{
  "schema_version": "veriharness_version_registry_v0.1",
  "project_id": "veriharness-v0.2",
  "versions": [
    {
      "id": "v0.2.0-prompt-redesign",
      "label": "v0.2.0 · Prompt 抽象重设计",
      "status": "complete 60/60",
      "page": "sp-veriharness-v0-2.html",
      "href": "sp-veriharness-v0-2.html#vh2-version-lineage",
      "page_id": "vh2-v020",
      "experiment_id": "veriharness-v0.2-formal-20260714",
      "attempt_group": {
        "kind": "abstract-verification-prompt-redesign",
        "contains": [
          "native claude -p inner loop",
          "one general CallMLLM tool",
          "P0 tool plus task baseline",
          "P1 appended verification policy",
          "10 cases, two conditions, three paired replicates"
        ]
      },
      "public_result": "0-Project/20260713-veriharness-v0-2/docs/public-result-summary-v0.2.md",
      "result_status": "complete_60_of_60",
      "scope_guardrail": "prompt-only abstract policy redesign; no harness hard gate"
    },
    {
      "id": "runtime-full-native-core1-v0.1",
      "label": "v0.2.1 · Core 1 / Mentor Prompt 尝试",
      "status": "complete mechanism screen",
      "page": "sp-veriharness-v0-2-core1.html",
      "experiment_id": "veriharness-v0.2-core1-case-h3-h6-v0.1",
      "attempt_group": {
        "kind": "runtime-and-first-verification-screen",
        "contains": [
          "controlled non-bare Claude Code baseline",
          "default native tools plus mcp__mllm__call",
          "B1 Lite shared prompt",
          "B2 v0.1 unchanged",
          "P0 baseline versus B3 Core-1",
          "case-h3 and case-h6, three paired replicates"
        ]
      },
      "goal_spec": "0-Project/20260713-veriharness-v0-2/docs/experiments/verification-core-ladder/GOAL_SPEC.md",
      "run_root": "0-Project/20260713-veriharness-v0-2/runs/core1-case-h3-h6-20260716",
      "public_result": "0-Project/20260713-veriharness-v0-2/docs/experiments/verification-core-ladder/core1-result-v0.1.md",
      "viewer": {
        "experiment_id": "veriharness-v0-2-core1-screen",
        "overview": "0-Project/20260704-vdiff-vharness-visual-trajectory-viewer/public/?view=overview&experiment=veriharness-v0-2-core1-screen&dataset=veriharness-v0-2-core1-c1-r1",
        "datasets": 6,
        "case_runs": 12
      },
      "result_status": "complete_12_of_12",
      "last_verified": "2026-07-16",
      "timing": {
        "generation_seconds": 467.38,
        "evaluation_seconds": 96.215,
        "active_pipeline_seconds": 563.595
      },
      "case_mean_offline_total_effect": 0.176667,
      "scope_guardrail": "two selected cases; mechanism screen, not benchmark-wide conclusion"
    },
    {
      "id": "verification-prompt-lab-p0-p3-v0.1",
      "label": "v0.2.2 · Verification Prompt Lab",
      "status": "complete P0–P3 · 54 case-runs",
      "page": "sp-verification-prompt-lab.html",
      "experiment_id": "veriharness-v0-2-core03-prompt-diagnosis",
      "attempt_group": {
        "kind": "verification-prompt-and-behavior-diagnosis-series",
        "contains": [
          "S0/S1 historical Core 0.1 short-base controls",
          "S2 short base plus Core V1 B3 in user prompt",
          "F0/F1 historical full Core V1 controls",
          "F2 full Core V1 base plus mentor verification in appended system",
          "case-h3 and case-h6, three replicates per condition",
          "P3 channel bridge, tool instruction isolation, task isolation, and system-channel 2x2 interaction"
        ]
      },
      "goal_spec": "0-Project/20260713-veriharness-v0-2/docs/experiments/verification-prompt-lab-p3/GOAL_SPEC.md",
      "run_roots": [
        "0-Project/20260713-veriharness-v0-2/runs/core03-s2-short-core1-b3-user-20260716",
        "0-Project/20260713-veriharness-v0-2/runs/core03-f2-full-mentor-system-20260716",
        "0-Project/20260713-veriharness-v0-2/runs/p3-mentor-transfer-20260716"
      ],
      "public_result": "0-Project/20260713-veriharness-v0-2/docs/experiments/verification-prompt-lab-p3/RESULT.md",
      "viewer": {
        "experiment_id": "veriharness-v0-2-core03-prompt-diagnosis",
        "overview": "0-Project/20260704-vdiff-vharness-visual-trajectory-viewer/public/?view=overview&experiment=veriharness-v0-2-core03-prompt-diagnosis&dataset=veriharness-v0-2-p3-o0-tm-vm-s-r1",
        "datasets": 27,
        "case_runs": 54
      },
      "result_status": "complete_54_of_54_with_18_new_p3",
      "last_verified": "2026-07-16",
      "deployment": {
        "preview": "https://c052de27.multimodal-verification-board.pages.dev/sp-verification-prompt-lab",
        "preview_alias": "https://prompt-lab-p3-20260716.multimodal-verification-board.pages.dev/sp-verification-prompt-lab",
        "stable": "https://multimodal-verification-board.pages.dev/sp-verification-prompt-lab",
        "qa": "0-Project/20260713-veriharness-v0-2/docs/qa/verification-prompt-lab-p3-20260716/README.md",
        "legacy_core03_page": "https://multimodal-verification-board.pages.dev/sp-veriharness-v0-2-core03",
        "production_recorded": "https://a7f0b95b.multimodal-verification-board.pages.dev/sp-verification-prompt-lab"
      },
      "shared_bank_sha256": {
        "case-h3": "fdfa0bb5dfe2ac2b616a63b4538bf2953f3835179bc13a117769f31da62447dc",
        "case-h6": "92ed0092c8eccc82d79fff93799e4064310e4a9f82451cf9986d1bf4c3a704a4"
      },
      "root_cause": "P3 isolates delivery channel as a major suppressor of the mentor long loop; Core1 tool instructions further reduce closed-loop behavior under the mentor task, while Core1 task effects vary by case.",
      "scope_guardrail": "two selected cases; diagnostic evidence, not benchmark-wide statistical significance"
    },
    {
      "id": "v0.2.3-verification-prompt-ablation",
      "label": "v0.2.3 · Verification Prompt Ablation",
      "status": "initial 60/60 complete · V3 follow-up generated 12/12",
      "page": "sp-veriharness-v0-2-3.html",
      "href": "sp-veriharness-v0-2-3.html",
      "page_id": "sp-veriharness-v0-2-3",
      "experiment_id": "veriharness-v0.2.3-verification-prompt-ablation",
      "attempt_group": {
        "kind": "verification-prompt-ablation-with-general-callmllm",
        "contains": [
          "Tool V1 and Task V1 fixed across conditions",
          "C0 no Verification prompt",
          "C1 V0 Mentor Minimal",
          "C2 V1 Initial Full",
          "C3 V2 Overall Compact",
          "five image-only cases and three replicates per condition",
          "Focused Validate versus Gate-like Checkpoint trajectory analysis",
          "V3 Two-Checkpoint Closure retained for the active follow-up iteration"
        ]
      },
      "result_status": "initial_complete_60_of_60_followup_generation_complete_evaluation_pending",
      "scope_guardrail": "initial result is five image-only cases; Gate-like labels describe observed agent behavior, not a harness hard gate",
      "run_root": "0-Project/20260718-veriharness-v0-2-3/runs/v023-verification-ablation-5case-20260718-01",
      "public_result": "0-Project/20260718-veriharness-v0-2-3/docs/results/20260718-v023-verification-ablation-5case-v0.1.md",
      "viewer": {
        "experiment_id": "veriharness-v0-2-3-verification-ablation",
        "overview": "0-Project/20260704-vdiff-vharness-visual-trajectory-viewer/public/?view=overview&experiment=veriharness-v0-2-3-verification-ablation&dataset=veriharness-v0-2-3-c1-r1",
        "datasets": 12,
        "case_runs": 60
      }
    }
  ]
}