roboolympics-fun masters: 6 EO Fun families re-rendered from their aef3b33b59 formal Astra results (board 0252)

masters/roboolympics-fun/dunk/meta.json

@@ -1,17 +1,20 @@
 {
  "suite": "roboolympics-fun",
  "task": "dunk",
- "source": "agent-success:codex-gpt6_astra-high-roboolympics-dunk-i00-privileged/roboolympics-dunk-i00-privileged__WMRy4fr",
- "renderer": "Blender 4.2.23 LTS Cycles path tracing (GPU OptiX), adaptive sampling (threshold 0.01, 64-128 spp), OpenImageDenoise with albedo + normal passes; kinematic playback of the exact MuJoCo replay poses (MuJoCo's own compiled meshes, primitives and colours); daylight: a sun lamp (2 degree disc) and a Nishita sky (no sun disc), the venue's own ground colour on a large ground plane with a faint world-space variation; AgX view transform, Medium High Contrast look; a tracking camera (fixed 22 degree elevation, per-frame fit of the robot and the object in play, smoothed in time) with a slow eased 10-degree orbit",
+ "family": "dunk",
+ "task_dir": "embodied-olympics-fun-dunk-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@aef3b33b59",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-dunk-i00-privileged/embodied-olympics-fun-dunk-i00-p__eruDdGs",
+ "renderer": "Blender 4.5.3 LTS Cycles path tracing (GPU OptiX), adaptive sampling (threshold 0.01, 64-128 spp), OpenImageDenoise with albedo + normal passes; kinematic playback of the exact MuJoCo replay poses (MuJoCo's own compiled meshes, primitives and colours); daylight: a sun lamp (2 degree disc) and a Nishita sky (no sun disc), the venue's own ground colour on a large ground plane with a faint world-space variation; AgX view transform, Medium High Contrast look; a tracking camera (per-frame fit of the robot and the object in play, smoothed in time, zooming out early) with a slow eased 10.0-degree orbit",
  "rendering": "path-tracing",
  "width": 2560,
  "height": 1440,
  "fps": 30,
- "frames": 491,
- "seconds": 16.37,
- "speed": 0.9,
+ "frames": 352,
+ "seconds": 11.73,
+ "speed": 0.3,
  "success_check": true,
- "success_check_how": "the task's own verdict (tests/rcb_rfo.py Episode.verdict(), the referee) at the end of this re-render's own replay of every recorded row from the task's frozen instance; a second plain replay with the task's own replay() agrees and ends in the same final state hash (91fedb1d4e1cd41b)",
+ "success_check_how": "the task's own verdict (tests/rcb_rfo.py Episode.verdict(), the referee) at the end of this re-render's own replay of every recorded row from the task's frozen instance on the new task version; a second plain replay with the task's own replay() agrees and ends in the same final state hash (af65930fbcc34626)",
  "end_check": {
   "success": true,
   "stages": [
@@ -21,59 +24,68 @@
   "stages_total": 2,
   "progress": 1.0,
   "metrics": {
-   "ball_height_m": 0.729
+   "ball_height_m": 1.641,
+   "basket_touch_s": 0.098
   },
-  "time_s": 12.89
+  "time_s": 2.91,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the verifier image of the version it was graded on (which is this master's version); both end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "aef3b33b59 (files 0.3.3; the same version as this master, so the two replays in it are the whole check)",
+  "final_state_sha": "af65930fbcc346264dca4eba63063c59f470d18a3032263ee41dbbd73178efae",
+  "steps": 291,
+  "pass": true
  },
  "replay": {
-  "trajectory": "a run of the exam",
-  "n_actions": 1289,
-  "control_steps_replayed": 1289,
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-dunk-i00-privileged/embodied-olympics-fun-dunk-i00-p__eruDdGs/artifacts/app/output/trajectory.npz",
+  "n_actions": 291,
+  "control_steps_replayed": 291,
   "control_hz": 100.0,
-  "every_control_steps": 3,
-  "final_state_sha": "91fedb1d4e1cd41b9557fe13c3f347ca96259b8e6cee12e9a31217e13a8f2419",
+  "every_control_steps": 1,
+  "final_state_sha": "af65930fbcc346264dca4eba63063c59f470d18a3032263ee41dbbd73178efae",
   "original_verdict": "counted success (verifier reward 1)"
  },
  "visual_changes": [
-  "daylight sky and a plain ground plane in the venue's ground colour instead of the venue's painted sky, hills, trees, grandstands and perimeter boards (drawn-only scenery)",
+  "daylight sky and a plain ground plane instead of the venue's painted sky, hills, trees, grandstands and perimeter boards (drawn-only scenery, geoms named rfo_*)",
   "the ball drawn as a basketball (orange, black channels)"
  ],
  "camera": {
   "mode": "director",
-  "az_deg": -83.3,
-  "az_rule": "object-travel-115",
+  "az_deg": -50.0,
+  "az_rule": "facing-50",
   "el_deg": 22.0,
   "fovy_deg": 34.0,
   "fill": 0.72,
   "distance_m": [
-   4.318,
-   12.322
+   4.352,
+   5.863
   ],
   "robot_root": "pelvis",
-  "travel_m": 0.423,
+  "travel_m": 0.213,
   "focus": {
-   "robot+object": 284,
-   "object": 28,
-   "robot": 119
+   "robot+object": 212,
+   "robot": 80
   },
   "objects": {
-   "ball": 8.55
+   "ball": 1.91
   }
  },
  "orbit_deg": 10.0,
  "holds": {
+  "start_frames": 15,
+  "end_frames": 45,
   "start_s": 0.5,
   "end_s": 1.5,
   "note": "first/last frame held, no extra simulation"
  },
- "seconds_per_frame": 3.59,
- "image": "eai-robofunolympics:0.2.1",
+ "seconds_per_frame": 2.97,
+ "image": "ghcr.io/mll-lab-nu/efe:embodied-olympics-fun-0.2.1 + embodied-olympics-fun-files-0.3.3 (the task's tests/ verifier image)",
  "agent": "codex + GPT-6 Astra (high)",
  "run": "codex-gpt6_astra-high",
  "mode": "privileged",
- "trial_choice": "a counted GPT-6 Astra success",
- "poster": "frame 309: the ball at the ring (the dunk)",
- "host": "gpu3 062523e6f067",
- "encode": "libx264 preset slow CRF 16, yuv420p, bt709",
- "created": "2026-10-18T07:23:33-0500"
+ "trial_choice": "a counted GPT-6 Astra success: its formal privileged result on this task version (the quick check of aef3b33b59, reviewed and promoted)",
+ "poster": "frame 140 (same scene, camera and settings as the video)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T21:09:41-0500"
 }

masters/roboolympics-fun/dunk/video.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7cec832633f667b9d1db4d995ac256a3d2dcb08377e95346860cfffb273f96c9
-size 8713925
+oid sha256:734a7b104ac60fb16d6ec4f5109a39f409b96c2a5159811624aae6cbb4311e5f
+size 6716384

masters/roboolympics-fun/dunk_between_legs/meta.json

@@ -1,19 +1,22 @@
 {
  "suite": "roboolympics-fun",
  "task": "dunk_between_legs",
+ "family": "dunk-between-legs",
+ "task_dir": "embodied-olympics-fun-dunk-between-legs-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@aef3b33b59",
  "source": "scene-orbit",
- "why_no_video": "No evaluated agent has solved this task (no counted success in any run when this was made). As for the other unsolved RoboOlympics-Fun tasks on the site (board 0161 rule), this video shows the best failed attempt instead, replayed exactly (not a solved run): the task's own check at the end says it failed, as the verifier did.",
- "shows": "agent-attempt (failed):codex-gpt6_astra-high-roboolympics-fun-dunk-between-legs-i00-privileged/roboolympics-fun-dunk-between-le__zHY69cc",
- "renderer": "Blender 4.2.23 LTS Cycles path tracing (GPU OptiX), adaptive sampling (threshold 0.01, 64-128 spp), OpenImageDenoise with albedo + normal passes; kinematic playback of the exact MuJoCo replay poses (MuJoCo's own compiled meshes, primitives and colours); daylight: a sun lamp (2 degree disc) and a Nishita sky (no sun disc), the venue's own ground colour on a large ground plane with a faint world-space variation; AgX view transform, Medium High Contrast look; a tracking camera (fixed 22 degree elevation, per-frame fit of the robot and the object in play, smoothed in time) with a slow eased 10-degree orbit",
+ "why_no_video": "No evaluated agent has a success on this task version yet. Site rule (boards 0161 / 0252): show GPT-6 Astra's failed attempt with the most progress instead. This video is that attempt replayed exactly (not a solved run): the task's own referee at the end says it failed, as the verifier did.",
+ "shows": "agent-attempt (failed):codex-gpt6_astra-high-embodied-olympics-fun-dunk-between-legs-i00-privileged/embodied-olympics-fun-dunk-betwe__xbhRyY5",
+ "renderer": "Blender 4.5.3 LTS Cycles path tracing (GPU OptiX), adaptive sampling (threshold 0.01, 64-128 spp), OpenImageDenoise with albedo + normal passes; kinematic playback of the exact MuJoCo replay poses (MuJoCo's own compiled meshes, primitives and colours); daylight: a sun lamp (2 degree disc) and a Nishita sky (no sun disc), the venue's own ground colour on a large ground plane with a faint world-space variation; AgX view transform, Medium High Contrast look; a tracking camera (per-frame fit of the robot and the object in play, smoothed in time, zooming out early) with a slow eased 10.0-degree orbit",
  "rendering": "path-tracing",
  "width": 2560,
  "height": 1440,
  "fps": 30,
- "frames": 293,
- "seconds": 9.77,
+ "frames": 550,
+ "seconds": 18.33,
  "speed": 0.9,
  "success_check": false,
- "success_check_how": "the task's own verdict (rcb_rfo.py Episode.verdict(), the referee, from the task dir's tests/ (privileged verifier copy)) at the end of this re-render's own replay of every recorded row from the task's frozen instance (t = 0 state hash checked); a second plain replay with the task's own replay() agrees and ends in the same final state hash (4bb20c99b755beb2)",
+ "success_check_how": "the task's own verdict (tests/rcb_rfo.py Episode.verdict(), the referee) at the end of this re-render's own replay of every recorded row from the task's frozen instance on the new task version; a second plain replay with the task's own replay() agrees and ends in the same final state hash (64bc0aad4a2f4c9b)",
  "end_check": {
   "success": false,
   "stages": [
@@ -24,79 +27,70 @@
   "stages_total": 5,
   "progress": 0.6,
   "metrics": {
-   "ball_height_m": 1.344,
+   "ball_height_m": 2.005,
    "basket_touch_s": 0.0,
-   "dribbles": 7,
-   "between_the_legs": 2
+   "dribbles": 6,
+   "between_the_legs": 1
   },
-  "time_s": 6.95
+  "time_s": 14.65,
+  "failed": null
  },
- "recorded_result": {
-  "verifier_reward_json": {
-   "success": 0,
-   "n_actions": 695,
-   "progress": 0.6,
-   "stages_reached": 3,
-   "stages_total": 5,
-   "metric_ball_height_m": 1.344,
-   "metric_basket_touch_s": 0.0,
-   "metric_between_the_legs": 2.0,
-   "metric_dribbles": 7.0
-  }
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the verifier image of the version it was graded on (which is this master's version); both end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "aef3b33b59 (files 0.3.3; the same version as this master, so the two replays in it are the whole check)",
+  "final_state_sha": "64bc0aad4a2f4c9be3c4c5fc5cfd05c73136ab7b4d7c7358857754cea133f748",
+  "steps": 1465,
+  "pass": true
  },
  "replay": {
-  "task_dir": "roboolympics-fun-dunk-between-legs-i00-privileged",
-  "trajectory": "jobs/codex/gpt6_astra/high/codex-gpt6_astra-high-roboolympics-fun-dunk-between-legs-i00-privileged/roboolympics-fun-dunk-between-le__zHY69cc/artifacts/app/output/trajectory.npz",
-  "n_actions": 695,
-  "control_steps_replayed": 695,
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-dunk-between-legs-i00-privileged/embodied-olympics-fun-dunk-betwe__xbhRyY5/artifacts/app/output/trajectory.npz",
+  "n_actions": 1465,
+  "control_steps_replayed": 1465,
   "control_hz": 100.0,
   "every_control_steps": 3,
-  "final_state_sha": "4bb20c99b755beb27d523b60b987ed763ecfa8c4009abc34a793eaaf4e98c523",
-  "replays_agree": true,
-  "mujoco": "3.3.7",
+  "final_state_sha": "64bc0aad4a2f4c9be3c4c5fc5cfd05c73136ab7b4d7c7358857754cea133f748",
   "original_verdict": "counted failure (verifier reward 0)"
  },
  "visual_changes": [
-  "daylight sky and a plain ground plane in the venue's ground colour instead of the venue's painted sky, hills, trees, grandstands and perimeter boards (drawn-only scenery)",
+  "daylight sky and a plain ground plane instead of the venue's painted sky, hills, trees, grandstands and perimeter boards (drawn-only scenery, geoms named rfo_*)",
   "the ball drawn as a basketball (orange, black channels)"
  ],
  "camera": {
   "mode": "director",
-  "az_deg": -75.18,
+  "az_deg": -70.57,
   "az_rule": "robot-travel-70",
   "el_deg": 22.0,
   "fovy_deg": 34.0,
   "fill": 0.72,
   "distance_m": [
-   4.203,
-   4.539
+   4.498,
+   6.681
   ],
   "robot_root": "pelvis",
-  "travel_m": 2.88,
+  "travel_m": 3.589,
   "focus": {
-   "robot": 9,
-   "robot+object": 224
+   "robot+object": 320,
+   "robot": 170
   },
   "objects": {
-   "ball": 2.89
+   "ball": 3.64
   }
  },
  "orbit_deg": 10.0,
  "holds": {
+  "start_frames": 15,
+  "end_frames": 45,
   "start_s": 0.5,
   "end_s": 1.5,
   "note": "first/last frame held, no extra simulation"
  },
- "seconds_per_frame": 3.11,
- "image": "ghcr.io/mll-lab-nu/efe:robofunolympics-0.2.1 (simulator) + ghcr.io/mll-lab-nu/efe:roboolympics-fun-files-0.1.0@sha256:49a33342a72b57b56742cc754e427205a30f39dfcd69a99d1d7275c0eae4a619 (rcb_rfo.py, rfo/)",
- "task_dir": "roboolympics-fun-dunk-between-legs-i00-privileged",
+ "seconds_per_frame": 2.91,
+ "image": "ghcr.io/mll-lab-nu/efe:embodied-olympics-fun-0.2.1 + embodied-olympics-fun-files-0.3.3 (the task's tests/ verifier image)",
  "agent": "codex + GPT-6 Astra (high)",
  "run": "codex-gpt6_astra-high",
  "mode": "privileged",
- "trial": "codex-gpt6_astra-high-roboolympics-fun-dunk-between-legs-i00-privileged/roboolympics-fun-dunk-between-le__zHY69cc",
- "trial_choice": "no counted success in any run; GPT-6 Astra's privileged attempt with the most progress",
+ "trial_choice": "no success on this task version; GPT-6 Astra's failed attempt (most progress)",
  "poster": "last frame (end of the failed attempt)",
- "host": "vast.ai RTX 4090 gpu2",
- "encode": "libx264 preset slow CRF 16, yuv420p, bt709, faststart",
- "created": "2026-10-19T06:14:58-0500"
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T21:09:41-0500"
 }

masters/roboolympics-fun/dunk_between_legs/video.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4616f481c1104f326f8c19c9bbd99addef171e86b228678abd5aee0189b384be
-size 6603886
+oid sha256:219ecca0e641d22d4f989d8927ff82b76831fc337dd5bda1fcab92554672b306
+size 10614174

masters/roboolympics-fun/freeski_360/meta.json

@@ -0,0 +1,108 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "freeski_360",
+ "family": "freeski-360",
+ "task_dir": "embodied-olympics-fun-freeski-360-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@aef3b33b59",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-freeski-360-i00-privileged/embodied-olympics-fun-freeski-36__pA2yUNn",
+ "renderer": "Blender 4.5.3 LTS Cycles path tracing (GPU OptiX), adaptive sampling (threshold 0.01, 64-128 spp), OpenImageDenoise with albedo + normal passes; kinematic playback of the exact MuJoCo replay poses (MuJoCo's own compiled meshes, primitives and colours); daylight: a sun lamp (2 degree disc) and a Nishita sky (no sun disc), the venue's own ground colour on a large ground plane with a faint world-space variation; AgX view transform, Medium High Contrast look; snow venue: a snow shader (bright blue-white, rough, slight subsurface, fine bump) on the course's snow geoms, snow-white valley floor; a chase camera: per-frame fit of the robot (height included, so jumps and spins stay framed), at least 6 m away, 24 degree elevation, smoothed in time",
+ "rendering": "path-tracing",
+ "width": 2560,
+ "height": 1440,
+ "fps": 30,
+ "frames": 649,
+ "seconds": 21.63,
+ "speed": 0.9,
+ "success_check": true,
+ "success_check_how": "the task's own verdict (tests/rcb_rfo.py Episode.verdict(), the referee) at the end of this re-render's own replay of every recorded row from the task's frozen instance on the new task version; a second plain replay with the task's own replay() agrees and ends in the same final state hash (4b118a2881f108ea)",
+ "end_check": {
+  "success": true,
+  "stages": [
+   "start",
+   "take-off",
+   "landing",
+   "ride-away"
+  ],
+  "stages_total": 4,
+  "progress": 1.0,
+  "metrics": {
+   "speed_m_s": 18.198,
+   "trick_spin_deg": 355.0,
+   "trick_flip_deg": -74.8,
+   "trick_short_deg": 5.0,
+   "trick_half_turns": 2,
+   "trick_flips": 0,
+   "trick_pelvis_spin_deg": 324.6,
+   "air_s": 1.92,
+   "height_m": 2.64,
+   "from_knoll_m": 3.17,
+   "landing_heading_deg": [
+    8.2,
+    9.1
+   ],
+   "ride_turn_deg": 13.6
+  },
+  "time_s": 17.63,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the verifier image of the version it was graded on (which is this master's version); both end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "aef3b33b59 (files 0.3.3; the same version as this master, so the two replays in it are the whole check)",
+  "final_state_sha": "4b118a2881f108eacac1567e97a8f6d18a2c56e47f42ac098a435ee5b7721169",
+  "steps": 1763,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-freeski-360-i00-privileged/embodied-olympics-fun-freeski-36__pA2yUNn/artifacts/app/output/trajectory.npz",
+  "n_actions": 1763,
+  "control_steps_replayed": 1763,
+  "control_hz": 100.0,
+  "every_control_steps": 3,
+  "final_state_sha": "4b118a2881f108eacac1567e97a8f6d18a2c56e47f42ac098a435ee5b7721169",
+  "original_verdict": "counted success (verifier reward 1)"
+ },
+ "visual_changes": [
+  "daylight sky and a plain ground plane instead of the venue's painted sky, hills, trees, grandstands and perimeter boards (drawn-only scenery, geoms named rfo_*)",
+  "textured snow drawn with a snow shader instead of its checker texture's mean colour",
+  "the start and finish arches and the landing's distance-marker boards (drawn-only, no contact) not drawn: they hid the robot from the chase camera; the safety nets are drawn"
+ ],
+ "camera": {
+  "mode": "director",
+  "az_deg": -67.85,
+  "az_rule": "robot-travel-70",
+  "el_deg": 24.0,
+  "fovy_deg": 34.0,
+  "fill": 0.72,
+  "distance_m": [
+   6.24,
+   14.608
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 143.407,
+  "focus": {
+   "robot": 87,
+   "robot+object": 502
+  },
+  "objects": {
+   "left_ski": 143.54,
+   "right_ski": 143.57
+  }
+ },
+ "orbit_deg": 0.0,
+ "holds": {
+  "start_frames": 15,
+  "end_frames": 45,
+  "start_s": 0.5,
+  "end_s": 1.5,
+  "note": "first/last frame held, no extra simulation"
+ },
+ "seconds_per_frame": 4.88,
+ "image": "ghcr.io/mll-lab-nu/efe:embodied-olympics-fun-0.2.1 + embodied-olympics-fun-files-0.3.3 (the task's tests/ verifier image)",
+ "agent": "codex + GPT-6 Astra (high)",
+ "run": "codex-gpt6_astra-high",
+ "mode": "privileged",
+ "trial_choice": "a counted GPT-6 Astra success: its formal privileged result on this task version (the quick check of aef3b33b59, reviewed and promoted)",
+ "poster": "last frame (solved state)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T21:09:41-0500"
+}

masters/roboolympics-fun/freeski_360/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a33bccfef63001ce1cecbe29fe0d1e2f403b5e062164d0c2150e01d39025df16
+size 13328293

masters/roboolympics-fun/freeski_straight_air/meta.json

@@ -0,0 +1,108 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "freeski_straight_air",
+ "family": "freeski-straight-air",
+ "task_dir": "embodied-olympics-fun-freeski-straight-air-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@aef3b33b59",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-freeski-straight-air-i00-privileged/embodied-olympics-fun-freeski-st__uJxLXfG",
+ "renderer": "Blender 4.5.3 LTS Cycles path tracing (GPU OptiX), adaptive sampling (threshold 0.01, 64-128 spp), OpenImageDenoise with albedo + normal passes; kinematic playback of the exact MuJoCo replay poses (MuJoCo's own compiled meshes, primitives and colours); daylight: a sun lamp (2 degree disc) and a Nishita sky (no sun disc), the venue's own ground colour on a large ground plane with a faint world-space variation; AgX view transform, Medium High Contrast look; snow venue: a snow shader (bright blue-white, rough, slight subsurface, fine bump) on the course's snow geoms, snow-white valley floor; a chase camera: per-frame fit of the robot (height included, so jumps and spins stay framed), at least 6 m away, 24 degree elevation, smoothed in time",
+ "rendering": "path-tracing",
+ "width": 2560,
+ "height": 1440,
+ "fps": 30,
+ "frames": 634,
+ "seconds": 21.13,
+ "speed": 0.9,
+ "success_check": true,
+ "success_check_how": "the task's own verdict (tests/rcb_rfo.py Episode.verdict(), the referee) at the end of this re-render's own replay of every recorded row from the task's frozen instance on the new task version; a second plain replay with the task's own replay() agrees and ends in the same final state hash (739a755245c7c0f3)",
+ "end_check": {
+  "success": true,
+  "stages": [
+   "start",
+   "take-off",
+   "landing",
+   "ride-away"
+  ],
+  "stages_total": 4,
+  "progress": 1.0,
+  "metrics": {
+   "speed_m_s": 18.519,
+   "trick_spin_deg": 0.1,
+   "trick_flip_deg": -9.0,
+   "trick_short_deg": 0.1,
+   "trick_half_turns": 0,
+   "trick_flips": 0,
+   "trick_pelvis_spin_deg": 5.7,
+   "air_s": 2.134,
+   "height_m": 2.6,
+   "from_knoll_m": 7.53,
+   "landing_heading_deg": [
+    0.3,
+    0.3
+   ],
+   "ride_turn_deg": 0.5
+  },
+  "time_s": 17.19,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the verifier image of the version it was graded on (which is this master's version); both end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "aef3b33b59 (files 0.3.3; the same version as this master, so the two replays in it are the whole check)",
+  "final_state_sha": "739a755245c7c0f323f4204dfcdacc00ee2a630de6eadecf27feb12f1dc362f1",
+  "steps": 1719,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-freeski-straight-air-i00-privileged/embodied-olympics-fun-freeski-st__uJxLXfG/artifacts/app/output/trajectory.npz",
+  "n_actions": 1719,
+  "control_steps_replayed": 1719,
+  "control_hz": 100.0,
+  "every_control_steps": 3,
+  "final_state_sha": "739a755245c7c0f323f4204dfcdacc00ee2a630de6eadecf27feb12f1dc362f1",
+  "original_verdict": "counted success (verifier reward 1)"
+ },
+ "visual_changes": [
+  "daylight sky and a plain ground plane instead of the venue's painted sky, hills, trees, grandstands and perimeter boards (drawn-only scenery, geoms named rfo_*)",
+  "textured snow drawn with a snow shader instead of its checker texture's mean colour",
+  "the start and finish arches and the landing's distance-marker boards (drawn-only, no contact) not drawn: they hid the robot from the chase camera; the safety nets are drawn"
+ ],
+ "camera": {
+  "mode": "director",
+  "az_deg": -70.64,
+  "az_rule": "robot-travel-70",
+  "el_deg": 24.0,
+  "fovy_deg": 34.0,
+  "fill": 0.72,
+  "distance_m": [
+   6.24,
+   14.96
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 143.424,
+  "focus": {
+   "robot": 65,
+   "robot+object": 509
+  },
+  "objects": {
+   "left_ski": 143.44,
+   "right_ski": 143.43
+  }
+ },
+ "orbit_deg": 0.0,
+ "holds": {
+  "start_frames": 15,
+  "end_frames": 45,
+  "start_s": 0.5,
+  "end_s": 1.5,
+  "note": "first/last frame held, no extra simulation"
+ },
+ "seconds_per_frame": 4.94,
+ "image": "ghcr.io/mll-lab-nu/efe:embodied-olympics-fun-0.2.1 + embodied-olympics-fun-files-0.3.3 (the task's tests/ verifier image)",
+ "agent": "codex + GPT-6 Astra (high)",
+ "run": "codex-gpt6_astra-high",
+ "mode": "privileged",
+ "trial_choice": "a counted GPT-6 Astra success: its formal privileged result on this task version (the quick check of aef3b33b59, reviewed and promoted)",
+ "poster": "last frame (solved state)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T21:09:41-0500"
+}

masters/roboolympics-fun/freeski_straight_air/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eee12262695d1c0d7a9c5479f0cfcfe61bef95a64c7d8c66abcbe7cc76e7b41f
+size 12635346

masters/roboolympics-fun/handstand/meta.json

@@ -0,0 +1,88 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "handstand",
+ "family": "handstand",
+ "task_dir": "embodied-olympics-fun-handstand-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@aef3b33b59",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-handstand-i00-privileged/embodied-olympics-fun-handstand__jNPsVm2",
+ "renderer": "Blender 4.5.3 LTS Cycles path tracing (GPU OptiX), adaptive sampling (threshold 0.01, 64-128 spp), OpenImageDenoise with albedo + normal passes; kinematic playback of the exact MuJoCo replay poses (MuJoCo's own compiled meshes, primitives and colours); daylight: a sun lamp (2 degree disc) and a Nishita sky (no sun disc), the venue's own ground colour on a large ground plane with a faint world-space variation; AgX view transform, Medium High Contrast look; a tracking camera (per-frame fit of the robot and the object in play, smoothed in time, zooming out early) with a slow eased 10.0-degree orbit",
+ "rendering": "path-tracing",
+ "width": 2560,
+ "height": 1440,
+ "fps": 30,
+ "frames": 665,
+ "seconds": 22.17,
+ "speed": 1.2,
+ "success_check": true,
+ "success_check_how": "the task's own verdict (tests/rcb_rfo.py Episode.verdict(), the referee) at the end of this re-render's own replay of every recorded row from the task's frozen instance on the new task version; a second plain replay with the task's own replay() agrees and ends in the same final state hash (0738c8c70a86406e)",
+ "end_check": {
+  "success": true,
+  "stages": [
+   "handstand",
+   "held 2 s",
+   "on the feet",
+   "stand"
+  ],
+  "stages_total": 4,
+  "progress": 1.0,
+  "metrics": {
+   "hold_s": 2.0
+  },
+  "time_s": 24.14,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the verifier image of the version it was graded on (which is this master's version); both end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "aef3b33b59 (files 0.3.3; the same version as this master, so the two replays in it are the whole check)",
+  "final_state_sha": "0738c8c70a86406e29172fd03c5ce6561f1127d63a3580f145a7f58876884aef",
+  "steps": 2414,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-handstand-i00-privileged/embodied-olympics-fun-handstand__jNPsVm2/artifacts/app/output/trajectory.npz",
+  "n_actions": 2414,
+  "control_steps_replayed": 2414,
+  "control_hz": 100.0,
+  "every_control_steps": 4,
+  "final_state_sha": "0738c8c70a86406e29172fd03c5ce6561f1127d63a3580f145a7f58876884aef",
+  "original_verdict": "counted success (verifier reward 1)"
+ },
+ "visual_changes": [
+  "daylight sky and a plain ground plane instead of the venue's painted sky, hills, trees, grandstands and perimeter boards (drawn-only scenery, geoms named rfo_*)"
+ ],
+ "camera": {
+  "mode": "director",
+  "az_deg": -50.0,
+  "az_rule": "facing-50",
+  "el_deg": 22.0,
+  "fovy_deg": 34.0,
+  "fill": 0.72,
+  "distance_m": [
+   2.132,
+   4.146
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 0.015,
+  "focus": {
+   "robot": 605
+  },
+  "objects": {}
+ },
+ "orbit_deg": 10.0,
+ "holds": {
+  "start_frames": 15,
+  "end_frames": 45,
+  "start_s": 0.5,
+  "end_s": 1.5,
+  "note": "first/last frame held, no extra simulation"
+ },
+ "seconds_per_frame": 2.62,
+ "image": "ghcr.io/mll-lab-nu/efe:embodied-olympics-fun-0.2.1 + embodied-olympics-fun-files-0.3.3 (the task's tests/ verifier image)",
+ "agent": "codex + GPT-6 Astra (high)",
+ "run": "codex-gpt6_astra-high",
+ "mode": "privileged",
+ "trial_choice": "a counted GPT-6 Astra success: its formal privileged result on this task version (the quick check of aef3b33b59, reviewed and promoted)",
+ "poster": "last frame (solved state)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T21:09:41-0500"
+}

masters/roboolympics-fun/handstand/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2252c46f846b36d70b7a7dbcdbedd6741fd3f99efa7bacc1998af7a721c67bf
+size 11327778

masters/roboolympics-fun/wall_ball/meta.json

@@ -0,0 +1,94 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "wall_ball",
+ "family": "wall-ball",
+ "task_dir": "embodied-olympics-fun-wall-ball-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@aef3b33b59",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-wall-ball-i00-privileged/embodied-olympics-fun-wall-ball__is3AtBU",
+ "renderer": "Blender 4.5.3 LTS Cycles path tracing (GPU OptiX), adaptive sampling (threshold 0.01, 64-128 spp), OpenImageDenoise with albedo + normal passes; kinematic playback of the exact MuJoCo replay poses (MuJoCo's own compiled meshes, primitives and colours); daylight: a sun lamp (2 degree disc) and a Nishita sky (no sun disc), the venue's own ground colour on a large ground plane with a faint world-space variation; AgX view transform, Medium High Contrast look; a tracking camera (per-frame fit of the robot and the object in play, smoothed in time, zooming out early) with a slow eased 10.0-degree orbit",
+ "rendering": "path-tracing",
+ "width": 2560,
+ "height": 1440,
+ "fps": 30,
+ "frames": 385,
+ "seconds": 12.83,
+ "speed": 0.3,
+ "success_check": true,
+ "success_check_how": "the task's own verdict (tests/rcb_rfo.py Episode.verdict(), the referee) at the end of this re-render's own replay of every recorded row from the task's frozen instance on the new task version; a second plain replay with the task's own replay() agrees and ends in the same final state hash (f78017a52e630b6b)",
+ "end_check": {
+  "success": true,
+  "stages": [
+   "release",
+   "target",
+   "catch",
+   "hold"
+  ],
+  "stages_total": 4,
+  "progress": 1.0,
+  "metrics": {
+   "release_speed_m_s": 16.42,
+   "target_offset_m": 0.001,
+   "wall_speed_m_s": 14.69,
+   "catch_height_m": 0.77
+  },
+  "time_s": 3.24,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the verifier image of the version it was graded on (which is this master's version); both end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "aef3b33b59 (files 0.3.3; the same version as this master, so the two replays in it are the whole check)",
+  "final_state_sha": "f78017a52e630b6bc03b3bd8743c5079f9a6d5c6d7b7db78ccc2d690322ecbb2",
+  "steps": 324,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-wall-ball-i00-privileged/embodied-olympics-fun-wall-ball__is3AtBU/artifacts/app/output/trajectory.npz",
+  "n_actions": 324,
+  "control_steps_replayed": 324,
+  "control_hz": 100.0,
+  "every_control_steps": 1,
+  "final_state_sha": "f78017a52e630b6bc03b3bd8743c5079f9a6d5c6d7b7db78ccc2d690322ecbb2",
+  "original_verdict": "counted success (verifier reward 1)"
+ },
+ "visual_changes": [
+  "daylight sky and a plain ground plane instead of the venue's painted sky, hills, trees, grandstands and perimeter boards (drawn-only scenery, geoms named rfo_*)"
+ ],
+ "camera": {
+  "mode": "director",
+  "az_deg": -50.0,
+  "az_rule": "facing-50",
+  "el_deg": 22.0,
+  "fovy_deg": 34.0,
+  "fill": 0.72,
+  "distance_m": [
+   4.339,
+   10.56
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 0.09,
+  "focus": {
+   "robot": 52,
+   "robot+object": 273
+  },
+  "objects": {
+   "ball": 0.43
+  }
+ },
+ "orbit_deg": 10.0,
+ "holds": {
+  "start_frames": 15,
+  "end_frames": 45,
+  "start_s": 0.5,
+  "end_s": 1.5,
+  "note": "first/last frame held, no extra simulation"
+ },
+ "seconds_per_frame": 2.68,
+ "image": "ghcr.io/mll-lab-nu/efe:embodied-olympics-fun-0.2.1 + embodied-olympics-fun-files-0.3.3 (the task's tests/ verifier image)",
+ "agent": "codex + GPT-6 Astra (high)",
+ "run": "codex-gpt6_astra-high",
+ "mode": "privileged",
+ "trial_choice": "a counted GPT-6 Astra success: its formal privileged result on this task version (the quick check of aef3b33b59, reviewed and promoted)",
+ "poster": "frame 175 (same scene, camera and settings as the video)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T21:09:41-0500"
+}

masters/roboolympics-fun/wall_ball/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8baaf1ddc443b1c727367af9c89ca8de57d8ab2a4cacf69fc63bf8a51e41c81b
+size 5859350