Complete mode1 twin_holds

minimax/mode1/twin_holds/evaluation/report.md

@@ -0,0 +1,118 @@
+# taskgen eval — brief / twin_holds
+
+resolved=no eligibility=failed | ranking_eligible=no (measurement coverage, not contract)
+verifier_profile=brief: judge evidence interpretation, authored GDD quality, executable self-clear causality, and GT-conditioned product fidelity
+
+Evidence domains: Objective Behavioral Evaluation / Perceptual Quality Assessment. Availability and scoring eligibility are reported separately.
+
+ranking_note: Compare only complete cards within the same mode, rubric and registry; independent calibration remains pending.
+
+visual_protocol: 2026-09-19.game-rubric-v1; game-specific M/D/V/A, GT and asset images, continuous per-item attainment
+
+
+Objective Behavioral Evaluation / 客观行为评测: 19.327 / 85.000 available points. This is the objective contribution, not a complete composite score.
+
+## Hierarchical scorecard (2026-09-19.mode1-vlm1)
+
+composite_score=not_measured; assessment_status=objective_only; headline_ceiling=100.000; currently_reachable_ceiling=80.143; measured_weight_share=0.801; ranking_eligible=no; strict.resolved=no; eligibility=failed
+
+The evaluator did not obtain a reading for every applicable source, so no complete composite is printed; the objective contribution above remains usable when measured. Unmeasured:
+- `numeric_contract` (mode1_fixed_axes/numeric_contract): no per-frame numeric samples were recorded for the candidate run
+- `item:task_visual` (visual_experience/task_visual): Game VLM requires --visual-judge vlm and an enabled engine
+
+The weighted total is diagnostic. The strict mode profile remains conjunctive; a visual or content score cannot compensate for a false clear, missing mechanic, or repair regression.
+
+| category | weight | score / 100 | measured share |
+|---|---:|---:|---:|
+| Mechanics and task fulfillment | 27% | 22.726 | 0.818 |
+| Interface, content, and asset realization | 42% | 12.143 | 1.000 |
+| Causal playability and demonstrations | 13% | 51.515 | 1.000 |
+| Candidate-authored design contract | 2% | 50.000 | 1.000 |
+| Game-specific visual quality | 15% | incomplete | 0.000 |
+
+## Independent feature demonstrations
+
+Action-caused coverage: 1/3. Every segment starts cold; repeated checks count once. Full clear is optional.
+
+| Demonstration | Status | Task-feature coverage / 100 | Action-caused checks |
+|---|---|---:|---|
+| cycle_build_menu | measured | 0.0 |  |
+| demolish_building | measured | 0.0 |  |
+| end_turn_rival_act | measured | 0.0 |  |
+| full_match_victory | measured | 33.333 | campaign_clear |
+| movement_cursor | measured | 0.0 |  |
+| no_input_defeat | measured | 0.0 |  |
+| place_mine | measured | 0.0 |  |
+
+Missing: economy_resolution, successful_attack
+
+## Perceptual Quality Assessment / 感知质量评审
+
+GameCraft-adapted judge; no GT reference images or submitter description. Experimental: not independently validated on this benchmark.
+
+Status: inconclusive; visual credit: not measured
+
+| Requirement / dimension | Aggregation | Credit / 100 |
+|---|---|---:|
+
+Game VLM requires --visual-judge vlm and an enabled engine
+
+behavior_score_lo=0.611 behavior_score_hi=0.611 coverage=1.000 denom=6.000 comparable=yes (diagnostic; strict ranking uses resolved)
+
+## Eligibility
+
+| id | verdict | detail |
+|---|---|---|
+| `layout` | passed | project.godot resolved |
+| `gdd` | passed | GDD.md present (13387 chars) |
+| `authored_gdd_quality` | passed | self-contained design, failure/recovery, progression, and acceptance present |
+| `authored_gdd_interface` | failed | extended/analog controls are mentioned without stable ids |
+| `brief_gdd_grounding` | passed | authored GDD retains every reference-conditioned control obligation |
+| `rubric_interface` | failed | required group gb_enemy is absent; required group gb_interactive is absent; required group gb_goal is absent |
+| `ops_present` | passed | feature demonstration protocol: /root/autodl-tmp/CodingBenchmark-server-results/fast_run_20260924_parallel/brief/twin_holds/cells/twin_holds__brief__claude__minimax-m3/submission/demos.json |
+| `ops_valid` | passed | 7 independent input-only demonstrations accepted |
+| `ops_not_idle` | passed | each demonstration contains player input |
+| `interface` | passed | Interface v2 conformance PASS |
+| `anti_grant_static` | passed | no blocking harness-flag on the load graph (3 files, authenticity=no_harness_flag_in_gameplay) |
+| `auto_win_ready` | passed | no declared-victory change in _ready |
+| `health_writable` | passed | 2 decreasing write(s) to declared `food` on its owner (0 owner, 2 unattributed) |
+| `verifier_profile_complete` | passed | brief emitted all 18 strict profile checks |
+
+## Behavioral verification
+
+| id | verdict | detail |
+|---|---|---|
+| `mechanic_trace` | passed | graded feature coverage: 1/3 |
+| `causal_witness` | passed | graded action-caused feature coverage: 1/3; whole-game clear not required |
+| `demonstrations_complete` | failed | incomplete feature coverage or no action-caused evidence: economy_resolution, successful_attack |
+| `null_no_win` | passed | idle stop_reason=budget_frames |
+| `extended_mash_no_win` | passed | mash of confirm_build, cycle_build, demolish, end_turn from frame 0 did not win |
+| `anti_grant_diff` | passed | idle grant-probe and ops pair did not diverge in the grant direction (flagged-wins / env-loses, or flagged much earlier) |
+
+## Reference fidelity (not part of resolved)
+
+| id | verdict | detail |
+|---|---|---|
+| `task_visual` | inconclusive | Game VLM requires --visual-judge vlm and an enabled engine |
+| `reproduction` | passed | evaluate_project objective_lo=0.507 |
+
+## Limits of the anti-grant gate
+
+- concatenated or computed flag names (`'--' + 'gb-route-plan'`)
+- PackedByteArray / encrypted predicates
+- checking that a plan JSON exists on disk
+- a node probe whose autoload name is computed rather than spelled out, or which recognises the evaluator by something other than its autoload names
+- unconditional invulnerability or zero damage with no harness test
+- a victory predicate that is simply always true
+- timed auto-clear outside the idle grant-probe budget (1100 frames)
+- spawn overlapping gb_goal without an immediate scene change
+- a grant in a file the project.godot load graph does not reach
+- C# using/class references without res:// are not followed
+- a _ready auto-win whose victory scene is not in gb_levels.json
+- concessions (vendored probe GB_ASSET_SAMPLING, acknowledged GB_ROUTE_* reads) do not fail ScanReport.ok or anti_grant_static
+- mash holds declared extras from frame 0; a delayed is_action_just_pressed win is missed
+- mash at spawn misses a win button that requires being somewhere else first
+- mash budget is 1100 frames, same gap as the idle grant probe
+
+A passing static scan is not a proof of honesty. See `docs/reference/TASKGEN_HARNESS.md` §5.
+

minimax/mode1/twin_holds/evaluation/reproduction/capture_manifest.json

@@ -0,0 +1,110 @@
+{
+  "anchor_frame": "",
+  "anchor_requested": false,
+  "extended_actions": [
+    {
+      "id": "confirm_build",
+      "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+    },
+    {
+      "id": "cycle_build",
+      "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+    },
+    {
+      "id": "demolish",
+      "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+    },
+    {
+      "id": "end_turn",
+      "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+    }
+  ],
+  "frame_paths": [
+    "/root/autodl-tmp/CodingBenchmark-server-results/fast_run_20260924_parallel/brief/twin_holds/cells/twin_holds__brief__claude__minimax-m3/evaluation/reproduction/captures/game__f00000042.png",
+    "/root/autodl-tmp/CodingBenchmark-server-results/fast_run_20260924_parallel/brief/twin_holds/cells/twin_holds__brief__claude__minimax-m3/evaluation/reproduction/captures/game__f00000072.png",
+    "/root/autodl-tmp/CodingBenchmark-server-results/fast_run_20260924_parallel/brief/twin_holds/cells/twin_holds__brief__claude__minimax-m3/evaluation/reproduction/captures/game__f00000108.png"
+  ],
+  "interface_version": 2,
+  "level_frame_paths": {},
+  "mode": "X_CPU",
+  "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+  "notes": [],
+  "probe_reported": true,
+  "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0",
+  "visual_evidence": {
+    "candidate_evaluator_capture": {
+      "capture_kind": "frame_sequence",
+      "frames": [
+        {
+          "frame_index": 42,
+          "path": "captures/game__f00000042.png",
+          "source_frame": "f00000042.png",
+          "timestamp_seconds": 1.4
+        },
+        {
+          "frame_index": 72,
+          "path": "captures/game__f00000072.png",
+          "source_frame": "f00000072.png",
+          "timestamp_seconds": 2.4
+        },
+        {
+          "frame_index": 108,
+          "path": "captures/game__f00000108.png",
+          "source_frame": "f00000108.png",
+          "timestamp_seconds": 3.6
+        }
+      ],
+      "is_reference_input": false,
+      "level": "res://scenes/main.tscn",
+      "mode": "X",
+      "owner": "evaluator",
+      "provenance": {
+        "argv": [
+          "xvfb-run",
+          "-a",
+          "-s",
+          "-screen 0 1280x720x24",
+          "/opt/godot451-bin/godot",
+          "--rendering-driver",
+          "opengl3",
+          "--path",
+          "/data2/gamebench/scratch/gb_ocard_scratch/game-frames-a39a3cda",
+          "res://scenes/main.tscn",
+          "--quit-after",
+          "120",
+          "--fixed-fps",
+          "30",
+          "--write-movie",
+          "/data2/gamebench/scratch/gb_ocard_scratch/frames-5629acae/f.png"
+        ],
+        "cwd": "/data2/gamebench/scratch/gb_ocard_scratch/game-frames-a39a3cda",
+        "extra_hashes": {},
+        "outcome": "reported",
+        "returncode": 0,
+        "seconds": 103.902,
+        "stderr_tail": "",
+        "tool": "godot",
+        "tool_path": "/opt/godot451-bin/godot",
+        "tool_sha256": "db07cae7de644278a1884d4552bdf2bca3f5d30131b18faf3a0c4d730080b199"
+      },
+      "role": "candidate_evaluator_capture",
+      "subject": "candidate_submission",
+      "timeline": {
+        "basis": "frame_index / fixed_fps from evaluator capture start",
+        "fixed_fps": 30.0,
+        "frames_generated": 120
+      },
+      "video_file": null,
+      "video_status": "not_recorded"
+    },
+    "reference_input": {
+      "artifact_status": "separate_task_material",
+      "is_candidate_evidence": false,
+      "note": "Reference gameplay belongs to the model-facing task package; it is not evidence captured from this candidate submission.",
+      "owner": "benchmark_task",
+      "role": "reference_input",
+      "subject": "reference_game"
+    },
+    "schema": "gamebench.candidate-visual-evidence.v1"
+  }
+}

minimax/mode1/twin_holds/evaluation/reproduction/card.json

@@ -0,0 +1,1374 @@
+{
+  "ceiling": {
+    "fidelity_subset": 1.0,
+    "note": "Cross-tier comparison uses normalised_lo only. A partial card normalises against `restricted`, computed over the channels that reported; mixing that numerator with `total` produced 1.011 once.",
+    "restricted": 1.0,
+    "total": 1.0,
+    "used_for_normalisation": 1.0
+  },
+  "channels": [
+    {
+      "by_verdict": {
+        "passed": 1.0
+      },
+      "channel": "O1",
+      "coverage": 1.0,
+      "denominator": 1.0,
+      "hi": 1.0,
+      "lo": 1.0,
+      "low_coverage": false,
+      "measured": true,
+      "mode_used": "X_CPU",
+      "name": "runnable",
+      "note": "cold_import=passed; boots=passed; draws_nontrivial=passed; responds_to_input=passed",
+      "weight": 0.063,
+      "width": 0.0
+    },
+    {
+      "by_verdict": {
+        "passed": 1.0
+      },
+      "channel": "O2",
+      "coverage": 1.0,
+      "denominator": 1.0,
+      "hi": 1.0,
+      "lo": 1.0,
+      "low_coverage": false,
+      "measured": true,
+      "mode_used": "H",
+      "name": "interface_conformance",
+      "note": "groups=passed; actions=passed; levels=passed; endings=passed; optional_fields_valid=passed; task_required_observables=passed",
+      "weight": 0.072,
+      "width": 0.0
+    },
+    {
+      "by_verdict": {
+        "exempt": 2.0,
+        "failed": 1.0,
+        "passed": 5.0
+      },
+      "channel": "O3",
+      "coverage": 1.0,
+      "denominator": 8.0,
+      "hi": 0.875,
+      "lo": 0.625,
+      "low_coverage": false,
+      "measured": true,
+      "mode_used": "H",
+      "name": "mechanic_fidelity",
+      "note": "universal assertions U1, U2, U3, U4, U5, U6, U7 over 1 levels; H-class 3 item(s) for task_id='twin_holds' 2 item(s) genre-exempt (weight 2.000); see by_verdict.",
+      "weight": 0.18,
+      "width": 0.25
+    },
+    {
+      "by_verdict": {
+        "passed": 1.0,
+        "unobservable": 2.0
+      },
+      "channel": "O4",
+      "coverage": 1.0,
+      "denominator": 1.0,
+      "hi": 1.0,
+      "lo": 1.0,
+      "low_coverage": false,
+      "measured": true,
+      "mode_used": "H",
+      "name": "level_topology",
+      "note": "declared-vs-built consistency (brief mode): 1 declared, 1 built, gdd states no count",
+      "weight": 0.099,
+      "width": 0.0
+    },
+    {
+      "by_verdict": {
+        "skipped": 2.0,
+        "unobservable": 1.0
+      },
+      "channel": "O5",
+      "coverage": 0.0,
+      "denominator": 2.0,
+      "hi": 1.0,
+      "lo": 0.0,
+      "low_coverage": true,
+      "measured": true,
+      "mode_used": "H",
+      "name": "spatial_ordinal",
+      "note": "collectible ordering, normalised goal distance and entity dispersion over 1 levels against expectations frozen 2026-08-23T10:43:04+00:00",
+      "weight": 0.081,
+      "width": 1.0
+    },
+    {
+      "by_verdict": {
+        "failed": 37.0
+      },
+      "channel": "O6",
+      "coverage": 1.0,
+      "denominator": 37.0,
+      "hi": 0.0,
+      "lo": 0.0,
+      "low_coverage": false,
+      "measured": true,
+      "mode_used": "H",
+      "name": "asset_use",
+      "note": "0/37 supplied assets observed in the engine",
+      "weight": 0.063,
+      "width": 0.0
+    },
+    {
+      "by_verdict": {
+        "failed": 1.0
+      },
+      "channel": "O7",
+      "coverage": 1.0,
+      "denominator": 1.0,
+      "hi": 0.0,
+      "lo": 0.0,
+      "low_coverage": false,
+      "measured": true,
+      "mode_used": "H",
+      "name": "authored_clear",
+      "note": "L5 authored clear over 1 RT-1-passed route(s)",
+      "weight": 0.126,
+      "width": 0.0
+    },
+    {
+      "by_verdict": {
+        "unobservable": 1.0
+      },
+      "channel": "O9",
+      "coverage": 0.0,
+      "denominator": 0.0,
+      "hi": 0.0,
+      "lo": 0.0,
+      "low_coverage": true,
+      "measured": false,
+      "mode_used": "X_CPU",
+      "name": "anchor_composition",
+      "note": "task has no O9 expectation",
+      "weight": 0.09,
+      "width": 0.0
+    },
+    {
+      "by_verdict": {
+        "unobservable": 1.0
+      },
+      "channel": "O8",
+      "coverage": 0.0,
+      "denominator": 0.0,
+      "hi": 0.0,
+      "lo": 0.0,
+      "low_coverage": true,
+      "measured": false,
+      "mode_used": "X_CPU",
+      "name": "gui_agent_routes",
+      "note": "excluded from the main card; independent external-agent routes consume the same levels/actions SubmissionInterface",
+      "weight": 0.126,
+      "width": 0.0
+    }
+  ],
+  "flags": [
+    "partial-card(0.76 of registered weight)",
+    "exempt:2(w=2.000)",
+    "zero:O5",
+    "zero:O6",
+    "zero:O7",
+    "failed:O3",
+    "failed:O6",
+    "failed:O7"
+  ],
+  "generated_at": "2026-09-24T06:05:36+00:00",
+  "main_table_eligible": false,
+  "main_table_refusal": "partial-card: only 76% of the registered O-card weight produced a reading ( reported nothing). Its normalised score has a different denominator from a complete card and the two must not share a column.",
+  "modality": "Mv",
+  "notes": [
+    "liveness deadline derived from this project's own structural pass: 57.8 ms/frame median, 57.8 ms/frame worst",
+    "liveness sweep: 5 live / 1 dead against the 420-frame idle control truth/L0#idle420",
+    "replayed 5 certified route(s) from a suite reviewed by 'project_owner' on 2026-09-09",
+    "only 76% of registered O-card weight produced any reading (O1, O2, O3, O4, O5, O6, O7); coverage=0.88 describes those channels alone and must not be read as coverage of the card.",
+    "2 genre-exempt item(s) (weight 2.000) stay in the denominator with lo=0/hi=full; interval width records the exemption. See exempt_items."
+  ],
+  "ocard": {
+    "by_verdict": {
+      "exempt": 2.0,
+      "failed": 39.0,
+      "passed": 8.0,
+      "skipped": 2.0,
+      "unobservable": 3.0
+    },
+    "coverage": 0.881579,
+    "denominator": 0.684,
+    "hi": 0.690789,
+    "lo": 0.506579,
+    "low_coverage": false,
+    "width": 0.184211
+  },
+  "provenance": {
+    "bot_skip_reason": null,
+    "channel_items": {
+      "O1": [
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "cold import clean (1 warnings)",
+          "id": "O1/cold_import",
+          "verdict": "passed",
+          "weight": 0.25
+        },
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "res://scenes/main.tscn reached a playable state (the level was playable at load, no menu to get past)",
+          "id": "O1/boots",
+          "verdict": "passed",
+          "weight": 0.25
+        },
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "1888 colours, 54% dominant, edge energy 7.8",
+          "id": "O1/draws_nontrivial",
+          "verdict": "passed",
+          "weight": 0.25
+        },
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "5/6 declared actions move the state signature away from the matched idle control (gb_left, gb_right, gb_up, gb_jump, gb_action)",
+          "id": "O1/responds_to_input",
+          "verdict": "passed",
+          "weight": 0.25
+        }
+      ],
+      "O2": [
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "fixed group vocabulary scanned; non-player categories may be empty",
+          "id": "O2/groups",
+          "verdict": "passed",
+          "weight": 0.2
+        },
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "all eight fixed aliases have an InputMap event and appear in source",
+          "id": "O2/actions",
+          "verdict": "passed",
+          "weight": 0.25
+        },
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "1 independently addressable level(s)",
+          "id": "O2/levels",
+          "verdict": "passed",
+          "weight": 0.2
+        },
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "success ending vocabulary is declared and distinguishes success from failure",
+          "id": "O2/endings",
+          "verdict": "passed",
+          "weight": 0.2
+        },
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "all declared optional fields are valid",
+          "id": "O2/optional_fields_valid",
+          "verdict": "passed",
+          "weight": 0.05
+        },
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "no task-conditional observables are registered",
+          "id": "O2/task_required_observables",
+          "verdict": "passed",
+          "weight": 0.1
+        }
+      ],
+      "O3": [
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "1 nodes tagged gb_player in res://scenes/main.tscn; public task contract permits player counts [1]",
+          "id": "U1/single_player[0]",
+          "verdict": "passed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "5/6 declared actions move the state signature away from the matched idle control (gb_left, gb_right, gb_up, gb_jump, gb_action)",
+          "id": "U2/actions_live",
+          "verdict": "passed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "genre_inapplicable (twin_holds): the frozen reference declares no gb_collectible in any level, so this is not one of this game's mechanics. measured: res://scenes/main.tscn declares no gb_collectible: there is nothing to pick up, which is this level not implementing collection rather than a reading the evaluator could not take",
+          "id": "U3/collect_removes_one[0]",
+          "verdict": "exempt",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "genre_inapplicable (twin_holds): gb_goal marks the rival castle hex; the level ends on capture, not overlap (GOLD_TRIAGE #4). measured: res://scenes/main.tscn declares no gb_goal: there is no win condition to reach, which is this level not implementing progression rather than a reading the evaluator could not take",
+          "id": "U4/goal_changes_scene[0]",
+          "verdict": "exempt",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "a zero-input control run stopped with `budget_frames`",
+          "id": "U5/no_input_no_win",
+          "verdict": "passed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "res://scenes/main.tscn was entered directly and the engine confirmed it was running it",
+          "id": "U6/levels_enterable[0]",
+          "verdict": "passed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "success-only ending scene(s): res://scenes/victory.tscn",
+          "id": "U7/distinct_endings",
+          "verdict": "passed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "failed on reached, used_required (stop_reason=ops_exhausted). segment_clean and the applicable required-device gate are pass/fail with no partial credit: they are anti-bypass gates, not a measure of progress.",
+          "id": "O3/H/L2/twin_holds/C1/score_route",
+          "verdict": "failed",
+          "weight": 0.3333333333333333
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "failed on reached, segment_clean, used_required (stop_reason=ops_exhausted). segment_clean and the applicable required-device gate are pass/fail with no partial credit: they are anti-bypass gates, not a measure of progress.",
+          "id": "O3/H/L3/twin_holds/C2/castle_capture_route",
+          "verdict": "failed",
+          "weight": 0.3333333333333333
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "failed on reached, segment_clean, used_required (stop_reason=ops_exhausted). segment_clean and the applicable required-device gate are pass/fail with no partial credit: they are anti-bypass gates, not a measure of progress.",
+          "id": "O3/H/L4/twin_holds/C3/advanced_score_route",
+          "verdict": "failed",
+          "weight": 0.3333333333333333
+        }
+      ],
+      "O4": [
+        {
+          "attribution": null,
+          "credit": 1.0,
+          "detail": "1 level(s) declared in gb_levels.json, 1 cold-loaded to a playable state on the declared scene",
+          "id": "E1/manifest_vs_built",
+          "verdict": "passed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "the authored GDD states no explicit level/zone/stage count to hold the manifest to; only manifest-vs-built is read",
+          "id": "E1/gdd_vs_manifest",
+          "verdict": "unobservable",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "no replay of the submission's own tape was available, so traversal of the declared levels was not read",
+          "id": "E4/declared_progression",
+          "verdict": "unobservable",
+          "weight": 1.0
+        }
+      ],
+      "O5": [
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "the reference has fewer than two levels, so there is no ordering between levels to reproduce",
+          "id": "E2/collectible_order",
+          "verdict": "unobservable",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "no normalised spawn-to-goal distance could be read from this level: it has no gb_goal, no gb_player, or degenerate bounds",
+          "id": "E3/goal_distance[0]",
+          "verdict": "skipped",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "no normalised gb_interactive positions were read from this level",
+          "id": "E5/dispersion[0/gb_interactive]",
+          "verdict": "skipped",
+          "weight": 1.0
+        }
+      ],
+      "O6": [
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/audio/ambience.ogg]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/blue/building_barracks_blue.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/blue/building_castle_blue.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/blue/building_church_blue.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/blue/building_home_A_blue.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/blue/building_home_B_blue.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/blue/building_lumbermill_blue.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/blue/building_market_blue.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/blue/building_mine_blue.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/blue/building_tavern_blue.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/blue/building_tower_base_blue.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/blue/building_well_blue.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/blue/building_windmill_blue.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/blue/hexagons_medieval.png]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/green/building_castle_green.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/green/hexagons_medieval.png]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/neutral/building_dirt.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/buildings/neutral/hexagons_medieval.png]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/nature/hexagons_medieval.png]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/nature/hill_single_A.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/nature/hills_A_trees.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/nature/mountain_A_grass_trees.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/nature/mountain_B_grass.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/nature/rock_single_A.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/nature/trees_A_medium.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/nature/trees_B_small.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/props/flag_blue.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/props/flag_green.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/props/flag_red.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/props/hexagons_medieval.png]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/tiles/hex_grass.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/tiles/hex_water.gltf]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/models/tiles/hexagons_medieval.png]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/ui/button_focus.png]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/ui/button_normal.png]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/ui/button_pressed.png]",
+          "verdict": "failed",
+          "weight": 1.0
+        },
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "supplied but never loaded at runtime",
+          "id": "O6/uses[res://assets/ui/panel_frame.png]",
+          "verdict": "failed",
+          "weight": 1.0
+        }
+      ],
+      "O7": [
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "failed on reached, used_required (stop_reason=ops_exhausted). segment_clean and the applicable required-device gate are pass/fail with no partial credit: they are anti-bypass gates, not a measure of progress.",
+          "id": "O7/L5/twin_holds/L5/whole_game_clear",
+          "verdict": "failed",
+          "weight": 1.0
+        }
+      ],
+      "O8": [
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "L0-L5 route pass rates from the route runner, with a per-game op cap derived from measured frame cost; not available here",
+          "id": "O8/not_implemented",
+          "verdict": "unobservable",
+          "weight": 1.0
+        }
+      ],
+      "O9": [
+        {
+          "attribution": null,
+          "credit": 0.0,
+          "detail": "this task has no frozen anchor composition expectation",
+          "id": "O9/anchor_composition",
+          "verdict": "unobservable",
+          "weight": 1.0
+        }
+      ]
+    },
+    "failed_axes": [
+      "O3",
+      "O6",
+      "O7"
+    ],
+    "harness": {
+      "contract_version": "1",
+      "gb_capture_probe.gd": "sha256:204e194e0071fa165070ba3af7b66553fa77122a973737bc293d25b4bc46d4df",
+      "gb_probe.gd": "sha256:39609c49a433ca5f7a05f2c3cc583693be12614dbe33b80cb5d115da290931c3",
+      "gb_route_driver.gd": "sha256:75645e1b23545ce6c18e3af0c519cd532956ac7cf4b9f50288da2403d39d6945",
+      "gb_truth_driver.gd": "sha256:7778e6190895bfeb10af8beffeec678ec3b03848b53a674477bcdf796a0c0742"
+    },
+    "interface": {
+      "extended_actions": [
+        {
+          "id": "confirm_build",
+          "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+        },
+        {
+          "id": "cycle_build",
+          "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+        },
+        {
+          "id": "demolish",
+          "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+        },
+        {
+          "id": "end_turn",
+          "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+        }
+      ],
+      "interface_version": 2,
+      "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+      "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0"
+    },
+    "interface_conformance": {
+      "checks": {
+        "actions": {
+          "bound": [
+            "gb_left",
+            "gb_right",
+            "gb_up",
+            "gb_down",
+            "gb_jump",
+            "gb_action",
+            "gb_pause",
+            "gb_reset"
+          ],
+          "detail": "all eight fixed aliases have an InputMap event and appear in source",
+          "missing": [],
+          "status": "pass",
+          "unread": []
+        },
+        "analog_axes": {
+          "declared": [],
+          "detail": "no analog axes declared",
+          "optional": true,
+          "status": "pass"
+        },
+        "anchor_camera": {
+          "address": null,
+          "detail": "optional anchor camera is a scene::NodePath address",
+          "optional": true,
+          "status": "pass"
+        },
+        "audio_buses": {
+          "buses": [],
+          "detail": "optional audio bus names are valid evaluator addresses",
+          "optional": true,
+          "status": "pass"
+        },
+        "device_ids": {
+          "declared": [],
+          "detail": "optional generic device ids map to submitted NodePaths",
+          "invalid": [],
+          "optional": true,
+          "status": "pass",
+          "vocabulary": [
+            "checkpoint",
+            "collectible",
+            "door",
+            "enemy",
+            "falling_platform",
+            "goal",
+            "hazard",
+            "interactive",
+            "movplat",
+            "pad",
+            "spring",
+            "tile"
+          ]
+        },
+        "endings": {
+          "declared": [
+            "defeat",
+            "victory"
+          ],
+          "detail": "success ending vocabulary is declared and distinguishes success from failure",
+          "discriminating_scenes": [
+            "res://scenes/victory.tscn"
+          ],
+          "failure": [
+            "defeat"
+          ],
+          "shared_scenes": [],
+          "status": "pass",
+          "success": [
+            "victory"
+          ]
+        },
+        "extended_actions": {
+          "declared": [
+            {
+              "id": "confirm_build",
+              "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+            },
+            {
+              "id": "cycle_build",
+              "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+            },
+            {
+              "id": "demolish",
+              "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+            },
+            {
+              "id": "end_turn",
+              "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+            }
+          ],
+          "detail": "4 declared extra action(s) on the load-bearing InputMap",
+          "optional": true,
+          "status": "pass"
+        },
+        "gb_levels": {
+          "accepted_form": "each levels entry must be a \"res://…\" scene-path string naming a shipped scene (no objects, no {id, name, scene} records)",
+          "detail": "1 independently addressable level(s)",
+          "levels": [
+            "res://scenes/main.tscn"
+          ],
+          "status": "pass",
+          "unresolved": []
+        },
+        "groups": {
+          "absent": [
+            "gb_collectible",
+            "gb_goal",
+            "gb_hazard",
+            "gb_enemy",
+            "gb_checkpoint",
+            "gb_door",
+            "gb_interactive"
+          ],
+          "detail": "fixed group vocabulary scanned; non-player categories may be empty",
+          "present": [
+            "gb_player"
+          ],
+          "required_player": "present",
+          "status": "pass"
+        },
+        "level_clear": {
+          "detail": "level_clear is optional; default is gb_goal transition",
+          "findings": [],
+          "optional": true,
+          "predicate": "",
+          "status": "pass"
+        },
+        "level_entry": {
+          "detail": "optional level-entry address is valid",
+          "optional": true,
+          "status": "pass"
+        },
+        "numeric": {
+          "accepted_form": "bare property name (^[A-Za-z_]\\w*$) of an int/float readable on an autoload, the current scene root, a gb_* node, or a run/state/run_state/game object held by one",
+          "detail": "generic numeric mapping is valid; every value is a bare property name the route driver reads with Object.get() on autoloads, the current scene root, gb_* nodes and their run/state/run_state/game holders",
+          "mapping": {
+            "health": "food",
+            "progress": "round",
+            "score": "gold"
+          },
+          "optional": true,
+          "status": "pass",
+          "values": {
+            "health": {
+              "declared_in_source": true,
+              "driver_resolves": true,
+              "form": "property_name",
+              "note": "read verbatim by Object.get() on autoloads / current scene root / gb_* nodes / run|state|run_state|game holders",
+              "property": "food",
+              "value": "food"
+            },
+            "progress": {
+              "declared_in_source": true,
+              "driver_resolves": true,
+              "form": "property_name",
+              "note": "read verbatim by Object.get() on autoloads / current scene root / gb_* nodes / run|state|run_state|game holders",
+              "property": "round",
+              "value": "round"
+            },
+            "score": {
+              "declared_in_source": true,
+              "driver_resolves": true,
+              "form": "property_name",
+              "note": "read verbatim by Object.get() on autoloads / current scene root / gb_* nodes / run|state|run_state|game holders",
+              "property": "gold",
+              "value": "gold"
+            }
+          }
+        },
+        "project_godot": {
+          "detail": "project.godot exists",
+          "status": "pass"
+        }
+      },
+      "missing": [],
+      "project": "/root/autodl-tmp/CodingBenchmark-server-results/fast_run_20260924_parallel/brief/twin_holds/cells/twin_holds__brief__claude__minimax-m3/submission/game",
+      "schema_version": 2,
+      "submission_obligation_excludes": [
+        "observe()",
+        "observe().kind",
+        "Agent Bridge"
+      ],
+      "verdict": "PASS"
+    },
+    "pass_a_seconds": 574.0,
+    "pass_b_seconds": 179.6,
+    "render_mode": "X_CPU",
+    "route_readings": [
+      {
+        "clean_clear": true,
+        "deaths": 0,
+        "detail": "",
+        "devices_required": [],
+        "devices_unresolved": [],
+        "devices_used": [],
+        "errors": [],
+        "frames": 420,
+        "frames_note": "frames is the trace row count; physics_frames_at_end keeps counting while the tree is paused for the agent and is not the game's clock; process_frames_at_end is Engine.get_process_frames() and can diverge from frames when --fixed-fps is off",
+        "goal_frame": -1,
+        "idle_baseline_id": "",
+        "injects": 0,
+        "interface": {
+          "extended_actions": [
+            {
+              "id": "confirm_build",
+              "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+            },
+            {
+              "id": "cycle_build",
+              "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+            },
+            {
+              "id": "demolish",
+              "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+            },
+            {
+              "id": "end_turn",
+              "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+            }
+          ],
+          "interface_version": 2,
+          "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+          "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0"
+        },
+        "invariants_failed": [],
+        "milestone_frames": {},
+        "milestones_reached": [],
+        "missing_groups": [],
+        "mode": "H",
+        "numeric_resolution": {
+          "health": "none",
+          "progress": "none",
+          "score": "none"
+        },
+        "observations_reached": [],
+        "observations_triggered": [],
+        "physics_frames_at_end": 427,
+        "process_frames_at_end": 426,
+        "reached": true,
+        "refusals": [],
+        "replans": 0,
+        "route_id": "twin_holds/L0/no_input_no_clear",
+        "segment_clean": true,
+        "stall_frames": 420,
+        "state_delta": 0.0,
+        "steps": 1,
+        "stop_reason": "budget_frames",
+        "success_before_all_levels": false,
+        "tier": 0,
+        "used_required": true,
+        "wall_ms": 4695.2
+      },
+      {
+        "clean_clear": false,
+        "deaths": 0,
+        "detail": "",
+        "devices_required": [
+          "tile#25"
+        ],
+        "devices_unresolved": [
+          "tile#25"
+        ],
+        "devices_used": [],
+        "errors": [],
+        "frames": 1299,
+        "frames_note": "frames is the trace row count; physics_frames_at_end keeps counting while the tree is paused for the agent and is not the game's clock; process_frames_at_end is Engine.get_process_frames() and can diverge from frames when --fixed-fps is off",
+        "goal_frame": -1,
+        "idle_baseline_id": "",
+        "injects": 0,
+        "interface": {
+          "extended_actions": [
+            {
+              "id": "confirm_build",
+              "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+            },
+            {
+              "id": "cycle_build",
+              "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+            },
+            {
+              "id": "demolish",
+              "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+            },
+            {
+              "id": "end_turn",
+              "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+            }
+          ],
+          "interface_version": 2,
+          "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+          "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0"
+        },
+        "invariants_failed": [],
+        "milestone_frames": {},
+        "milestones_reached": [],
+        "missing_groups": [
+          "numeric:score"
+        ],
+        "mode": "H",
+        "numeric_resolution": {
+          "health": "none",
+          "progress": "none",
+          "score": "none"
+        },
+        "observations_reached": [],
+        "observations_triggered": [],
+        "physics_frames_at_end": 1365,
+        "process_frames_at_end": 1364,
+        "reached": false,
+        "refusals": [],
+        "replans": 0,
+        "route_id": "twin_holds/C1/score_route",
+        "segment_clean": true,
+        "stall_frames": 1299,
+        "state_delta": 1.0,
+        "steps": 197,
+        "stop_reason": "ops_exhausted",
+        "success_before_all_levels": false,
+        "tier": 2,
+        "used_required": false,
+        "wall_ms": 7794.4
+      },
+      {
+        "clean_clear": false,
+        "deaths": 0,
+        "detail": "",
+        "devices_required": [
+          "tile#39"
+        ],
+        "devices_unresolved": [
+          "tile#39"
+        ],
+        "devices_used": [],
+        "errors": [
+          "inject.autoload_state could not be applied completely"
+        ],
+        "frames": 2717,
+        "frames_note": "frames is the trace row count; physics_frames_at_end keeps counting while the tree is paused for the agent and is not the game's clock; process_frames_at_end is Engine.get_process_frames() and can diverge from frames when --fixed-fps is off",
+        "goal_frame": -1,
+        "idle_baseline_id": "",
+        "injects": 0,
+        "interface": {
+          "extended_actions": [
+            {
+              "id": "confirm_build",
+              "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+            },
+            {
+              "id": "cycle_build",
+              "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+            },
+            {
+              "id": "demolish",
+              "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+            },
+            {
+              "id": "end_turn",
+              "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+            }
+          ],
+          "interface_version": 2,
+          "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+          "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0"
+        },
+        "invariants_failed": [],
+        "milestone_frames": {},
+        "milestones_reached": [],
+        "missing_groups": [
+          "numeric:score"
+        ],
+        "mode": "H",
+        "numeric_resolution": {
+          "health": "none",
+          "progress": "none",
+          "score": "none"
+        },
+        "observations_reached": [],
+        "observations_triggered": [],
+        "physics_frames_at_end": 2783,
+        "process_frames_at_end": 2782,
+        "reached": false,
+        "refusals": [],
+        "replans": 0,
+        "route_id": "twin_holds/C2/castle_capture_route",
+        "segment_clean": false,
+        "stall_frames": 2717,
+        "state_delta": 1.0,
+        "steps": 745,
+        "stop_reason": "ops_exhausted",
+        "success_before_all_levels": false,
+        "tier": 3,
+        "used_required": false,
+        "wall_ms": 11786.7
+      },
+      {
+        "clean_clear": false,
+        "deaths": 0,
+        "detail": "",
+        "devices_required": [
+          "tile#56"
+        ],
+        "devices_unresolved": [
+          "tile#56"
+        ],
+        "devices_used": [],
+        "errors": [
+          "inject.autoload_state could not be applied completely"
+        ],
+        "frames": 4775,
+        "frames_note": "frames is the trace row count; physics_frames_at_end keeps counting while the tree is paused for the agent and is not the game's clock; process_frames_at_end is Engine.get_process_frames() and can diverge from frames when --fixed-fps is off",
+        "goal_frame": -1,
+        "idle_baseline_id": "",
+        "injects": 0,
+        "interface": {
+          "extended_actions": [
+            {
+              "id": "confirm_build",
+              "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+            },
+            {
+              "id": "cycle_build",
+              "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+            },
+            {
+              "id": "demolish",
+              "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+            },
+            {
+              "id": "end_turn",
+              "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+            }
+          ],
+          "interface_version": 2,
+          "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+          "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0"
+        },
+        "invariants_failed": [],
+        "milestone_frames": {},
+        "milestones_reached": [],
+        "missing_groups": [
+          "numeric:score"
+        ],
+        "mode": "H",
+        "numeric_resolution": {
+          "health": "none",
+          "progress": "none",
+          "score": "none"
+        },
+        "observations_reached": [],
+        "observations_triggered": [],
+        "physics_frames_at_end": 4841,
+        "process_frames_at_end": 4840,
+        "reached": false,
+        "refusals": [],
+        "replans": 0,
+        "route_id": "twin_holds/C3/advanced_score_route",
+        "segment_clean": false,
+        "stall_frames": 4775,
+        "state_delta": 1.0,
+        "steps": 1475,
+        "stop_reason": "ops_exhausted",
+        "success_before_all_levels": false,
+        "tier": 4,
+        "used_required": false,
+        "wall_ms": 17106.8
+      },
+      {
+        "clean_clear": false,
+        "deaths": 0,
+        "detail": "",
+        "devices_required": [
+          "L1/tile#25",
+          "L2/tile#39",
+          "L3/tile#56"
+        ],
+        "devices_unresolved": [
+          "L1/tile#25",
+          "L2/tile#39",
+          "L3/tile#56"
+        ],
+        "devices_used": [],
+        "errors": [],
+        "frames": 8921,
+        "frames_note": "frames is the trace row count; physics_frames_at_end keeps counting while the tree is paused for the agent and is not the game's clock; process_frames_at_end is Engine.get_process_frames() and can diverge from frames when --fixed-fps is off",
+        "goal_frame": -1,
+        "idle_baseline_id": "",
+        "injects": 0,
+        "interface": {
+          "extended_actions": [
+            {
+              "id": "confirm_build",
+              "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+            },
+            {
+              "id": "cycle_build",
+              "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+            },
+            {
+              "id": "demolish",
+              "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+            },
+            {
+              "id": "end_turn",
+              "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+            }
+          ],
+          "interface_version": 2,
+          "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+          "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0"
+        },
+        "invariants_failed": [],
+        "milestone_frames": {},
+        "milestones_reached": [],
+        "missing_groups": [],
+        "mode": "H",
+        "numeric_resolution": {
+          "health": "none",
+          "progress": "none",
+          "score": "none"
+        },
+        "observations_reached": [],
+        "observations_triggered": [],
+        "physics_frames_at_end": 8928,
+        "process_frames_at_end": 8927,
+        "reached": false,
+        "refusals": [],
+        "replans": 0,
+        "route_id": "twin_holds/L5/whole_game_clear",
+        "segment_clean": true,
+        "stall_frames": 8921,
+        "state_delta": 1.0,
+        "steps": 2423,
+        "stop_reason": "ops_exhausted",
+        "success_before_all_levels": false,
+        "tier": 5,
+        "used_required": false,
+        "wall_ms": 17611.5
+      }
+    ]
+  },
+  "replay_reading": null,
+  "scard": null,
+  "scard_state": "absent",
+  "score": {
+    "combined_hi": 0.690789,
+    "combined_lo": 0.506579,
+    "coverage": 0.881579,
+    "measured_weight_share": 0.76,
+    "normalised_lo": 0.506579,
+    "objective_lo": 0.506579
+  },
+  "separately_reported": {
+    "exempt_count": 2,
+    "exempt_items": [
+      {
+        "channel": "O3",
+        "detail": "genre_inapplicable (twin_holds): the frozen reference declares no gb_collectible in any level, so this is not one of this game's mechanics. measured: res://scenes/main.tscn declares no gb_collectible: there is nothing to pick up, which is this level not implementing collection rather than a reading the evaluator could not take",
+        "family": "U3",
+        "id": "U3/collect_removes_one[0]",
+        "weight": 1.0
+      },
+      {
+        "channel": "O3",
+        "detail": "genre_inapplicable (twin_holds): gb_goal marks the rival castle hex; the level ends on capture, not overlap (GOLD_TRIAGE #4). measured: res://scenes/main.tscn declares no gb_goal: there is no win condition to reach, which is this level not implementing progression rather than a reading the evaluator could not take",
+        "family": "U4",
+        "id": "U4/goal_changes_scene[0]",
+        "weight": 1.0
+      }
+    ],
+    "exempt_weight": 2.0,
+    "inconclusive_rate": 0.0,
+    "shortcut_detected": false,
+    "shortcut_evidence": [],
+    "unattributable_error_rate": 0.0,
+    "unmeasurable_channels": [],
+    "unobservable_hit_rate": null
+  },
+  "submission": "game",
+  "task_id": "twin_holds",
+  "tier": "D2",
+  "weights_registry": "2026-09-04.0"
+}

minimax/mode1/twin_holds/evaluation/reproduction/env.json

@@ -0,0 +1,393 @@
+{
+  "asset_observation": {
+    "detail": "observed 3 live resource path(s) in an isolated pass",
+    "loaded_asset_digests": {
+      "res://gb_asset_probe.gd": "0ee55cd7e3c94e0ddecb695220fde12b150e5c8515433f8a0f99ae7a5a502b6d",
+      "res://scripts/main.gd": "b93964020c98a61360ea181da7e81e5dd96ab67d84eaaac6af2371cc79ecde2f",
+      "res://scripts/main_controller.gd": "b5e0615aafbd7a9f6033c69ec222d8f66952986171d591c8f0d0bbb98197a3c6"
+    },
+    "loaded_assets": [
+      "res://scripts/main.gd",
+      "res://scripts/main_controller.gd",
+      "res://gb_asset_probe.gd"
+    ],
+    "report_produced": true
+  },
+  "capture": {
+    "anchor_frame": "",
+    "anchor_requested": false,
+    "extended_actions": [
+      {
+        "id": "confirm_build",
+        "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+      },
+      {
+        "id": "cycle_build",
+        "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+      },
+      {
+        "id": "demolish",
+        "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+      },
+      {
+        "id": "end_turn",
+        "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+      }
+    ],
+    "frame_paths": [
+      "/root/autodl-tmp/CodingBenchmark-server-results/fast_run_20260924_parallel/brief/twin_holds/cells/twin_holds__brief__claude__minimax-m3/evaluation/reproduction/captures/game__f00000042.png",
+      "/root/autodl-tmp/CodingBenchmark-server-results/fast_run_20260924_parallel/brief/twin_holds/cells/twin_holds__brief__claude__minimax-m3/evaluation/reproduction/captures/game__f00000072.png",
+      "/root/autodl-tmp/CodingBenchmark-server-results/fast_run_20260924_parallel/brief/twin_holds/cells/twin_holds__brief__claude__minimax-m3/evaluation/reproduction/captures/game__f00000108.png"
+    ],
+    "interface_version": 2,
+    "level_frame_paths": {},
+    "mode": "X_CPU",
+    "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+    "notes": [],
+    "probe_reported": true,
+    "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0",
+    "visual_evidence": {
+      "candidate_evaluator_capture": {
+        "capture_kind": "frame_sequence",
+        "frames": [
+          {
+            "frame_index": 42,
+            "path": "captures/game__f00000042.png",
+            "source_frame": "f00000042.png",
+            "timestamp_seconds": 1.4
+          },
+          {
+            "frame_index": 72,
+            "path": "captures/game__f00000072.png",
+            "source_frame": "f00000072.png",
+            "timestamp_seconds": 2.4
+          },
+          {
+            "frame_index": 108,
+            "path": "captures/game__f00000108.png",
+            "source_frame": "f00000108.png",
+            "timestamp_seconds": 3.6
+          }
+        ],
+        "is_reference_input": false,
+        "level": "res://scenes/main.tscn",
+        "mode": "X",
+        "owner": "evaluator",
+        "provenance": {
+          "argv": [
+            "xvfb-run",
+            "-a",
+            "-s",
+            "-screen 0 1280x720x24",
+            "/opt/godot451-bin/godot",
+            "--rendering-driver",
+            "opengl3",
+            "--path",
+            "/data2/gamebench/scratch/gb_ocard_scratch/game-frames-a39a3cda",
+            "res://scenes/main.tscn",
+            "--quit-after",
+            "120",
+            "--fixed-fps",
+            "30",
+            "--write-movie",
+            "/data2/gamebench/scratch/gb_ocard_scratch/frames-5629acae/f.png"
+          ],
+          "cwd": "/data2/gamebench/scratch/gb_ocard_scratch/game-frames-a39a3cda",
+          "extra_hashes": {},
+          "outcome": "reported",
+          "returncode": 0,
+          "seconds": 103.902,
+          "stderr_tail": "",
+          "tool": "godot",
+          "tool_path": "/opt/godot451-bin/godot",
+          "tool_sha256": "db07cae7de644278a1884d4552bdf2bca3f5d30131b18faf3a0c4d730080b199"
+        },
+        "role": "candidate_evaluator_capture",
+        "subject": "candidate_submission",
+        "timeline": {
+          "basis": "frame_index / fixed_fps from evaluator capture start",
+          "fixed_fps": 30.0,
+          "frames_generated": 120
+        },
+        "video_file": null,
+        "video_status": "not_recorded"
+      },
+      "reference_input": {
+        "artifact_status": "separate_task_material",
+        "is_candidate_evidence": false,
+        "note": "Reference gameplay belongs to the model-facing task package; it is not evidence captured from this candidate submission.",
+        "owner": "benchmark_task",
+        "role": "reference_input",
+        "subject": "reference_game"
+      },
+      "schema": "gamebench.candidate-visual-evidence.v1"
+    }
+  },
+  "conformance": {
+    "checks": {
+      "actions": {
+        "bound": [
+          "gb_left",
+          "gb_right",
+          "gb_up",
+          "gb_down",
+          "gb_jump",
+          "gb_action",
+          "gb_pause",
+          "gb_reset"
+        ],
+        "detail": "all eight fixed aliases have an InputMap event and appear in source",
+        "missing": [],
+        "status": "pass",
+        "unread": []
+      },
+      "analog_axes": {
+        "declared": [],
+        "detail": "no analog axes declared",
+        "optional": true,
+        "status": "pass"
+      },
+      "anchor_camera": {
+        "address": null,
+        "detail": "optional anchor camera is a scene::NodePath address",
+        "optional": true,
+        "status": "pass"
+      },
+      "audio_buses": {
+        "buses": [],
+        "detail": "optional audio bus names are valid evaluator addresses",
+        "optional": true,
+        "status": "pass"
+      },
+      "device_ids": {
+        "declared": [],
+        "detail": "optional generic device ids map to submitted NodePaths",
+        "invalid": [],
+        "optional": true,
+        "status": "pass",
+        "vocabulary": [
+          "checkpoint",
+          "collectible",
+          "door",
+          "enemy",
+          "falling_platform",
+          "goal",
+          "hazard",
+          "interactive",
+          "movplat",
+          "pad",
+          "spring",
+          "tile"
+        ]
+      },
+      "endings": {
+        "declared": [
+          "defeat",
+          "victory"
+        ],
+        "detail": "success ending vocabulary is declared and distinguishes success from failure",
+        "discriminating_scenes": [
+          "res://scenes/victory.tscn"
+        ],
+        "failure": [
+          "defeat"
+        ],
+        "shared_scenes": [],
+        "status": "pass",
+        "success": [
+          "victory"
+        ]
+      },
+      "extended_actions": {
+        "declared": [
+          {
+            "id": "confirm_build",
+            "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+          },
+          {
+            "id": "cycle_build",
+            "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+          },
+          {
+            "id": "demolish",
+            "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+          },
+          {
+            "id": "end_turn",
+            "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+          }
+        ],
+        "detail": "4 declared extra action(s) on the load-bearing InputMap",
+        "optional": true,
+        "status": "pass"
+      },
+      "gb_levels": {
+        "accepted_form": "each levels entry must be a \"res://…\" scene-path string naming a shipped scene (no objects, no {id, name, scene} records)",
+        "detail": "1 independently addressable level(s)",
+        "levels": [
+          "res://scenes/main.tscn"
+        ],
+        "status": "pass",
+        "unresolved": []
+      },
+      "groups": {
+        "absent": [
+          "gb_collectible",
+          "gb_goal",
+          "gb_hazard",
+          "gb_enemy",
+          "gb_checkpoint",
+          "gb_door",
+          "gb_interactive"
+        ],
+        "detail": "fixed group vocabulary scanned; non-player categories may be empty",
+        "present": [
+          "gb_player"
+        ],
+        "required_player": "present",
+        "status": "pass"
+      },
+      "level_clear": {
+        "detail": "level_clear is optional; default is gb_goal transition",
+        "findings": [],
+        "optional": true,
+        "predicate": "",
+        "status": "pass"
+      },
+      "level_entry": {
+        "detail": "optional level-entry address is valid",
+        "optional": true,
+        "status": "pass"
+      },
+      "numeric": {
+        "accepted_form": "bare property name (^[A-Za-z_]\\w*$) of an int/float readable on an autoload, the current scene root, a gb_* node, or a run/state/run_state/game object held by one",
+        "detail": "generic numeric mapping is valid; every value is a bare property name the route driver reads with Object.get() on autoloads, the current scene root, gb_* nodes and their run/state/run_state/game holders",
+        "mapping": {
+          "health": "food",
+          "progress": "round",
+          "score": "gold"
+        },
+        "optional": true,
+        "status": "pass",
+        "values": {
+          "health": {
+            "declared_in_source": true,
+            "driver_resolves": true,
+            "form": "property_name",
+            "note": "read verbatim by Object.get() on autoloads / current scene root / gb_* nodes / run|state|run_state|game holders",
+            "property": "food",
+            "value": "food"
+          },
+          "progress": {
+            "declared_in_source": true,
+            "driver_resolves": true,
+            "form": "property_name",
+            "note": "read verbatim by Object.get() on autoloads / current scene root / gb_* nodes / run|state|run_state|game holders",
+            "property": "round",
+            "value": "round"
+          },
+          "score": {
+            "declared_in_source": true,
+            "driver_resolves": true,
+            "form": "property_name",
+            "note": "read verbatim by Object.get() on autoloads / current scene root / gb_* nodes / run|state|run_state|game holders",
+            "property": "gold",
+            "value": "gold"
+          }
+        }
+      },
+      "project_godot": {
+        "detail": "project.godot exists",
+        "status": "pass"
+      }
+    },
+    "missing": [],
+    "project": "/root/autodl-tmp/CodingBenchmark-server-results/fast_run_20260924_parallel/brief/twin_holds/cells/twin_holds__brief__claude__minimax-m3/submission/game",
+    "schema_version": 2,
+    "submission_obligation_excludes": [
+      "observe()",
+      "observe().kind",
+      "Agent Bridge"
+    ],
+    "verdict": "PASS"
+  },
+  "evalsys_commit": "11e2848deb22940d3840325de462f4c256e2c74d",
+  "expectations_generated_at": "2026-08-23T10:43:04+00:00",
+  "expectations_version": "2",
+  "godot_version": "4.5.1",
+  "harness": {
+    "contract_version": "1",
+    "gb_capture_probe.gd": "sha256:204e194e0071fa165070ba3af7b66553fa77122a973737bc293d25b4bc46d4df",
+    "gb_probe.gd": "sha256:39609c49a433ca5f7a05f2c3cc583693be12614dbe33b80cb5d115da290931c3",
+    "gb_route_driver.gd": "sha256:75645e1b23545ce6c18e3af0c519cd532956ac7cf4b9f50288da2403d39d6945",
+    "gb_truth_driver.gd": "sha256:7778e6190895bfeb10af8beffeec678ec3b03848b53a674477bcdf796a0c0742"
+  },
+  "interface": {
+    "extended_actions": [
+      {
+        "id": "confirm_build",
+        "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+      },
+      {
+        "id": "cycle_build",
+        "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+      },
+      {
+        "id": "demolish",
+        "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+      },
+      {
+        "id": "end_turn",
+        "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+      }
+    ],
+    "interface_version": 2,
+    "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+    "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0"
+  },
+  "modality": "Mv",
+  "mode": "X_CPU",
+  "reference_runtime_assets": {
+    "res://assets/audio/ambience.ogg": "35697cbfde073f92df06e828fe2c8676a1f98c07404193e6477ca9202213f80a",
+    "res://assets/models/buildings/blue/building_barracks_blue.gltf": "a4f3aaa7ebb33cf0d86a139941b350e580e3e1376e3ead99a5aaecb5de2f3817",
+    "res://assets/models/buildings/blue/building_castle_blue.gltf": "f23ede8641122d2226f02a5fdb4957dfedec99accf27240f07ef7fb312696759",
+    "res://assets/models/buildings/blue/building_church_blue.gltf": "ac2d9dfbdcffd1604e1522ddb26fe27c1feae2bc01bf90664c8044b22926867b",
+    "res://assets/models/buildings/blue/building_home_A_blue.gltf": "60b3568e086d586b491eed8168c46acb61bb9670fb2c296faa64929b18dfcbea",
+    "res://assets/models/buildings/blue/building_home_B_blue.gltf": "bf503f4edcd3a744a021ccdd5b9215f6c8934ef87fa44d95632cb653422797eb",
+    "res://assets/models/buildings/blue/building_lumbermill_blue.gltf": "bf6ff5811c7e4aac10dbd32d0cb26b2dda6e31b894726bf9d567371a0fb1f1b4",
+    "res://assets/models/buildings/blue/building_market_blue.gltf": "24874a4651dfff0d1aea31e15a32b9c35122958764c6f968ca55f286285ffbe9",
+    "res://assets/models/buildings/blue/building_mine_blue.gltf": "fe21db192f3ee7d40bd5f64a1d3782c6301b92a6676ca8a4d43e8c3e85ab7695",
+    "res://assets/models/buildings/blue/building_tavern_blue.gltf": "58a3573cf468f9c5f1d03a6da74fbc884f5f662ccddbfef2a10860abfe50109b",
+    "res://assets/models/buildings/blue/building_tower_base_blue.gltf": "60c9ea992b4dd584f57183a6c9b89619ff36eda2c81be71cefdb8704bdf84f4a",
+    "res://assets/models/buildings/blue/building_well_blue.gltf": "66305823322b2e6a7f1cb8d8bf0d91ed9b94b8d96f61d0f7af77bd0d09f6463e",
+    "res://assets/models/buildings/blue/building_windmill_blue.gltf": "a98412d4bcd016ce63f66b5d7193adf664867c3649829707b04586fb73e31f38",
+    "res://assets/models/buildings/blue/hexagons_medieval.png": "5301a06866ba3be68dcc14cc429bd6d0f69004b72884283715d7e233f2e9782e",
+    "res://assets/models/buildings/green/building_castle_green.gltf": "6739f651daf33b406c0abb0b91af49315059a0e70012bd01b77b30695f4bd98c",
+    "res://assets/models/buildings/green/hexagons_medieval.png": "5301a06866ba3be68dcc14cc429bd6d0f69004b72884283715d7e233f2e9782e",
+    "res://assets/models/buildings/neutral/building_dirt.gltf": "dd7d0f3e4740b0465615d69024a553f87314385998431c8555788bef354aed16",
+    "res://assets/models/buildings/neutral/hexagons_medieval.png": "5301a06866ba3be68dcc14cc429bd6d0f69004b72884283715d7e233f2e9782e",
+    "res://assets/models/nature/hexagons_medieval.png": "5301a06866ba3be68dcc14cc429bd6d0f69004b72884283715d7e233f2e9782e",
+    "res://assets/models/nature/hill_single_A.gltf": "00eb206cb0239257d05c107568bcf1fd6548d2f3da2cfb6ec8ecd41c377bbf06",
+    "res://assets/models/nature/hills_A_trees.gltf": "6e7e71bd37856c442dc7cf562a1cc1e9eb172050dc002d2d65c3fb5b90ae1d28",
+    "res://assets/models/nature/mountain_A_grass_trees.gltf": "defb29e0132afd4ec5cb33281efb890c9ed9603b5181471b31ff4f21d71c143a",
+    "res://assets/models/nature/mountain_B_grass.gltf": "4df9f7e2479b581061f3369e9da18de672e89dd896d7cb116d6078c823150afc",
+    "res://assets/models/nature/rock_single_A.gltf": "ddf307987fb18aa8e7594bb399de3943e497b50bc01aae0e7292a38418944915",
+    "res://assets/models/nature/trees_A_medium.gltf": "c330d7c775c50b7ccbc6ce7799e699fbdbbeb14a390315e85419dc0dfe5178a6",
+    "res://assets/models/nature/trees_B_small.gltf": "f9a491097a116f98a11022b1b1adb563946ba28751f9184d689a8019ef4d7e41",
+    "res://assets/models/props/flag_blue.gltf": "de33035f0e9945f8d4e6ca3242e18f4822227cc8cc3a1afd9f0d193419573f08",
+    "res://assets/models/props/flag_green.gltf": "3d3192dc48ae58179391adc15d38faadcc1680e2d6b8bab582da054cf01c82bd",
+    "res://assets/models/props/flag_red.gltf": "c1f76b1ca3d47a30a868500ecbe2844e89b75b4bebc7283f41bd37561aa58c1f",
+    "res://assets/models/props/hexagons_medieval.png": "5301a06866ba3be68dcc14cc429bd6d0f69004b72884283715d7e233f2e9782e",
+    "res://assets/models/tiles/hex_grass.gltf": "9197814431b5becf69b22bb7a20e340d63155c4aeddfb3a678c3f041ca585e41",
+    "res://assets/models/tiles/hex_water.gltf": "ccc5b423abcd43dfb562ef0fda1f518e24e0696364d6beb2f2da704da9d2acfa",
+    "res://assets/models/tiles/hexagons_medieval.png": "5301a06866ba3be68dcc14cc429bd6d0f69004b72884283715d7e233f2e9782e",
+    "res://assets/ui/button_focus.png": "f8d084a5caaae0599d0c5fbfd52385ce2932af2d1375180c5e771b0a2aa41e0a",
+    "res://assets/ui/button_normal.png": "3fc1d9bedf07da0f1a49bed143311e6edf3bebd9a937e95a6ab4b812c8448ec0",
+    "res://assets/ui/button_pressed.png": "424f7743b51666915429a32b76732aeea7079c053cabbe7874d826fed210dacc",
+    "res://assets/ui/panel_frame.png": "4239da5e604c18d0410847df936040dd821e80d85ad89efe02635d93286f5574"
+  },
+  "route_sha256": "f47e2bcb8a54c0f3e0de97b33fe68a540677c529b27e4afd3bf275bd97d29952",
+  "task_id": "twin_holds",
+  "tier": "D2",
+  "weights_registry_version": "2026-09-04.0"
+}

minimax/mode1/twin_holds/evaluation/reproduction/report.md

@@ -0,0 +1,63 @@
+# game — twin_holds @ D2/Mv
+
+Weight registry `2026-09-04.0`.
+
+## Score
+
+| | value |
+|---|---|
+| ranked score (`score_lo`) | **0.507** |
+| interval | [0.507, 0.691] |
+| Objective Behavioral Evaluation (objective only) | 0.507 |
+| normalised (`/ restricted ceiling, over the channels that reported = 1.000`) | 0.507 |
+| coverage | 0.88 |
+| share of registered weight actually measured | 0.76 |
+| flags | partial-card(0.76 of registered weight), exempt:2(w=2.000), zero:O5, zero:O6, zero:O7, failed:O3, failed:O6, failed:O7 |
+
+> **Not eligible for a cross-model table.** partial-card: only 76% of the registered Objective Behavioral Evaluation weight produced a reading ( reported nothing). Its normalised score has a different denominator from a complete card and the two must not share a column.
+
+The interval is 0.184 wide. That width is the part of the submission we did not manage to measure, and it is counted against the score rather than around it: ranking uses the low end, so a check that was skipped costs exactly as much as a check that failed.
+
+## Objective Behavioral Evaluation
+
+| channel | what it reads | w | score | coverage | mode | note |
+|---|---|---:|---|---:|:---:|---|
+| O1 | runnable | 0.063 | 1.000 | 1.00 | X_CPU | cold_import=passed; boots=passed; draws_nontrivial=passed; responds_to_input=passed |
+| O2 | interface_conformance | 0.072 | 1.000 | 1.00 | H | groups=passed; actions=passed; levels=passed; endings=passed; optional_fields_valid=passed; task_required_observables=passed |
+| O3 | mechanic_fidelity | 0.180 | 0.625 [0.625, 0.875] | 1.00 | H | universal assertions U1, U2, U3, U4, U5, U6, U7 over 1 levels; H-class 3 item(s) for task_id='twin_holds' 2 item(s) genre-exempt (weight 2.000); see by_verdict. |
+| O4 | level_topology | 0.099 | 1.000 | 1.00 | H | declared-vs-built consistency (brief mode): 1 declared, 1 built, gdd states no count |
+| O5 | spatial_ordinal | 0.081 | 0.000 [0.000, 1.000] | 0.00 ! | H | collectible ordering, normalised goal distance and entity dispersion over 1 levels against expectations frozen 2026-08-23T10:43:04+00:00 |
+| O6 | asset_use | 0.063 | 0.000 | 1.00 | H | 0/37 supplied assets observed in the engine |
+| O7 | authored_clear | 0.126 | 0.000 | 1.00 | H | L5 authored clear over 1 RT-1-passed route(s) |
+| O8 | gui_agent_routes | 0.126 | not measured | - | X_CPU | excluded from the main card; independent external-agent routes consume the same levels/actions SubmissionInterface |
+| O9 | anchor_composition | 0.090 | not measured | - | X_CPU | task has no O9 expectation |
+
+## Reported separately, never folded into the score
+
+| reading | value | why it is not in the score |
+|---|---|---|
+| `shortcut_detected` | False | A categorical fact, not a magnitude. Any number of points docked for it would be the wrong number. |
+| `unmeasurable_channels` | none | Distinguishes 'did not do it' from 'we could not read it'. |
+| `inconclusive_rate` | 0.000 | Our failures, not the submission's. Out of the denominator, but public. |
+| `unattributable_error_rate` | 0.000 | Errors we could not blame on either side. Flagged for human review. |
+| `unobservable_hit_rate` | n/a | Pass rate on items this tier cannot convey. Measures the design prior — how much it guessed right with no information — which separates reproducing from pattern-matching. |
+| `exempt_count` / `exempt_weight` | 2 / 2.000 | Genre exemptions that stay in the denominator (lo=0, hi=full). Interval width records them; they must not vanish. |
+
+## Genre exemptions
+
+- `U3/collect_removes_one[0]` (O3, w=1.000): genre_inapplicable (twin_holds): the frozen reference declares no gb_collectible in any level, so this is not one of this game's mechanics. measured: res://scen
+- `U4/goal_changes_scene[0]` (O3, w=1.000): genre_inapplicable (twin_holds): gb_goal marks the rival castle hex; the level ends on capture, not overlap (GOLD_TRIAGE #4). measured: res://scenes/main.tscn d
+
+## Perceptual Quality Assessment
+
+Assessment state: `absent`.
+
+The Perceptual Quality Assessment is `absent` and contributes nothing to the total. An uncalibrated judge in a score is a number with no shown relationship to anything it claims to measure.
+
+## Notes
+
+- liveness deadline derived from this project's own structural pass: 57.8 ms/frame median, 57.8 ms/frame worst
+- liveness sweep: 5 live / 1 dead against the 420-frame idle control truth/L0#idle420
+- replayed 5 certified route(s) from a suite reviewed by 'project_owner' on 2026-09-09
+- only 76% of registered Objective Behavioral Evaluation weight produced any reading (O1, O2, O3, O4, O5, O6, O7); coverage=0.88 describes those channels alone and must not be read as coverage of the card.
+- 2 genre-exempt item(s) (weight 2.000) stay in the denominator with lo=0/hi=full; interval width records the exemption. See exempt_items.

minimax/mode1/twin_holds/evaluation/reproduction/routes/readings.json

@@ -0,0 +1,352 @@
+[
+  {
+    "clean_clear": true,
+    "deaths": 0,
+    "detail": "",
+    "devices_required": [],
+    "devices_unresolved": [],
+    "devices_used": [],
+    "errors": [],
+    "frames": 420,
+    "frames_note": "frames is the trace row count; physics_frames_at_end keeps counting while the tree is paused for the agent and is not the game's clock; process_frames_at_end is Engine.get_process_frames() and can diverge from frames when --fixed-fps is off",
+    "goal_frame": -1,
+    "idle_baseline_id": "",
+    "injects": 0,
+    "interface": {
+      "extended_actions": [
+        {
+          "id": "confirm_build",
+          "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+        },
+        {
+          "id": "cycle_build",
+          "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+        },
+        {
+          "id": "demolish",
+          "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+        },
+        {
+          "id": "end_turn",
+          "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+        }
+      ],
+      "interface_version": 2,
+      "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+      "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0"
+    },
+    "invariants_failed": [],
+    "milestone_frames": {},
+    "milestones_reached": [],
+    "missing_groups": [],
+    "mode": "H",
+    "numeric_resolution": {
+      "health": "none",
+      "progress": "none",
+      "score": "none"
+    },
+    "observations_reached": [],
+    "observations_triggered": [],
+    "physics_frames_at_end": 427,
+    "process_frames_at_end": 426,
+    "reached": true,
+    "refusals": [],
+    "replans": 0,
+    "route_id": "twin_holds/L0/no_input_no_clear",
+    "segment_clean": true,
+    "stall_frames": 420,
+    "state_delta": 0.0,
+    "steps": 1,
+    "stop_reason": "budget_frames",
+    "success_before_all_levels": false,
+    "tier": 0,
+    "used_required": true,
+    "wall_ms": 4695.2
+  },
+  {
+    "clean_clear": false,
+    "deaths": 0,
+    "detail": "",
+    "devices_required": [
+      "tile#25"
+    ],
+    "devices_unresolved": [
+      "tile#25"
+    ],
+    "devices_used": [],
+    "errors": [],
+    "frames": 1299,
+    "frames_note": "frames is the trace row count; physics_frames_at_end keeps counting while the tree is paused for the agent and is not the game's clock; process_frames_at_end is Engine.get_process_frames() and can diverge from frames when --fixed-fps is off",
+    "goal_frame": -1,
+    "idle_baseline_id": "",
+    "injects": 0,
+    "interface": {
+      "extended_actions": [
+        {
+          "id": "confirm_build",
+          "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+        },
+        {
+          "id": "cycle_build",
+          "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+        },
+        {
+          "id": "demolish",
+          "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+        },
+        {
+          "id": "end_turn",
+          "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+        }
+      ],
+      "interface_version": 2,
+      "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+      "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0"
+    },
+    "invariants_failed": [],
+    "milestone_frames": {},
+    "milestones_reached": [],
+    "missing_groups": [
+      "numeric:score"
+    ],
+    "mode": "H",
+    "numeric_resolution": {
+      "health": "none",
+      "progress": "none",
+      "score": "none"
+    },
+    "observations_reached": [],
+    "observations_triggered": [],
+    "physics_frames_at_end": 1365,
+    "process_frames_at_end": 1364,
+    "reached": false,
+    "refusals": [],
+    "replans": 0,
+    "route_id": "twin_holds/C1/score_route",
+    "segment_clean": true,
+    "stall_frames": 1299,
+    "state_delta": 1.0,
+    "steps": 197,
+    "stop_reason": "ops_exhausted",
+    "success_before_all_levels": false,
+    "tier": 2,
+    "used_required": false,
+    "wall_ms": 7794.4
+  },
+  {
+    "clean_clear": false,
+    "deaths": 0,
+    "detail": "",
+    "devices_required": [
+      "tile#39"
+    ],
+    "devices_unresolved": [
+      "tile#39"
+    ],
+    "devices_used": [],
+    "errors": [
+      "inject.autoload_state could not be applied completely"
+    ],
+    "frames": 2717,
+    "frames_note": "frames is the trace row count; physics_frames_at_end keeps counting while the tree is paused for the agent and is not the game's clock; process_frames_at_end is Engine.get_process_frames() and can diverge from frames when --fixed-fps is off",
+    "goal_frame": -1,
+    "idle_baseline_id": "",
+    "injects": 0,
+    "interface": {
+      "extended_actions": [
+        {
+          "id": "confirm_build",
+          "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+        },
+        {
+          "id": "cycle_build",
+          "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+        },
+        {
+          "id": "demolish",
+          "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+        },
+        {
+          "id": "end_turn",
+          "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+        }
+      ],
+      "interface_version": 2,
+      "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+      "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0"
+    },
+    "invariants_failed": [],
+    "milestone_frames": {},
+    "milestones_reached": [],
+    "missing_groups": [
+      "numeric:score"
+    ],
+    "mode": "H",
+    "numeric_resolution": {
+      "health": "none",
+      "progress": "none",
+      "score": "none"
+    },
+    "observations_reached": [],
+    "observations_triggered": [],
+    "physics_frames_at_end": 2783,
+    "process_frames_at_end": 2782,
+    "reached": false,
+    "refusals": [],
+    "replans": 0,
+    "route_id": "twin_holds/C2/castle_capture_route",
+    "segment_clean": false,
+    "stall_frames": 2717,
+    "state_delta": 1.0,
+    "steps": 745,
+    "stop_reason": "ops_exhausted",
+    "success_before_all_levels": false,
+    "tier": 3,
+    "used_required": false,
+    "wall_ms": 11786.7
+  },
+  {
+    "clean_clear": false,
+    "deaths": 0,
+    "detail": "",
+    "devices_required": [
+      "tile#56"
+    ],
+    "devices_unresolved": [
+      "tile#56"
+    ],
+    "devices_used": [],
+    "errors": [
+      "inject.autoload_state could not be applied completely"
+    ],
+    "frames": 4775,
+    "frames_note": "frames is the trace row count; physics_frames_at_end keeps counting while the tree is paused for the agent and is not the game's clock; process_frames_at_end is Engine.get_process_frames() and can diverge from frames when --fixed-fps is off",
+    "goal_frame": -1,
+    "idle_baseline_id": "",
+    "injects": 0,
+    "interface": {
+      "extended_actions": [
+        {
+          "id": "confirm_build",
+          "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+        },
+        {
+          "id": "cycle_build",
+          "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+        },
+        {
+          "id": "demolish",
+          "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+        },
+        {
+          "id": "end_turn",
+          "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+        }
+      ],
+      "interface_version": 2,
+      "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+      "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0"
+    },
+    "invariants_failed": [],
+    "milestone_frames": {},
+    "milestones_reached": [],
+    "missing_groups": [
+      "numeric:score"
+    ],
+    "mode": "H",
+    "numeric_resolution": {
+      "health": "none",
+      "progress": "none",
+      "score": "none"
+    },
+    "observations_reached": [],
+    "observations_triggered": [],
+    "physics_frames_at_end": 4841,
+    "process_frames_at_end": 4840,
+    "reached": false,
+    "refusals": [],
+    "replans": 0,
+    "route_id": "twin_holds/C3/advanced_score_route",
+    "segment_clean": false,
+    "stall_frames": 4775,
+    "state_delta": 1.0,
+    "steps": 1475,
+    "stop_reason": "ops_exhausted",
+    "success_before_all_levels": false,
+    "tier": 4,
+    "used_required": false,
+    "wall_ms": 17106.8
+  },
+  {
+    "clean_clear": false,
+    "deaths": 0,
+    "detail": "",
+    "devices_required": [
+      "L1/tile#25",
+      "L2/tile#39",
+      "L3/tile#56"
+    ],
+    "devices_unresolved": [
+      "L1/tile#25",
+      "L2/tile#39",
+      "L3/tile#56"
+    ],
+    "devices_used": [],
+    "errors": [],
+    "frames": 8921,
+    "frames_note": "frames is the trace row count; physics_frames_at_end keeps counting while the tree is paused for the agent and is not the game's clock; process_frames_at_end is Engine.get_process_frames() and can diverge from frames when --fixed-fps is off",
+    "goal_frame": -1,
+    "idle_baseline_id": "",
+    "injects": 0,
+    "interface": {
+      "extended_actions": [
+        {
+          "id": "confirm_build",
+          "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+        },
+        {
+          "id": "cycle_build",
+          "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+        },
+        {
+          "id": "demolish",
+          "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+        },
+        {
+          "id": "end_turn",
+          "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+        }
+      ],
+      "interface_version": 2,
+      "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+      "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0"
+    },
+    "invariants_failed": [],
+    "milestone_frames": {},
+    "milestones_reached": [],
+    "missing_groups": [],
+    "mode": "H",
+    "numeric_resolution": {
+      "health": "none",
+      "progress": "none",
+      "score": "none"
+    },
+    "observations_reached": [],
+    "observations_triggered": [],
+    "physics_frames_at_end": 8928,
+    "process_frames_at_end": 8927,
+    "reached": false,
+    "refusals": [],
+    "replans": 0,
+    "route_id": "twin_holds/L5/whole_game_clear",
+    "segment_clean": true,
+    "stall_frames": 8921,
+    "state_delta": 1.0,
+    "steps": 2423,
+    "stop_reason": "ops_exhausted",
+    "success_before_all_levels": false,
+    "tier": 5,
+    "used_required": false,
+    "wall_ms": 17611.5
+  }
+]

minimax/mode1/twin_holds/evaluation/reproduction/summary.txt

@@ -0,0 +1,10 @@
+submission=game
+task_id=twin_holds
+tier=D2
+mode=X_CPU
+modality=Mv
+score_lo=0.506579
+score_hi=0.690789
+measured_weight_share=0.760000
+coverage=0.881579
+route_sha256=f47e2bcb8a54c0f3e0de97b33fe68a540677c529b27e4afd3bf275bd97d29952

minimax/mode1/twin_holds/evaluation/reproduction/truth_snapshot.json

@@ -0,0 +1,281 @@
+{
+  "declared_actions": [
+    "gb_left",
+    "gb_right",
+    "gb_up",
+    "gb_down",
+    "gb_jump",
+    "gb_action",
+    "gb_pause",
+    "gb_reset"
+  ],
+  "driver_sha256": "7778e6190895bfeb10af8beffeec678ec3b03848b53a674477bcdf796a0c0742",
+  "endings": {
+    "defeat": "res://scenes/defeat.tscn",
+    "victory": "res://scenes/victory.tscn"
+  },
+  "engine": "4.5.1",
+  "harness": {
+    "contract_version": "1",
+    "gb_capture_probe.gd": "sha256:204e194e0071fa165070ba3af7b66553fa77122a973737bc293d25b4bc46d4df",
+    "gb_probe.gd": "sha256:39609c49a433ca5f7a05f2c3cc583693be12614dbe33b80cb5d115da290931c3",
+    "gb_route_driver.gd": "sha256:75645e1b23545ce6c18e3af0c519cd532956ac7cf4b9f50288da2403d39d6945",
+    "gb_truth_driver.gd": "sha256:7778e6190895bfeb10af8beffeec678ec3b03848b53a674477bcdf796a0c0742"
+  },
+  "interface": {
+    "extended_actions": [
+      {
+        "id": "confirm_build",
+        "why": "Places the currently selected building under the cursor and consumes the cursor selection state in one action; a verb the six canonical actions cannot express in one binding."
+      },
+      {
+        "id": "cycle_build",
+        "why": "Advances the build menu index without moving the cursor and updates the building preview; a separate verb the six canonical actions cannot express."
+      },
+      {
+        "id": "demolish",
+        "why": "Removes the building under the cursor and refunds half its cost; a verb the six canonical actions cannot express."
+      },
+      {
+        "id": "end_turn",
+        "why": "Commits the player's build actions, runs the rival script, and advances the round counter as one combined verb; the six canonical actions cannot express."
+      }
+    ],
+    "interface_version": 2,
+    "normalized_sha256": "16a3fb98937e30c36c7af21aa3d3283acc0271c6853615fbcf6edc2092cebaaa",
+    "source_sha256": "5b1fdbb523f094673e998cf77c09b0a801d4a0260c55befabdd4868fc9a163f0"
+  },
+  "levels": [
+    {
+      "bound_actions": [
+        "gb_left",
+        "gb_right",
+        "gb_up",
+        "gb_down",
+        "gb_jump",
+        "gb_action",
+        "gb_pause",
+        "gb_reset"
+      ],
+      "bounds": {
+        "max3": [
+          0.0,
+          0.0,
+          0.0
+        ],
+        "min3": [
+          0.0,
+          0.0,
+          0.0
+        ],
+        "plane": "xy",
+        "valid": true
+      },
+      "collect": {
+        "alive_after": 0,
+        "alive_before": 0,
+        "attempted": false,
+        "delta": 0,
+        "detail": "this level declares no gb_collectible, so there is nothing to pick up",
+        "effect": {},
+        "target": ""
+      },
+      "cycles": [],
+      "cycles_capped": false,
+      "declared_scene": "res://scenes/main.tscn",
+      "entry_method": "none",
+      "errors": [],
+      "goal": {
+        "attempted": false,
+        "changed": false,
+        "detail": "this level declares no gb_goal, so there is no transition to observe",
+        "effect": {
+          "complete_trigger": false,
+          "kind": "observable_progress",
+          "observation_missing": false,
+          "progress_before": null
+        },
+        "frames_to_change": -1,
+        "scene_after": "res://scenes/main.tscn",
+        "scene_before": "res://scenes/main.tscn",
+        "target": ""
+      },
+      "groups": {
+        "gb_checkpoint": {
+          "alive": 0,
+          "normalised": [],
+          "total": 0,
+          "world": []
+        },
+        "gb_collectible": {
+          "alive": 0,
+          "normalised": [],
+          "total": 0,
+          "world": []
+        },
+        "gb_door": {
+          "alive": 0,
+          "normalised": [],
+          "total": 0,
+          "world": []
+        },
+        "gb_enemy": {
+          "alive": 0,
+          "normalised": [],
+          "total": 0,
+          "world": []
+        },
+        "gb_goal": {
+          "alive": 0,
+          "normalised": [],
+          "total": 0,
+          "world": []
+        },
+        "gb_hazard": {
+          "alive": 0,
+          "normalised": [],
+          "total": 0,
+          "world": []
+        },
+        "gb_interactive": {
+          "alive": 0,
+          "normalised": [],
+          "total": 0,
+          "world": []
+        },
+        "gb_player": {
+          "alive": 1,
+          "normalised": [
+            {
+              "valid": false,
+              "x": 0.0,
+              "y": 0.0
+            }
+          ],
+          "total": 1,
+          "world": [
+            [
+              0.0,
+              0.0,
+              0.0
+            ]
+          ]
+        }
+      },
+      "numeric": {},
+      "numeric_source": {},
+      "reached": true,
+      "scene": "res://scenes/main.tscn",
+      "series": [
+        {
+          "alive": {
+            "gb_checkpoint": 0,
+            "gb_collectible": 0,
+            "gb_door": 0,
+            "gb_enemy": 0,
+            "gb_goal": 0,
+            "gb_hazard": 0,
+            "gb_interactive": 0,
+            "gb_player": 1
+          },
+          "frame": 90,
+          "numeric": {},
+          "phase": "settle",
+          "player": [
+            0.0,
+            0.0,
+            0.0
+          ],
+          "scene_path": "res://scenes/main.tscn"
+        },
+        {
+          "alive": {
+            "gb_checkpoint": 0,
+            "gb_collectible": 0,
+            "gb_door": 0,
+            "gb_enemy": 0,
+            "gb_goal": 0,
+            "gb_hazard": 0,
+            "gb_interactive": 0,
+            "gb_player": 1
+          },
+          "frame": 90,
+          "numeric": {},
+          "phase": "settle",
+          "player": [
+            0.0,
+            0.0,
+            0.0
+          ],
+          "scene_path": "res://scenes/main.tscn"
+        }
+      ],
+      "spawn": [
+        0.0,
+        0.0,
+        0.0
+      ],
+      "spawn_norm": {
+        "valid": false,
+        "x": 0.0,
+        "y": 0.0
+      },
+      "stop_reason": "no_goal"
+    }
+  ],
+  "liveness": [
+    {
+      "action": "gb_left",
+      "detail": "gb_left differs from the idle control on text (delta 1.0 > 0.5)",
+      "idle_baseline_id": "truth/L0#idle420",
+      "live": true,
+      "via": "text"
+    },
+    {
+      "action": "gb_right",
+      "detail": "gb_right differs from the idle control on text (delta 1.0 > 0.5)",
+      "idle_baseline_id": "truth/L0#idle420",
+      "live": true,
+      "via": "text"
+    },
+    {
+      "action": "gb_up",
+      "detail": "gb_up differs from the idle control on text (delta 1.0 > 0.5)",
+      "idle_baseline_id": "truth/L0#idle420",
+      "live": true,
+      "via": "text"
+    },
+    {
+      "action": "gb_down",
+      "detail": "gb_down is indistinguishable from the idle control on every signature component; whatever the world did, it did without being asked; also no effect under repeated taps (1 frame on, 5 off)",
+      "idle_baseline_id": "truth/L0#idle420",
+      "live": false,
+      "via": "none"
+    },
+    {
+      "action": "gb_jump",
+      "detail": "gb_jump differs from the idle control on text (delta 1.0 > 0.5)",
+      "idle_baseline_id": "truth/L0#idle420",
+      "live": true,
+      "via": "text"
+    },
+    {
+      "action": "gb_action",
+      "detail": "gb_action differs from the idle control on text (delta 1.0 > 0.5)",
+      "idle_baseline_id": "truth/L0#idle420",
+      "live": true,
+      "via": "text"
+    }
+  ],
+  "no_input_no_win": true,
+  "no_input_stop_reason": "budget_frames",
+  "notes": [
+    "liveness deadline derived from this project's own structural pass: 57.8 ms/frame median, 57.8 ms/frame worst",
+    "liveness sweep: 5 live / 1 dead against the 420-frame idle control truth/L0#idle420"
+  ],
+  "probe_sha256": "39609c49a433ca5f7a05f2c3cc583693be12614dbe33b80cb5d115da290931c3",
+  "project": "game",
+  "read_failure": "",
+  "taken_at": "2026-09-24T05:54:21+00:00",
+  "version": "1"
+}

minimax/mode1/twin_holds/metadata.json

@@ -0,0 +1,13 @@
+{
+  "game": "twin_holds",
+  "mode": "brief",
+  "case_id": "",
+  "status": "completed",
+  "phase": "completed",
+  "agent_exit_code": 0,
+  "source_result": "remote_sync",
+  "source_cell": "remote_sync/twin_holds__brief",
+  "reference_video_requested": true,
+  "reference_video_copied_into_package": true,
+  "materialization": "remote_evaluation_import"
+}