sha256": "91b8d018ef7becb5fb26e3b1abb27c53663f207865ff65e4a3fa6fef56f9d6cd",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/family_social/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "9ec05f0985a3e02824fe77283f361c34bfc35c4ffe6e43802a13c8198b2377c3",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/family_social/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "7fbd00444b0174d355d76c1027aecd0e84caf5152029dfcdef38389e31ab8167",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/family_social/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "47f1dd845b20bf12eeaf845a0767c810c45e4d135f0c48cbf872c5be01556d59",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/family_social/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "fleet_management",
      "label": "Fleet Management",
      "prd": {
        "sha256": "7d289dd4be8b8a09c4197494f63370b08c45e155e4b866c6e6665a433fd3d868",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/fleet_management/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "c15160fdb9fba23b3718d854f1c655448a5044e4f2fb42389df820fd1ff87660",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/fleet_management/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "932ae1ccba455a7d2102ce1ed928e245f56e7f3f908943f06fce409dd2283166",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/fleet_management/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "fbb48b545a13929f2dd0a22b9133e0563fde8e84b96fb42fb41bae59f6128f96",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/fleet_management/tests/mvp/test3.txt"
        },
        {
          "id": "test4",
          "sha256": "15f87e5cd384b920bd70b83c3e2501bc3d8718a2b071b1cee92edd5d1ab947e3",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/fleet_management/tests/mvp/test4.txt"
        }
      ]
    },
    {
      "id": "furniture_freight",
      "label": "Furniture Freight",
      "prd": {
        "sha256": "bad19dbc3e6df661be47aad9299ea36ac8ccf7e207850e4c9f7f87732df5f937",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/furniture_freight/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "1db9c81dd113cb71fe2906c80e0b0634c4c02e224bb966ad918dc70583127ee5",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/furniture_freight/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "b27ebbe23e93d5153dfd1c64c1a660be76f367024ed9794c05e36924a1d5715e",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/furniture_freight/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "0d7edb65fd160482697bb3079137cb8df6d965170f29795cbefb7ffe4683b66c",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/furniture_freight/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "hvac",
      "label": "HVAC Operations",
      "prd": {
        "sha256": "be2fa71c731dbf5754daaae9dacd3ced9cf0af1f589f6da1b937bf5f490b7af3",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/hvac/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "57818ef3f0fdf0527daff8ddd629daa016c96d5419ecebee4b70b64987583d24",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/hvac/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "bffc068c4f4cf7e8a4a7f9b8bb087d8dd1c0fc92ffaa11e8d2e3b0e2f4527f55",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/hvac/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "19b1097b69364bbf5e7edf2b3cebc354cf6af8ff8b81e7c320222ab44d7ad732",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/hvac/tests/mvp/test3.txt"
        },
        {
          "id": "test4",
          "sha256": "e8485d4ac76e43238dab53cb2074b537a916c5d1e0f2a96bbd96b7e465475c02",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/hvac/tests/mvp/test4.txt"
        }
      ]
    },
    {
      "id": "language_learning",
      "label": "Language Learning",
      "prd": {
        "sha256": "e06bcaf39eb910a01fbafc765728c3143a91fdbbe1d0eee62850baac02fa5348",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/language_learning/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "df376b5283f9456fd2ea221a9bf0120729cef393726ed8812baf70874737fd27",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/language_learning/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "289785f1a3cb77876d8783ea4931aa04af53f8ecc4c0666aab903cfe7f0399a5",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/language_learning/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "7a35cef4cd819c595b89227ebca3c330b676ab372d6f7c5db51edb3fa33e44da",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/language_learning/tests/mvp/test3.txt"
        },
        {
          "id": "test4",
          "sha256": "8d6f7efb92d4337ad9e5b2715100fc3addaed8145b3a4ca6b902d0f1088b2bb7",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/language_learning/tests/mvp/test4.txt"
        }
      ]
    },
    {
      "id": "logistics",
      "label": "Logistics",
      "prd": {
        "sha256": "5a50d39ddc6f2ce2a51fc2c958d95d420c7526dddf13ec4694461e0beb59a299",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/logistics/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "7b83907bb5f174ca23020bba8b66a81e6fcc9d71099482a5c90052b93e68d636",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/logistics/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "79b60d6ce29bed271b539e425ea31b2e961711cebb47c209eb0c40b9b2ad62ac",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/logistics/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "627afbd4895464218584a443b565a9cd77e3fe9fb6a18cd104f28996e47798e0",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/logistics/tests/mvp/test3.txt"
        },
        {
          "id": "test4",
          "sha256": "a5e821703360339809fa0ebc6126232feacdfa6e00fa244bf79d97b90dab2e05",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/logistics/tests/mvp/test4.txt"
        }
      ]
    },
    {
      "id": "mafia",
      "label": "Mafia",
      "prd": {
        "sha256": "d142a3437fedf897db0af8632f7e6aa6a30b35711ca8c5ff73617ba8aa8b3ce7",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/mafia/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "c7e4b171b9a3ae93a31ca9d316b84c0335ac23a99baa290e5f9c0c5fbddb642f",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/mafia/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "4a45c36ca7ccde0e402edd7dd881913257fa43a18a3b33124d51a68106ae55d1",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/mafia/tests/mvp/test2.txt"
        }
      ]
    },
    {
      "id": "market_place",
      "label": "Marketplace",
      "prd": {
        "sha256": "356d4d0d0f4a5cace5dca4d7533c16ae175b013f31d81dd8d8cb6f465312163c",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/market_place/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "ba403ed51c95acd093a8f9f32892540b36f821d542f843dc6451529ef3140754",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/market_place/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "aae697c568ca53d09702b1b5bcf3d40452a046831cd0efb09b8924286f31f8e5",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/market_place/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "7bc7628dd8b3438054dc12a5f41c28ecefafb2ec0342dacebd570e49664ea12a",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/market_place/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "monopoly",
      "label": "Monopoly",
      "prd": {
        "sha256": "b069e936cc59e5bca0fbdbec1df862a6509719cb4146b57837b88cf3f0f31423",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/monopoly/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "6969ab2a280225d9f3fc8eb5773925e41848f77000c7ef6404fab43950a6e361",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/monopoly/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "1d34d5f893c44ad7a58405b1d7dd210df9d47b9ea8c37131603753bd5f622087",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/monopoly/tests/mvp/test2.txt"
        }
      ]
    },
    {
      "id": "notes",
      "label": "Notes",
      "prd": {
        "sha256": "cf04c26624a580621f534385e1cd8889c353f4a9e7e2f023d17f1b9431b38430",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/notes/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "9b4a50346ea755edf3c170003f1232bd6ad6c1750963c886e55abfb6920c18a7",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/notes/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "601ae0e67e1c654d9413c7f7b7d24489b1ae7af076ec4b8c7f12c01434a2aae1",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/notes/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "6fcb86c4c9aab1caa5d07574a9073ed0719dd08ff8edd557a787cca8839c40b7",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/notes/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "online_whiteboard",
      "label": "Online Whiteboard",
      "prd": {
        "sha256": "c9a786092128c294e2038e23aef24d5be82f26e13edcf678d17bc9080079a9f3",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/online_whiteboard/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "98f4ab6be50ada436c8b38d829b4a1ce29e609c9c754b7ee55cee6285592712d",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/online_whiteboard/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "585f71ee7f6d64a934494fe450f996ac88ab656d9e0ac83c5fbd6b692b01be3d",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/online_whiteboard/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "bd8940d23f4807d79e4e36362547b632c9130a1bc02cc4acf6042fb3b4996a1a",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/online_whiteboard/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "pilot_logbook",
      "label": "Pilot Logbook",
      "prd": {
        "sha256": "0040fa953d5e169d9519812423e7b14861011cc5d58ee60dae51c30ff3e493af",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/pilot_logbook/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "0c324b15f995df000f7d1d1a80dd5072af7ffb27a35a6e28d25e2f51de88cc46",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/pilot_logbook/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "add8ba2f2a136df45c3506336f29c7362fe922ca988e04ebb45110716025a064",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/pilot_logbook/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "72bf3cf95f608a745e17ced01ea766a816d918bb4db40a9d9804798350605f55",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/pilot_logbook/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "quiz",
      "label": "Quiz",
      "prd": {
        "sha256": "2e63c602a6ffd4d3738dc01185bb61aa929274af5e13cda62107c28f034a7b2b",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/quiz/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "14951fa0c8d80b32867f859065a3cb54d847b892d9330d359cf5bcee6e9a7249",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/quiz/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "9dae77d59dc9fc143c33582b55a784131e868c94600b914f638cdd6f5e4193e3",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/quiz/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "57c23e4106447aa0f668e97535d708f480adfe99deb98c92b62ca4839b632ffa",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/quiz/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "resume_builder",
      "label": "Resume Builder",
      "prd": {
        "sha256": "948f112a38ed30ce41e7e0de9c6cc6f645d3338c160a0489d009b54c501775e2",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/resume_builder/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "12bee34083f2b696cdc140c0402f2c34840613e9b4cdaa673d54bb6518af93d4",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/resume_builder/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "3d57523bd314da5cc1b5a698f7d7f51d8d87e81ccb560684139cdac6abcadb9d",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/resume_builder/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "caab4000b9d0c10b8775a0dad1e80955691925f2938162a27293ca307a53da4a",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/resume_builder/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "slack",
      "label": "Team Chat",
      "prd": {
        "sha256": "d613fafd4d5e275bf1ef991b5dbf1b2c0c5861238ffc99d3248314d174cc4034",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/slack/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "6adc999f89307611d6bc677ce3fa097748e4d64bf91bd2bedd5be51244050ef6",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/slack/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "2f34e4fca7b76451f817ca44ca66a879fc3d389714273dc55f7d2f56a69ddfdd",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/slack/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "8b58cf235824629e5b52dd332b463fa60bed708eb417d0af83443aea27273b9d",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/slack/tests/mvp/test3.txt"
        }
      ]
    },
    {
      "id": "srm",
      "label": "Supplier Relationship Management",
      "prd": {
        "sha256": "0b1e8ca0739c4b05d5f274e32d69de8dccb652e87989831bfcfdadf794668b08",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/srm/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "0f602f63e73b18f33341740578cd19186d7c9b65a398e59979b0be61d499c0a2",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/srm/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "433c27eb027e28a408885e3894d8faa05513e807f614aa6bb8fa65ae99f1950e",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/srm/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "dca9698545f70208a74dfd0206da0a5b1c11e0c0f6003a63c0f913ab495e1e65",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/srm/tests/mvp/test3.txt"
        },
        {
          "id": "test4",
          "sha256": "f0b36c9a8a60b6e8008f22adbe8cbb890a0b5f47aa1e71b202d69f0fe324bd2c",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/srm/tests/mvp/test4.txt"
        }
      ]
    },
    {
      "id": "wedding",
      "label": "Wedding Venue Planning",
      "prd": {
        "sha256": "e9e4f4dc8035c9fb79145cc113b09b6831a8b8c2c8fbbb40f0964a9534a68413",
        "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/wedding/prd/mvp.txt"
      },
      "test_plans": [
        {
          "id": "test1",
          "sha256": "0ae61a6bae6431bcba48b2004e634faefdca2cdfb15253206dfe86771457d01d",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/wedding/tests/mvp/test1.txt"
        },
        {
          "id": "test2",
          "sha256": "2b254bd20b2342b4a473ff34304bd66f0b8278a46022792a2c39772775e7ef2c",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/wedding/tests/mvp/test2.txt"
        },
        {
          "id": "test3",
          "sha256": "aa3890a07a0865a0e328c6cd73c2ac5929f8aa0d1efd1cdb69668019f8da87e1",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/wedding/tests/mvp/test3.txt"
        },
        {
          "id": "test4",
          "sha256": "2fdb7ffdcc42d9b20400947695b9891758a0529b22106dbb8deabc1b19debf76",
          "url": "https://github.com/ViBench/vibench-public/blob/5baa6892bad7dcf6ec0bee4f44e4e46445271f19/prds/wedding/tests/mvp/test4.txt"
        }
      ]
    }
  ],
  "treatment_rollups": [
    {
      "audit_diagnostic": {
        "app_macro_percent": null,
        "apps": [
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "barber"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "book_journey"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 1,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "canary"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "collabrative_kaban"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "creative_community"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "energy_audit"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_friendly_venue"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_social"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "fleet_management"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "furniture_freight"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "hvac"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "language_learning"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "logistics"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "mafia"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "market_place"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "monopoly"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "notes"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "online_whiteboard"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "pilot_logbook"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "quiz"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "resume_builder"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "slack"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "srm"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "wedding"
          }
        ],
        "complete": false,
        "excluded_plans": [
          {
            "plan_id": "test1",
            "quality_flags": [
              "declared_points_do_not_match_steps"
            ],
            "task_id": "wedding"
          }
        ],
        "label": "anomaly_excluded_diagnostic_not_upstream_primary",
        "micro_percent": null,
        "plan_macro_percent": null,
        "plans_resolved": 0,
        "plans_total": 73,
        "provisional": {
          "app_macro_percent": null,
          "micro_percent": null,
          "plan_macro_percent": null,
          "resolved_full_points": 0,
          "resolved_points": 0
        }
      },
      "coverage": {
        "complete": false,
        "plans_resolved": 0,
        "plans_total": 74,
        "status_counts": {
          "candidate_build_failure": 0,
          "candidate_seeding_failure": 0,
          "evaluated": 0,
          "evaluator_error": 0,
          "not_run": 74
        }
      },
      "label": "Codex Sol High · native CLI",
      "semantics": "pinned_vibench_analyze_results_compatible",
      "treatment_id": "codex-sol-high-native",
      "universe": {
        "apps_total": 24,
        "declared_full_points": 3794,
        "plans_total": 74,
        "plans_with_quality_flags": 1
      },
      "vibench": {
        "app_macro_percent": null,
        "apps": [
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "barber"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "book_journey"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 1,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "canary"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "collabrative_kaban"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "creative_community"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "energy_audit"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_friendly_venue"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_social"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "fleet_management"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "furniture_freight"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "hvac"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "language_learning"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "logistics"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "mafia"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "market_place"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "monopoly"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "notes"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "online_whiteboard"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "pilot_logbook"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "quiz"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "resume_builder"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "slack"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "srm"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "wedding"
          }
        ],
        "complete": false,
        "micro_percent": null,
        "plan_macro_percent": null,
        "plans_resolved": 0,
        "plans_total": 74,
        "provisional": {
          "app_macro_percent": null,
          "micro_percent": null,
          "plan_macro_percent": null,
          "resolved_full_points": 0,
          "resolved_points": 0
        }
      }
    },
    {
      "audit_diagnostic": {
        "app_macro_percent": null,
        "apps": [
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "barber"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "book_journey"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 1,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "canary"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "collabrative_kaban"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "creative_community"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "energy_audit"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_friendly_venue"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_social"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "fleet_management"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "furniture_freight"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "hvac"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "language_learning"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "logistics"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "mafia"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "market_place"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "monopoly"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "notes"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "online_whiteboard"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "pilot_logbook"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "quiz"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "resume_builder"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "slack"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "srm"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "wedding"
          }
        ],
        "complete": false,
        "excluded_plans": [
          {
            "plan_id": "test1",
            "quality_flags": [
              "declared_points_do_not_match_steps"
            ],
            "task_id": "wedding"
          }
        ],
        "label": "anomaly_excluded_diagnostic_not_upstream_primary",
        "micro_percent": null,
        "plan_macro_percent": null,
        "plans_resolved": 0,
        "plans_total": 73,
        "provisional": {
          "app_macro_percent": null,
          "micro_percent": null,
          "plan_macro_percent": null,
          "resolved_full_points": 0,
          "resolved_points": 0
        }
      },
      "coverage": {
        "complete": false,
        "plans_resolved": 0,
        "plans_total": 74,
        "status_counts": {
          "candidate_build_failure": 0,
          "candidate_seeding_failure": 0,
          "evaluated": 0,
          "evaluator_error": 0,
          "not_run": 74
        }
      },
      "label": "Grok 4.6 X-High · native CLI",
      "semantics": "pinned_vibench_analyze_results_compatible",
      "treatment_id": "grok-4.6-xhigh-native",
      "universe": {
        "apps_total": 24,
        "declared_full_points": 3794,
        "plans_total": 74,
        "plans_with_quality_flags": 1
      },
      "vibench": {
        "app_macro_percent": null,
        "apps": [
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "barber"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "book_journey"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 1,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "canary"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "collabrative_kaban"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "creative_community"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "energy_audit"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_friendly_venue"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_social"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "fleet_management"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "furniture_freight"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "hvac"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "language_learning"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "logistics"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "mafia"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "market_place"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "monopoly"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "notes"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "online_whiteboard"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "pilot_logbook"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "quiz"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "resume_builder"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "slack"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "srm"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "wedding"
          }
        ],
        "complete": false,
        "micro_percent": null,
        "plan_macro_percent": null,
        "plans_resolved": 0,
        "plans_total": 74,
        "provisional": {
          "app_macro_percent": null,
          "micro_percent": null,
          "plan_macro_percent": null,
          "resolved_full_points": 0,
          "resolved_points": 0
        }
      }
    },
    {
      "audit_diagnostic": {
        "app_macro_percent": null,
        "apps": [
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "barber"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "book_journey"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 1,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "canary"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "collabrative_kaban"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "creative_community"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "energy_audit"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_friendly_venue"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_social"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "fleet_management"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "furniture_freight"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "hvac"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "language_learning"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "logistics"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "mafia"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "market_place"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "monopoly"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "notes"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "online_whiteboard"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "pilot_logbook"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "quiz"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "resume_builder"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "slack"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "srm"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "wedding"
          }
        ],
        "complete": false,
        "excluded_plans": [
          {
            "plan_id": "test1",
            "quality_flags": [
              "declared_points_do_not_match_steps"
            ],
            "task_id": "wedding"
          }
        ],
        "label": "anomaly_excluded_diagnostic_not_upstream_primary",
        "micro_percent": null,
        "plan_macro_percent": null,
        "plans_resolved": 0,
        "plans_total": 73,
        "provisional": {
          "app_macro_percent": null,
          "micro_percent": null,
          "plan_macro_percent": null,
          "resolved_full_points": 0,
          "resolved_points": 0
        }
      },
      "coverage": {
        "complete": false,
        "plans_resolved": 0,
        "plans_total": 74,
        "status_counts": {
          "candidate_build_failure": 0,
          "candidate_seeding_failure": 0,
          "evaluated": 0,
          "evaluator_error": 0,
          "not_run": 74
        }
      },
      "label": "RepoPrompt CE · Agent Mode defaults",
      "semantics": "pinned_vibench_analyze_results_compatible",
      "treatment_id": "repoprompt-agent-default",
      "universe": {
        "apps_total": 24,
        "declared_full_points": 3794,
        "plans_total": 74,
        "plans_with_quality_flags": 1
      },
      "vibench": {
        "app_macro_percent": null,
        "apps": [
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "barber"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "book_journey"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 1,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "canary"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "collabrative_kaban"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "creative_community"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "energy_audit"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_friendly_venue"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_social"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "fleet_management"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "furniture_freight"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "hvac"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "language_learning"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "logistics"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "mafia"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "market_place"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "monopoly"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "notes"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "online_whiteboard"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "pilot_logbook"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "quiz"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "resume_builder"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "slack"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "srm"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "wedding"
          }
        ],
        "complete": false,
        "micro_percent": null,
        "plan_macro_percent": null,
        "plans_resolved": 0,
        "plans_total": 74,
        "provisional": {
          "app_macro_percent": null,
          "micro_percent": null,
          "plan_macro_percent": null,
          "resolved_full_points": 0,
          "resolved_points": 0
        }
      }
    },
    {
      "audit_diagnostic": {
        "app_macro_percent": null,
        "apps": [
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "barber"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "book_journey"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 1,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "canary"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "collabrative_kaban"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "creative_community"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "energy_audit"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_friendly_venue"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_social"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "fleet_management"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "furniture_freight"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "hvac"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "language_learning"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "logistics"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "mafia"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "market_place"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "monopoly"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "notes"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "online_whiteboard"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "pilot_logbook"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "quiz"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "resume_builder"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "slack"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "srm"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "wedding"
          }
        ],
        "complete": false,
        "excluded_plans": [
          {
            "plan_id": "test1",
            "quality_flags": [
              "declared_points_do_not_match_steps"
            ],
            "task_id": "wedding"
          }
        ],
        "label": "anomaly_excluded_diagnostic_not_upstream_primary",
        "micro_percent": null,
        "plan_macro_percent": null,
        "plans_resolved": 0,
        "plans_total": 73,
        "provisional": {
          "app_macro_percent": null,
          "micro_percent": null,
          "plan_macro_percent": null,
          "resolved_full_points": 0,
          "resolved_points": 0
        }
      },
      "coverage": {
        "complete": false,
        "plans_resolved": 0,
        "plans_total": 74,
        "status_counts": {
          "candidate_build_failure": 0,
          "candidate_seeding_failure": 0,
          "evaluated": 0,
          "evaluator_error": 0,
          "not_run": 74
        }
      },
      "label": "Claude Fable → Opus 5 High · native orchestration",
      "semantics": "pinned_vibench_analyze_results_compatible",
      "treatment_id": "claude-fable-opus5-high-native-orchestration",
      "universe": {
        "apps_total": 24,
        "declared_full_points": 3794,
        "plans_total": 74,
        "plans_with_quality_flags": 1
      },
      "vibench": {
        "app_macro_percent": null,
        "apps": [
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "barber"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "book_journey"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 1,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "canary"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "collabrative_kaban"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "creative_community"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "energy_audit"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_friendly_venue"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "family_social"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "fleet_management"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "furniture_freight"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "hvac"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "language_learning"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "logistics"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "mafia"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "market_place"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 2,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "monopoly"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "notes"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "online_whiteboard"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "pilot_logbook"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "quiz"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "resume_builder"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 3,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "slack"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "srm"
          },
          {
            "complete": false,
            "plans_resolved": 0,
            "plans_total": 4,
            "provisional_score_percent": null,
            "score_percent": null,
            "task_id": "wedding"
          }
        ],
        "complete": false,
        "micro_percent": null,
        "plan_macro_percent": null,
        "plans_resolved": 0,
        "plans_total": 74,
        "provisional": {
          "app_macro_percent": null,
          "micro_percent": null,
          "plan_macro_percent": null,
          "resolved_full_points": 0,
          "resolved_points": 0
        }
      }
    }
  ],
  "treatments": [
    {
      "id": "codex-sol-high-native",
      "label": "Codex Sol High · native CLI",
      "requested_effort": "high",
      "requested_model": "gpt-5.6-sol",
      "runner": {
        "adapter": "codex",
        "agent_config_id": "codex-sol-high",
        "blocker": null,
        "kind": "native_cli",
        "status": "ready"
      },
      "strategy": "native_single_agent"
    },
    {
      "id": "grok-4.6-xhigh-native",
      "label": "Grok 4.6 X-High · native CLI",
      "requested_effort": "xhigh",
      "requested_model": "grok-4.6",
      "runner": {
        "adapter": "grok",
        "agent_config_id": "grok-4.6-xhigh",
        "blocker": "Native Grok CLI OAuth is not authenticated, and effective effort attestation is not available.",
        "kind": "native_cli",
        "status": "blocked"
      },
      "strategy": "native_single_agent"
    },
    {
      "id": "repoprompt-agent-default",
      "label": "RepoPrompt CE · Agent Mode defaults",
      "requested_effort": "product default",
      "requested_model": "operator-configured Agent Mode stack",
      "runner": {
        "adapter": "repoprompt_ce",
        "agent_config_id": null,
        "blocker": "The operator must drive RepoPrompt CE Agent Mode and finalize each session capture.",
        "kind": "repoprompt_gui",
        "status": "manual"
      },
      "strategy": "repoprompt_agent_mode"
    },
    {
      "id": "claude-fable-opus5-high-native-orchestration",
      "label": "Claude Fable → Opus 5 High · native orchestration",
      "requested_effort": "high",
      "requested_model": "Fable orchestrating Opus 5",
      "runner": {
        "adapter": "claude",
        "agent_config_id": "claude-fable-opus5-high",
        "blocker": "Claude Code OAuth is not authenticated on this host.",
        "kind": "native_cli",
        "status": "blocked"
      },
      "strategy": "native_same_provider_orchestration"
    }
  ]
}