roboolympics-fun masters: EO Fun 4fde6ba91c (72 tasks, board 0252), teaser + highlights

masters/roboolympics-fun/alpine_slalom/meta.json

@@ -0,0 +1,140 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "alpine_slalom",
+ "family": "alpine-slalom",
+ "task_dir": "embodied-olympics-fun-alpine-slalom-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra_quick-high-embodied-olympics-fun-alpine-slalom-i00-privileged/embodied-olympics-fun-alpine-sla__eV2gD7X",
+ "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": 820,
+ "seconds": 27.33,
+ "speed": 3.3,
+ "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 (d52b119c47b1deaf)",
+ "end_check": {
+  "success": true,
+  "stages": [
+   "start",
+   "gate 1",
+   "gate 2",
+   "gate 3",
+   "gate 4",
+   "gate 5",
+   "gate 6",
+   "gate 7",
+   "gate 8",
+   "gate 9",
+   "gate 10",
+   "gate 11",
+   "gate 12",
+   "gate 13",
+   "gate 14",
+   "gate 15",
+   "gate 16",
+   "gate 17",
+   "gate 18",
+   "gate 19",
+   "gate 20",
+   "gate 21",
+   "gate 22",
+   "gate 23",
+   "gate 24",
+   "gate 25",
+   "gate 26",
+   "gate 27",
+   "gate 28",
+   "gate 29",
+   "gate 30",
+   "gate 31",
+   "gate 32",
+   "finish"
+  ],
+  "stages_total": 34,
+  "progress": 1.0,
+  "metrics": {
+   "speed_m_s": 5.35,
+   "gates": 32,
+   "time_s": 79.058
+  },
+  "time_s": 83.41,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "0dca23c35f (files 0.3.3)",
+  "final_state_sha": "d52b119c47b1deaf514fcdd4c0b2ac5dfa81b8ff3ca0aa278a1bb90505da9309",
+  "steps": 8341,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra_quick-high-embodied-olympics-fun-alpine-slalom-i00-privileged/embodied-olympics-fun-alpine-sla__eV2gD7X/artifacts/app/output/trajectory.npz",
+  "n_actions": 8341,
+  "control_steps_replayed": 8341,
+  "control_hz": 100.0,
+  "every_control_steps": 11,
+  "final_state_sha": "d52b119c47b1deaf514fcdd4c0b2ac5dfa81b8ff3ca0aa278a1bb90505da9309",
+  "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 house, the start and finish arches, the finish posts and banner (drawn-only, no contact) not drawn: they hid the robot from the chase camera; the safety nets and the gates are drawn"
+ ],
+ "camera": {
+  "mode": "director",
+  "az_deg": -70.61,
+  "az_rule": "robot-travel-70",
+  "el_deg": 24.0,
+  "fovy_deg": 34.0,
+  "fill": 0.72,
+  "distance_m": [
+   6.24,
+   9.304
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 307.125,
+  "focus": {
+   "robot+object": 742,
+   "robot": 18
+  },
+  "objects": {
+   "gate1_turn": 0.0,
+   "gate2_turn": 0.0,
+   "gate3_turn": 0.0,
+   "gate8_turn": 0.0,
+   "gate9_turn": 0.0,
+   "gate17_turn": 0.0,
+   "gate19_turn": 0.0,
+   "gate23_turn": 0.0,
+   "gate24_turn": 0.0,
+   "gate26_turn": 0.0,
+   "gate27_turn": 0.01,
+   "gate29_turn": 0.02,
+   "gate30_turn": 0.04,
+   "gate31_turn": 0.03,
+   "left_ski": 307.43,
+   "right_ski": 307.29
+  }
+ },
+ "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": 4.72,
+ "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), quick-check run",
+ "run": "codex-gpt6_astra_quick-high",
+ "mode": "privileged",
+ "trial_choice": "GPT-6 Astra's privileged success in the quick check of this task version (no formal run on it yet)",
+ "poster": "last frame (solved state)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:37-0500"
+}

masters/roboolympics-fun/alpine_slalom/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43c60360d5509c0445d422f0ff9ae6618a2b2cabc9aa9dc28ab2903cd655c8f9
+size 12693590

masters/roboolympics-fun/backflip/meta.json

@@ -1,17 +1,20 @@
 {
  "suite": "roboolympics-fun",
  "task": "backflip",
- "source": "agent-success:codex-gpt6_astra-high-roboolympics-backflip-i00-privileged/roboolympics-backflip-i00-privil__fuKrrEN",
- "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": "backflip",
+ "task_dir": "embodied-olympics-fun-backflip-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-backflip-i00-privileged/embodied-olympics-fun-backflip-i__L6om7HM",
+ "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": 227,
- "seconds": 7.57,
+ "frames": 237,
+ "seconds": 7.9,
  "speed": 0.6,
  "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 (f5c0b3c390fd3db4)",
+ "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 (c1ad751ccaaf63b9)",
  "end_check": {
   "success": true,
   "stages": [
@@ -23,22 +26,30 @@
   "stages_total": 4,
   "progress": 1.0,
   "metrics": {
-   "turn_deg": 319.3,
-   "air_s": 0.673
+   "turn_deg": 304.2,
+   "air_s": 0.436
   },
-  "time_s": 3.31
+  "time_s": 3.52,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "6109cac419 (files 0.3.2)",
+  "final_state_sha": "c1ad751ccaaf63b91879e4b69137dcd58180235c8bd7ddf39a078bb84bdd4bf1",
+  "steps": 352,
+  "pass": true
  },
  "replay": {
-  "trajectory": "a run of the exam",
-  "n_actions": 331,
-  "control_steps_replayed": 331,
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-backflip-i00-privileged/embodied-olympics-fun-backflip-i__L6om7HM/artifacts/app/output/trajectory.npz",
+  "n_actions": 352,
+  "control_steps_replayed": 352,
   "control_hz": 100.0,
   "every_control_steps": 2,
-  "final_state_sha": "f5c0b3c390fd3db47e93a8f22414ca675082bb9b748945b35436320990730f4c",
+  "final_state_sha": "c1ad751ccaaf63b91879e4b69137dcd58180235c8bd7ddf39a078bb84bdd4bf1",
   "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_*)"
  ],
  "camera": {
   "mode": "director",
@@ -48,30 +59,31 @@
   "fovy_deg": 34.0,
   "fill": 0.72,
   "distance_m": [
-   3.264,
-   3.385
+   3.094,
+   3.358
   ],
   "robot_root": "pelvis",
-  "travel_m": 0.017,
+  "travel_m": 0.813,
   "focus": {
-   "robot": 167
+   "robot": 177
   },
   "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": 2.8,
- "image": "eai-robofunolympics:0.2.1",
+ "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",
  "poster": "last frame (solved state)",
- "host": "gpu0 062523e6f067",
- "encode": "libx264 preset slow CRF 16, yuv420p, bt709",
- "created": "2026-10-18T07:23:31-0500"
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:37-0500"
 }

masters/roboolympics-fun/backflip/video.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a3695ad8018ec20e9298dda93e279bdf920ccfe997a75a631fc86bb8b1e49481
-size 4075675
+oid sha256:f9c39256f792c5d233bc2545f7807b7157727972fe35cb0cea667e6253e12067
+size 4084672

masters/roboolympics-fun/badminton_serve/meta.json

@@ -1,17 +1,20 @@
 {
  "suite": "roboolympics-fun",
  "task": "badminton_serve",
- "source": "agent-success:codex-gpt6_astra-high-roboolympics-fun-badminton-serve-i00-privileged/roboolympics-badminton-serve-i00__D5pDUDF",
- "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": "badminton-serve",
+ "task_dir": "embodied-olympics-fun-badminton-serve-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-badminton-serve-i00-privileged/embodied-olympics-fun-badminton__dU3SuCW",
+ "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": 279,
- "seconds": 9.3,
+ "frames": 359,
+ "seconds": 11.97,
  "speed": 0.3,
  "success_check": true,
- "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 (9afedd98535665a1)",
+ "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 (8ad0c87716c42927)",
  "end_check": {
   "success": true,
   "stages": [
@@ -23,81 +26,71 @@
   "stages_total": 4,
   "progress": 1.0,
   "metrics": {
-   "hit_height_m": 0.4881,
-   "shuttle_speed_m_s": 11.9426,
-   "shuttle_angle_deg": 45.48,
-   "landing_x_m": 5.788,
-   "landing_y_m": 1.716
+   "hit_height_m": 0.7396,
+   "shuttle_speed_m_s": 11.6684,
+   "shuttle_angle_deg": 38.16,
+   "landing_x_m": 5.933,
+   "landing_y_m": 1.554
   },
-  "time_s": 2.18
+  "time_s": 2.98,
+  "failed": null
  },
- "recorded_result": {
-  "verifier_reward_json": {
-   "success": 1,
-   "n_actions": 218,
-   "progress": 1.0,
-   "stages_reached": 4,
-   "stages_total": 4,
-   "metric_hit_height_m": 0.4881,
-   "metric_landing_x_m": 5.788,
-   "metric_landing_y_m": 1.716,
-   "metric_shuttle_angle_deg": 45.48,
-   "metric_shuttle_speed_m_s": 11.9426
-  }
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "6109cac419 (files 0.3.2)",
+  "final_state_sha": "8ad0c87716c429273574da561b752e29645f02dedfe0e43cb6ae80d442ae5be9",
+  "steps": 298,
+  "pass": true
  },
  "replay": {
-  "task_dir": "roboolympics-fun-badminton-serve-i00-privileged",
-  "trajectory": "jobs/codex/gpt6_astra/high/codex-gpt6_astra-high-roboolympics-fun-badminton-serve-i00-privileged/roboolympics-badminton-serve-i00__D5pDUDF/artifacts/app/output/trajectory.npz",
-  "n_actions": 218,
-  "control_steps_replayed": 218,
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-badminton-serve-i00-privileged/embodied-olympics-fun-badminton__dU3SuCW/artifacts/app/output/trajectory.npz",
+  "n_actions": 298,
+  "control_steps_replayed": 298,
   "control_hz": 100.0,
   "every_control_steps": 1,
-  "final_state_sha": "9afedd98535665a1dee70b866bc1073569264d702c09337714ba6088628908f6",
-  "replays_agree": true,
-  "mujoco": "3.3.7",
+  "final_state_sha": "8ad0c87716c429273574da561b752e29645f02dedfe0e43cb6ae80d442ae5be9",
   "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_*)"
  ],
  "camera": {
   "mode": "director",
-  "az_deg": -99.04,
+  "az_deg": -100.81,
   "az_rule": "object-travel-115",
   "el_deg": 22.0,
   "fovy_deg": 34.0,
   "fill": 0.72,
   "distance_m": [
-   4.504,
-   7.156
+   3.576,
+   6.469
   ],
   "robot_root": "pelvis",
-  "travel_m": 0.006,
+  "travel_m": 0.242,
   "focus": {
-   "robot+object": 212,
-   "robot": 7
+   "robot+object": 259,
+   "robot": 40
   },
   "objects": {
-   "shuttle": 5.62,
-   "racket": 0.31
+   "shuttle": 5.81,
+   "racket": 0.3
   }
  },
  "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": 3.43,
- "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-badminton-serve-i00-privileged",
+ "seconds_per_frame": 3.03,
+ "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-badminton-serve-i00-privileged/roboolympics-badminton-serve-i00__D5pDUDF",
- "trial_choice": "a counted GPT-6 Astra success in privileged mode",
+ "trial_choice": "a counted GPT-6 Astra success",
  "poster": "last frame (solved state)",
- "host": "vast.ai RTX 4090 gpu0",
- "encode": "libx264 preset slow CRF 16, yuv420p, bt709, faststart",
- "created": "2026-10-19T06:14:57-0500"
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:38-0500"
 }

masters/roboolympics-fun/badminton_serve/video.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:53d49e8d1404649fbd1331dffae3745cae61bbd1f82ff29e816261746aa4860a
-size 5400934
+oid sha256:e139dff6733b24ac6bfc6a8baf4b85417431909909ba556a02ee8e95e90cb815
+size 6319444

masters/roboolympics-fun/balance_beam/meta.json

@@ -0,0 +1,88 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "balance_beam",
+ "family": "balance-beam",
+ "task_dir": "embodied-olympics-fun-balance-beam-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-balance-beam-i00-privileged/embodied-olympics-fun-balance-be__FJ5XCVd",
+ "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": 780,
+ "seconds": 26.0,
+ "speed": 3.0,
+ "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 (1dbc37fa79cb9961)",
+ "end_check": {
+  "success": true,
+  "stages": [
+   "half way",
+   "far end",
+   "stand"
+  ],
+  "stages_total": 3,
+  "progress": 1.0,
+  "metrics": {
+   "distance_m": 4.541,
+   "time_s": 71.89
+  },
+  "time_s": 71.89,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "6109cac419 (files 0.3.2)",
+  "final_state_sha": "1dbc37fa79cb99619dea2129218d3366ca0592f3dcdf9d80838046ae0072e3b0",
+  "steps": 7189,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-balance-beam-i00-privileged/embodied-olympics-fun-balance-be__FJ5XCVd/artifacts/app/output/trajectory.npz",
+  "n_actions": 7189,
+  "control_steps_replayed": 7189,
+  "control_hz": 100.0,
+  "every_control_steps": 10,
+  "final_state_sha": "1dbc37fa79cb99619dea2129218d3366ca0592f3dcdf9d80838046ae0072e3b0",
+  "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": -70.02,
+  "az_rule": "robot-travel-70",
+  "el_deg": 22.0,
+  "fovy_deg": 34.0,
+  "fill": 0.72,
+  "distance_m": [
+   3.087,
+   3.22
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 4.451,
+  "focus": {
+   "robot": 720
+  },
+  "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": 2.68,
+ "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",
+ "poster": "last frame (solved state)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:38-0500"
+}

masters/roboolympics-fun/balance_beam/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3cc3f0294c632d4838dc59f7d742a0875ebc3c2ed19d141e1a16a1aeb0ea08b
+size 11572565

masters/roboolympics-fun/balance_board/meta.json

@@ -0,0 +1,91 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "balance_board",
+ "family": "balance-board",
+ "task_dir": "embodied-olympics-fun-balance-board-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-balance-board-i00-privileged/embodied-olympics-fun-balance-bo__RkHCQBb",
+ "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": 811,
+ "seconds": 27.03,
+ "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 (1d23a15ed5f9c597)",
+ "end_check": {
+  "success": true,
+  "stages": [
+   "10 s",
+   "20 s",
+   "30 s"
+  ],
+  "stages_total": 3,
+  "progress": 1.0,
+  "metrics": {
+   "max_tilt_deg": 10.15,
+   "time_on_board_s": 30.0
+  },
+  "time_s": 30.0,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "6109cac419 (files 0.3.2)",
+  "final_state_sha": "1d23a15ed5f9c597c27a2c93f86851b09296a0c5f0c37554600b31af5019f914",
+  "steps": 3000,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-balance-board-i00-privileged/embodied-olympics-fun-balance-bo__RkHCQBb/artifacts/app/output/trajectory.npz",
+  "n_actions": 3000,
+  "control_steps_replayed": 3000,
+  "control_hz": 100.0,
+  "every_control_steps": 4,
+  "final_state_sha": "1d23a15ed5f9c597c27a2c93f86851b09296a0c5f0c37554600b31af5019f914",
+  "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": [
+   3.737,
+   3.927
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 0.039,
+  "focus": {
+   "robot": 751
+  },
+  "objects": {
+   "roller": 0.12,
+   "deck": 0.23
+  }
+ },
+ "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": 2.58,
+ "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": "last frame (solved state)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:38-0500"
+}

masters/roboolympics-fun/balance_board/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b5671816567883b76efe463240635c7a0e9180c6da0177c90d62a908a532a637
+size 8246896

masters/roboolympics-fun/basketball_dribble/meta.json

@@ -0,0 +1,93 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "basketball_dribble",
+ "family": "basketball-dribble",
+ "task_dir": "embodied-olympics-fun-basketball-dribble-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra_quick-high-embodied-olympics-fun-basketball-dribble-i00-privileged/embodied-olympics-fun-basketball__yqWK2fP",
+ "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": 565,
+ "seconds": 18.83,
+ "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 (75298314bd477697)",
+ "end_check": {
+  "success": true,
+  "stages": [
+   "dribble",
+   "past the free-throw line",
+   "in the frontcourt"
+  ],
+  "stages_total": 3,
+  "progress": 1.0,
+  "metrics": {
+   "distance_m": 14.358,
+   "ball_height_m": 0.146,
+   "dribbles": 28
+  },
+  "time_s": 15.11,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "0dca23c35f (files 0.3.3)",
+  "final_state_sha": "75298314bd4776976ff97d2b893e81d3817bf85a109132d1b1c384e617e77909",
+  "steps": 1511,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra_quick-high-embodied-olympics-fun-basketball-dribble-i00-privileged/embodied-olympics-fun-basketball__yqWK2fP/artifacts/app/output/trajectory.npz",
+  "n_actions": 1511,
+  "control_steps_replayed": 1511,
+  "control_hz": 100.0,
+  "every_control_steps": 3,
+  "final_state_sha": "75298314bd4776976ff97d2b893e81d3817bf85a109132d1b1c384e617e77909",
+  "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_*)",
+  "the ball drawn as a basketball (orange, black channels)"
+ ],
+ "camera": {
+  "mode": "director",
+  "az_deg": -67.0,
+  "az_rule": "robot-travel-70",
+  "el_deg": 22.0,
+  "fovy_deg": 34.0,
+  "fill": 0.72,
+  "distance_m": [
+   4.298,
+   4.388
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 13.877,
+  "focus": {
+   "robot+object": 503,
+   "robot": 2
+  },
+  "objects": {
+   "ball": 13.9
+  }
+ },
+ "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": 2.86,
+ "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), quick-check run",
+ "run": "codex-gpt6_astra_quick-high",
+ "mode": "privileged",
+ "trial_choice": "GPT-6 Astra's privileged success in the quick check of this task version (no formal run on it yet)",
+ "poster": "last frame (solved state)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:38-0500"
+}

masters/roboolympics-fun/basketball_dribble/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ba3ff70303fe1825d1f8c4488897762a9a6789a7cc4399e4eff0cb596e96077
+size 11305885

masters/roboolympics-fun/bowling/meta.json

@@ -1,17 +1,20 @@
 {
  "suite": "roboolympics-fun",
  "task": "bowling",
- "source": "agent-success:codex-gpt6_luna-xhigh-roboolympics-bowling-i00-privileged/roboolympics-bowling-i00-privile__wK9Ujtv",
- "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": "bowling",
+ "task_dir": "embodied-olympics-fun-bowling-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-bowling-i00-privileged/embodied-olympics-fun-bowling-i0__nD6fgoT",
+ "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": 427,
- "seconds": 14.23,
+ "frames": 402,
+ "seconds": 13.4,
  "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; a second plain replay with the task's own replay() agrees and ends in the same final state hash (369fbd0f61bc2aa9)",
+ "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 (e0df2c83e0638eca)",
  "end_check": {
   "success": true,
   "stages": [
@@ -24,72 +27,81 @@
   "metrics": {
    "pins_down": 10,
    "release_speed_m_s": 5.06,
-   "entry_speed_m_s": 3.0,
-   "entry_y_m": -0.058,
-   "entry_angle_deg": 2.25
+   "entry_speed_m_s": 3.83,
+   "entry_y_m": -0.12,
+   "entry_angle_deg": -0.47
   },
-  "time_s": 10.98
+  "time_s": 10.22,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "6109cac419 (files 0.3.2)",
+  "final_state_sha": "e0df2c83e0638ecad98881b4688e1c53d1f4714ece2174affdbfcefd26c5fcdd",
+  "steps": 1022,
+  "pass": true
  },
  "replay": {
-  "trajectory": "a run of the exam",
-  "n_actions": 1098,
-  "control_steps_replayed": 1098,
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-bowling-i00-privileged/embodied-olympics-fun-bowling-i0__nD6fgoT/artifacts/app/output/trajectory.npz",
+  "n_actions": 1022,
+  "control_steps_replayed": 1022,
   "control_hz": 100.0,
   "every_control_steps": 3,
-  "final_state_sha": "369fbd0f61bc2aa9631a38b6bd217be0232e741da17cb30936ce994358f45960",
+  "final_state_sha": "e0df2c83e0638ecad98881b4688e1c53d1f4714ece2174affdbfcefd26c5fcdd",
   "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_*)"
  ],
  "camera": {
   "mode": "director",
-  "az_deg": -115.86,
+  "az_deg": -116.4,
   "az_rule": "object-travel-115",
   "el_deg": 22.0,
   "fovy_deg": 34.0,
   "fill": 0.72,
   "distance_m": [
-   4.599,
-   35.626
+   4.697,
+   36.677
   ],
   "robot_root": "pelvis",
-  "travel_m": 0.128,
+  "travel_m": 0.086,
   "focus": {
-   "robot": 16,
-   "robot(incoming)": 3,
-   "object": 201,
-   "robot+object": 71,
-   "robot+landed": 76
+   "robot+landed": 81,
+   "robot+object": 49,
+   "robot": 29,
+   "object": 173,
+   "robot(incoming)": 10
   },
   "objects": {
-   "pin1": 0.55,
-   "pin2": 0.52,
-   "pin3": 0.93,
-   "pin4": 0.87,
-   "pin5": 0.72,
-   "pin6": 0.32,
-   "pin7": 0.17,
-   "pin8": 0.87,
-   "pin9": 0.64,
-   "pin10": 0.84,
-   "ball": 19.71
+   "pin1": 0.59,
+   "pin2": 0.28,
+   "pin3": 0.44,
+   "pin4": 0.56,
+   "pin5": 0.32,
+   "pin6": 0.55,
+   "pin7": 0.43,
+   "pin8": 0.32,
+   "pin9": 0.39,
+   "pin10": 0.48,
+   "ball": 20.16
   }
  },
  "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": 4.45,
- "image": "eai-robofunolympics:0.2.1",
- "agent": "codex + GPT-6 Luna (xhigh)",
- "run": "codex-gpt6_luna-xhigh",
+ "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": "no counted Astra success; a counted success of another run",
- "poster": "frame 46: just after the release / strike, robot and object in view (the last frame no longer shows the robot)",
- "host": "gpu0 062523e6f067",
- "encode": "libx264 preset slow CRF 16, yuv420p, bt709",
- "created": "2026-10-18T07:23:32-0500"
+ "trial_choice": "a counted GPT-6 Astra success",
+ "poster": "frame 38 (same scene, camera and settings as the video)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:38-0500"
 }

masters/roboolympics-fun/bowling/video.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1ef37f2a4b8868940e9286310b7aac43301b9a63434d806e6d1070997a19461
-size 7391936
+oid sha256:d600112ba6c9fa9ff5651698830de3ad4142a4eda94e1ae2811c7972ad859a44
+size 7445595

masters/roboolympics-fun/box_jump/meta.json

@@ -0,0 +1,91 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "box_jump",
+ "family": "box-jump",
+ "task_dir": "embodied-olympics-fun-box-jump-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-box-jump-i00-privileged/embodied-olympics-fun-box-jump-i__o9U84U8",
+ "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": 244,
+ "seconds": 8.13,
+ "speed": 0.6,
+ "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 (7dba1a5cc42e13fe)",
+ "end_check": {
+  "success": true,
+  "stages": [
+   "take-off",
+   "landing",
+   "stand"
+  ],
+  "stages_total": 3,
+  "progress": 1.0,
+  "metrics": {
+   "air_s": 0.514,
+   "pelvis_rise_m": 0.531,
+   "pelvis_above_box_m": 1.027
+  },
+  "time_s": 3.65,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "6109cac419 (files 0.3.2)",
+  "final_state_sha": "7dba1a5cc42e13fe30558cf7684a49b635f8d495b2317dcff26d53cd3d7bee7e",
+  "steps": 365,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-box-jump-i00-privileged/embodied-olympics-fun-box-jump-i__o9U84U8/artifacts/app/output/trajectory.npz",
+  "n_actions": 365,
+  "control_steps_replayed": 365,
+  "control_hz": 100.0,
+  "every_control_steps": 2,
+  "final_state_sha": "7dba1a5cc42e13fe30558cf7684a49b635f8d495b2317dcff26d53cd3d7bee7e",
+  "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": [
+   5.354,
+   5.638
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 0.958,
+  "focus": {
+   "robot": 184
+  },
+  "objects": {
+   "box": 0.0
+  }
+ },
+ "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",
+ "poster": "last frame (solved state)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:38-0500"
+}

masters/roboolympics-fun/box_jump/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c5b723c54603a03565367ec78d3927231dd6646a899e4ce70bc415911ed4370c
+size 3383986

masters/roboolympics-fun/boxing/meta.json

@@ -1,17 +1,20 @@
 {
  "suite": "roboolympics-fun",
  "task": "boxing",
- "source": "agent-success:codex-gpt6_luna-xhigh-roboolympics-boxing-i00-standard/roboolympics-boxing-i00-standard__a7gYtdP",
- "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": "boxing",
+ "task_dir": "embodied-olympics-fun-boxing-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-boxing-i00-privileged/embodied-olympics-fun-boxing-i00__pTjF4SX",
+ "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": 302,
- "seconds": 10.07,
+ "frames": 257,
+ "seconds": 8.57,
  "speed": 0.6,
  "success_check": true,
- "success_check_how": "the task's own verdict (environment/sim/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 (7ac9e6b973578113)",
+ "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 (bee9e7f8378fcca5)",
  "end_check": {
   "success": true,
   "stages": [
@@ -21,24 +24,32 @@
   "stages_total": 2,
   "progress": 1.0,
   "metrics": {
-   "hit_height_m": 1.231,
-   "impulse_n_s": 13.39,
-   "contact_impulse_n_s": 15.12,
-   "contact_s": 0.089
+   "hit_height_m": 1.4,
+   "impulse_n_s": 20.64,
+   "contact_impulse_n_s": 26.35,
+   "contact_s": 0.059
   },
-  "time_s": 4.82
+  "time_s": 3.92,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "6109cac419 (files 0.3.2)",
+  "final_state_sha": "bee9e7f8378fcca5da7f32d97f4377381ec226b82dd2e0695270f802ee6a5550",
+  "steps": 392,
+  "pass": true
  },
  "replay": {
-  "trajectory": "a run of the exam",
-  "n_actions": 482,
-  "control_steps_replayed": 482,
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-boxing-i00-privileged/embodied-olympics-fun-boxing-i00__pTjF4SX/artifacts/app/output/trajectory.npz",
+  "n_actions": 392,
+  "control_steps_replayed": 392,
   "control_hz": 100.0,
   "every_control_steps": 2,
-  "final_state_sha": "7ac9e6b973578113abfa0fd35a2cb6f66ad6d75ef834e83ec2ca8ada36fc449e",
+  "final_state_sha": "bee9e7f8378fcca5da7f32d97f4377381ec226b82dd2e0695270f802ee6a5550",
   "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_*)"
  ],
  "camera": {
   "mode": "director",
@@ -48,32 +59,33 @@
   "fovy_deg": 34.0,
   "fill": 0.72,
   "distance_m": [
-   4.919,
-   5.205
+   4.872,
+   5.158
   ],
   "robot_root": "pelvis",
-  "travel_m": 0.489,
+  "travel_m": 0.762,
   "focus": {
-   "robot": 242
+   "robot": 197
   },
   "objects": {
-   "bag": 0.06
+   "bag": 0.04
   }
  },
  "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": 3.79,
- "image": "eai-robofunolympics:0.2.1",
- "agent": "codex + GPT-6 Luna (xhigh)",
- "run": "codex-gpt6_luna-xhigh",
- "mode": "standard",
- "trial_choice": "no counted Astra success; a counted success of another run",
+ "seconds_per_frame": 2.66,
+ "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": "last frame (solved state)",
- "host": "gpu1 062523e6f067",
- "encode": "libx264 preset slow CRF 16, yuv420p, bt709",
- "created": "2026-10-18T07:23:32-0500"
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:38-0500"
 }

masters/roboolympics-fun/boxing/video.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f2d2fb7df606b44637eaaad02b05369aa8cbb35de0739cb61c008d3765d4728c
-size 4264670
+oid sha256:ed2d114a79d4593712748a225191dec068cfbe47d9c4ff2cd0850c0c56015666
+size 4041069

masters/roboolympics-fun/breaking_windmill/meta.json

@@ -0,0 +1,89 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "breaking_windmill",
+ "family": "breaking-windmill",
+ "task_dir": "embodied-olympics-fun-breaking-windmill-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "scene-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_quick-high-embodied-olympics-fun-breaking-windmill-i00-privileged/embodied-olympics-fun-breaking-w__mxtoRx8",
+ "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": 748,
+ "seconds": 24.93,
+ "speed": 0.9,
+ "success_check": false,
+ "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 (b35bdb9e0bd9d696)",
+ "end_check": {
+  "success": false,
+  "stages": [
+   "down",
+   "1 windmill",
+   "2 windmills"
+  ],
+  "stages_total": 4,
+  "progress": 0.75,
+  "metrics": {
+   "windmills": 2
+  },
+  "time_s": 20.6,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "0dca23c35f (files 0.3.3)",
+  "final_state_sha": "b35bdb9e0bd9d69658ac77d30d57ad97af8c5903ab8b883e8d1af4b46d77e9f2",
+  "steps": 2060,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra_quick-high-embodied-olympics-fun-breaking-windmill-i00-privileged/embodied-olympics-fun-breaking-w__mxtoRx8/artifacts/app/output/trajectory.npz",
+  "n_actions": 2060,
+  "control_steps_replayed": 2060,
+  "control_hz": 100.0,
+  "every_control_steps": 3,
+  "final_state_sha": "b35bdb9e0bd9d69658ac77d30d57ad97af8c5903ab8b883e8d1af4b46d77e9f2",
+  "original_verdict": "counted failure (verifier reward 0)"
+ },
+ "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": [
+   2.072,
+   3.389
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 0.655,
+  "focus": {
+   "robot": 688
+  },
+  "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": 2.64,
+ "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), quick-check run",
+ "run": "codex-gpt6_astra_quick-high",
+ "mode": "privileged",
+ "trial_choice": "no success on this task version; GPT-6 Astra's failed attempt (most progress)",
+ "poster": "last frame (end of the failed attempt)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:39-0500"
+}

masters/roboolympics-fun/breaking_windmill/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d15b4138e5fae82de7c099ee1833b5ff6ead3771b547b78dffd8e394cc4d646f
+size 15904376

masters/roboolympics-fun/catch/meta.json

@@ -1,17 +1,20 @@
 {
  "suite": "roboolympics-fun",
  "task": "catch",
- "source": "agent-success:codex-gpt6_luna-xhigh-roboolympics-catch-i00-privileged/roboolympics-catch-i00-privilege__n9GbsX7",
- "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": "catch",
+ "task_dir": "embodied-olympics-fun-catch-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra_quick-high-embodied-olympics-fun-catch-i00-privileged/embodied-olympics-fun-catch-i00__qBKtgfp",
+ "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": 306,
- "seconds": 10.2,
- "speed": 0.6,
+ "frames": 227,
+ "seconds": 7.57,
+ "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; a second plain replay with the task's own replay() agrees and ends in the same final state hash (738f01722d4c5c4f)",
+ "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 (7143bce9fa2d82ad)",
  "end_check": {
   "success": true,
   "stages": [
@@ -21,58 +24,68 @@
   "stages_total": 2,
   "progress": 1.0,
   "metrics": {
-   "catch_height_m": 1.186
+   "catch_height_m": 1.042,
+   "catch_speed_m_s": 9.938
   },
-  "time_s": 4.9
+  "time_s": 4.96,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "0dca23c35f (files 0.3.3)",
+  "final_state_sha": "7143bce9fa2d82ad900e6e5d780b651fc77a8b98cdc4c93ba9633157e3447da7",
+  "steps": 496,
+  "pass": true
  },
  "replay": {
-  "trajectory": "a run of the exam",
-  "n_actions": 490,
-  "control_steps_replayed": 490,
+  "trajectory": "codex-gpt6_astra_quick-high-embodied-olympics-fun-catch-i00-privileged/embodied-olympics-fun-catch-i00__qBKtgfp/artifacts/app/output/trajectory.npz",
+  "n_actions": 496,
+  "control_steps_replayed": 496,
   "control_hz": 100.0,
-  "every_control_steps": 2,
-  "final_state_sha": "738f01722d4c5c4f0582593ef988a9d1960dd893a892e62ebb2a91046478b10f",
+  "every_control_steps": 3,
+  "final_state_sha": "7143bce9fa2d82ad900e6e5d780b651fc77a8b98cdc4c93ba9633157e3447da7",
   "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_*)"
  ],
  "camera": {
   "mode": "director",
-  "az_deg": -89.44,
-  "az_rule": "robot-travel-70",
+  "az_deg": -293.81,
+  "az_rule": "object-travel-115",
   "el_deg": 22.0,
   "fovy_deg": 34.0,
   "fill": 0.72,
   "distance_m": [
-   4.446,
-   17.158
+   4.356,
+   16.306
   ],
   "robot_root": "pelvis",
-  "travel_m": 2.245,
+  "travel_m": 1.66,
   "focus": {
-   "robot(incoming)": 124,
-   "robot+object": 71,
-   "robot": 51
+   "robot+object": 48,
+   "robot": 33,
+   "robot(incoming)": 86
   },
   "objects": {
-   "ball": 37.58
+   "ball": 38.23
   }
  },
  "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": 8.15,
- "image": "eai-robofunolympics:0.2.1",
- "agent": "codex + GPT-6 Luna (xhigh)",
- "run": "codex-gpt6_luna-xhigh",
+ "seconds_per_frame": 3.92,
+ "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), quick-check run",
+ "run": "codex-gpt6_astra_quick-high",
  "mode": "privileged",
- "trial_choice": "no counted Astra success; a counted success of another run",
+ "trial_choice": "GPT-6 Astra's privileged success in the quick check of this task version (no formal run on it yet)",
  "poster": "last frame (solved state)",
- "host": "gpu2 062523e6f067",
- "encode": "libx264 preset slow CRF 16, yuv420p, bt709",
- "created": "2026-10-18T07:23:32-0500"
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:39-0500"
 }

masters/roboolympics-fun/catch/video.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eed355e941a9d4720fa02f51106a13cb45d2b49c8e06ab3ebb9dff0408e6b8ca
-size 4733797
+oid sha256:4fcf5e3a498b657b438460f17ac9e231f2076972e0ab9d577e309b8fcd5d895b
+size 3564063

masters/roboolympics-fun/catch_and_shoot/meta.json

@@ -0,0 +1,92 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "catch_and_shoot",
+ "family": "catch-and-shoot",
+ "task_dir": "embodied-olympics-fun-catch-and-shoot-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra_quick-high-embodied-olympics-fun-catch-and-shoot-i00-privileged/embodied-olympics-fun-catch-and__GZgBxmp",
+ "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": 282,
+ "seconds": 9.4,
+ "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 (c4b69de29627ac7b)",
+ "end_check": {
+  "success": true,
+  "stages": [
+   "catch",
+   "release",
+   "through the ring"
+  ],
+  "stages_total": 3,
+  "progress": 1.0,
+  "metrics": {
+   "ball_height_m": 2.775
+  },
+  "time_s": 6.61,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "0dca23c35f (files 0.3.3)",
+  "final_state_sha": "c4b69de29627ac7bfb36bec68f4b47ea21a2275a4b0a6221f7bfdfd4579bdb91",
+  "steps": 661,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra_quick-high-embodied-olympics-fun-catch-and-shoot-i00-privileged/embodied-olympics-fun-catch-and__GZgBxmp/artifacts/app/output/trajectory.npz",
+  "n_actions": 661,
+  "control_steps_replayed": 661,
+  "control_hz": 100.0,
+  "every_control_steps": 3,
+  "final_state_sha": "c4b69de29627ac7bfb36bec68f4b47ea21a2275a4b0a6221f7bfdfd4579bdb91",
+  "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_*)",
+  "the ball drawn as a basketball (orange, black channels)"
+ ],
+ "camera": {
+  "mode": "director",
+  "az_deg": -155.62,
+  "az_rule": "object-travel-115",
+  "el_deg": 22.0,
+  "fovy_deg": 34.0,
+  "fill": 0.72,
+  "distance_m": [
+   5.886,
+   16.553
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 0.024,
+  "focus": {
+   "robot+object": 217,
+   "robot": 5
+  },
+  "objects": {
+   "passer": 0.03,
+   "ball": 6.5
+  }
+ },
+ "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": 2.91,
+ "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), quick-check run",
+ "run": "codex-gpt6_astra_quick-high",
+ "mode": "privileged",
+ "trial_choice": "GPT-6 Astra's privileged success in the quick check of this task version (no formal run on it yet)",
+ "poster": "frame 162 (same scene, camera and settings as the video)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:39-0500"
+}

masters/roboolympics-fun/catch_and_shoot/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28001c0fb9bd2e9d63658dc845f8755d317c2451ceacb388f69b45f908a86cda
+size 5684331

masters/roboolympics-fun/clean_and_jerk/meta.json

@@ -0,0 +1,95 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "clean_and_jerk",
+ "family": "clean-and-jerk",
+ "task_dir": "embodied-olympics-fun-clean-and-jerk-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra_quick-high-embodied-olympics-fun-clean-and-jerk-i00-privileged/embodied-olympics-fun-clean-and__HRohaub",
+ "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": 387,
+ "seconds": 12.9,
+ "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 (e574484c3443a628)",
+ "end_check": {
+  "success": true,
+  "stages": [
+   "lift-off",
+   "clean",
+   "overhead",
+   "down signal"
+  ],
+  "stages_total": 4,
+  "progress": 1.0,
+  "metrics": {
+   "bar_height_m": 2.122,
+   "bar_max_height_m": 2.123,
+   "elbow_flex_deg": 0.3,
+   "knee_flex_deg": 3.1,
+   "time_s": 9.76
+  },
+  "time_s": 9.76,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "0dca23c35f (files 0.3.3)",
+  "final_state_sha": "e574484c3443a628a9ff27f8df93ad0fbb777b314b89318a1bc1430d6647c76f",
+  "steps": 976,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra_quick-high-embodied-olympics-fun-clean-and-jerk-i00-privileged/embodied-olympics-fun-clean-and__HRohaub/artifacts/app/output/trajectory.npz",
+  "n_actions": 976,
+  "control_steps_replayed": 976,
+  "control_hz": 100.0,
+  "every_control_steps": 3,
+  "final_state_sha": "e574484c3443a628a9ff27f8df93ad0fbb777b314b89318a1bc1430d6647c76f",
+  "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": [
+   4.151,
+   6.408
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 0.076,
+  "focus": {
+   "robot": 191,
+   "robot+object": 136
+  },
+  "objects": {
+   "bar": 0.21
+  }
+ },
+ "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": 2.64,
+ "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), quick-check run",
+ "run": "codex-gpt6_astra_quick-high",
+ "mode": "privileged",
+ "trial_choice": "GPT-6 Astra's privileged success in the quick check of this task version (no formal run on it yet)",
+ "poster": "last frame (solved state)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:39-0500"
+}

masters/roboolympics-fun/clean_and_jerk/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad7cdfce402d206c89478423922feb7322dbd479f5e3da34ebceb3b96caa2b52
+size 5720370

masters/roboolympics-fun/crawl/meta.json

@@ -1,17 +1,20 @@
 {
  "suite": "roboolympics-fun",
  "task": "crawl",
- "source": "agent-success:codex-gpt6_luna-xhigh-roboolympics-crawl-i00-privileged/roboolympics-crawl-i00-privilege__Che2uUd",
- "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": "crawl",
+ "task_dir": "embodied-olympics-fun-crawl-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-crawl-i00-privileged/embodied-olympics-fun-crawl-i00__YDvTC8q",
+ "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": 806,
- "seconds": 26.87,
- "speed": 2.7,
+ "frames": 815,
+ "seconds": 27.17,
+ "speed": 3.0,
  "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 (e3c198584948a82e)",
+ "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 (faa48f5cc065243c)",
  "end_check": {
   "success": true,
   "stages": [
@@ -22,55 +25,64 @@
   "stages_total": 3,
   "progress": 1.0,
   "metrics": {
-   "rear_x_m": 11.004,
-   "time_s": 67.01
+   "rear_x_m": 11.029,
+   "time_s": 75.38
   },
-  "time_s": 67.01
+  "time_s": 75.38,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "6109cac419 (files 0.3.2)",
+  "final_state_sha": "faa48f5cc065243cb786397eb654755d85ac321ded2ad43b197fc3e384130c93",
+  "steps": 7538,
+  "pass": true
  },
  "replay": {
-  "trajectory": "a run of the exam",
-  "n_actions": 6701,
-  "control_steps_replayed": 6701,
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-crawl-i00-privileged/embodied-olympics-fun-crawl-i00__YDvTC8q/artifacts/app/output/trajectory.npz",
+  "n_actions": 7538,
+  "control_steps_replayed": 7538,
   "control_hz": 100.0,
-  "every_control_steps": 9,
-  "final_state_sha": "e3c198584948a82ed464af7e6034f377f6d997d95e75dd3b591284913baa82b5",
+  "every_control_steps": 10,
+  "final_state_sha": "faa48f5cc065243cb786397eb654755d85ac321ded2ad43b197fc3e384130c93",
   "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_*)"
  ],
  "camera": {
   "mode": "director",
-  "az_deg": -68.56,
+  "az_deg": -70.09,
   "az_rule": "robot-travel-70",
   "el_deg": 22.0,
   "fovy_deg": 34.0,
   "fill": 0.72,
   "distance_m": [
-   1.64,
-   3.345
+   2.141,
+   3.303
   ],
   "robot_root": "pelvis",
-  "travel_m": 11.186,
+  "travel_m": 11.431,
   "focus": {
-   "robot": 746
+   "robot": 755
   },
   "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": 4.58,
- "image": "eai-robofunolympics:0.2.1",
- "agent": "codex + GPT-6 Luna (xhigh)",
- "run": "codex-gpt6_luna-xhigh",
+ "seconds_per_frame": 2.81,
+ "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": "no counted Astra success; a counted success of another run",
+ "trial_choice": "a counted GPT-6 Astra success",
  "poster": "last frame (solved state)",
- "host": "gpu0 062523e6f067",
- "encode": "libx264 preset slow CRF 16, yuv420p, bt709",
- "created": "2026-10-18T07:23:32-0500"
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:39-0500"
 }

masters/roboolympics-fun/crawl/video.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2c4a29b327feff7223f85ae06c8e6dcf23d418a9c17722cc6a525a39c1d4d9de
-size 31976838
+oid sha256:fb0a674a9f857c644ae9c9df03b7d39e26d6e2baa09326728770e56e1417aa14
+size 28413111

masters/roboolympics-fun/discus/meta.json

@@ -1,17 +1,20 @@
 {
  "suite": "roboolympics-fun",
  "task": "discus",
- "source": "agent-success:codex-gpt6_astra-high-roboolympics-discus-i00-privileged/roboolympics-discus-i00-privileg__rnFDUxc",
- "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": "discus",
+ "task_dir": "embodied-olympics-fun-discus-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra_quick-high-embodied-olympics-fun-discus-i00-privileged/embodied-olympics-fun-discus-i00__Uhxr9X9",
+ "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": 332,
- "seconds": 11.07,
+ "frames": 289,
+ "seconds": 9.63,
  "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; a second plain replay with the task's own replay() agrees and ends in the same final state hash (01d601fa5a0b19e1)",
+ "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 (fb017b6a0c005f22)",
  "end_check": {
   "success": true,
   "stages": [
@@ -22,62 +25,72 @@
   "stages_total": 3,
   "progress": 1.0,
   "metrics": {
-   "release_speed_m_s": 15.58,
-   "release_height_m": 1.463,
-   "release_tilt_deg": 24.3,
-   "release_spin_rad_s": 27.8,
-   "distance_m": 21.73
+   "release_time_s": 1.118,
+   "release_speed_m_s": 12.68,
+   "release_height_m": 2.087,
+   "release_tilt_deg": 71.7,
+   "release_spin_rad_s": 24.4,
+   "distance_m": 16.267
   },
-  "time_s": 8.11
+  "time_s": 6.83,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "0dca23c35f (files 0.3.3)",
+  "final_state_sha": "fb017b6a0c005f2289db7a3c21dd584ce7e7760fecfa933897e2c74709c462af",
+  "steps": 683,
+  "pass": true
  },
  "replay": {
-  "trajectory": "a run of the exam",
-  "n_actions": 811,
-  "control_steps_replayed": 811,
+  "trajectory": "codex-gpt6_astra_quick-high-embodied-olympics-fun-discus-i00-privileged/embodied-olympics-fun-discus-i00__Uhxr9X9/artifacts/app/output/trajectory.npz",
+  "n_actions": 683,
+  "control_steps_replayed": 683,
   "control_hz": 100.0,
   "every_control_steps": 3,
-  "final_state_sha": "01d601fa5a0b19e1e9b352392b51dc840637256ae9ef2faf0d302bffdac3f53f",
+  "final_state_sha": "fb017b6a0c005f2289db7a3c21dd584ce7e7760fecfa933897e2c74709c462af",
   "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_*)"
  ],
  "camera": {
   "mode": "director",
-  "az_deg": -105.14,
+  "az_deg": -64.76,
   "az_rule": "object-travel-115",
   "el_deg": 22.0,
   "fovy_deg": 34.0,
   "fill": 0.72,
   "distance_m": [
-   4.527,
-   17.207
+   5.578,
+   14.867
   ],
   "robot_root": "pelvis",
-  "travel_m": 1.495,
+  "travel_m": 1.332,
   "focus": {
-   "robot": 8,
-   "robot+object": 103,
-   "object": 161
+   "robot": 7,
+   "object": 171,
+   "robot+object": 51
   },
   "objects": {
-   "discus": 45.2
+   "discus": 25.02
   }
  },
  "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": 4.32,
- "image": "eai-robofunolympics:0.2.1",
- "agent": "codex + GPT-6 Astra (high)",
- "run": "codex-gpt6_astra-high",
+ "seconds_per_frame": 2.7,
+ "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), quick-check run",
+ "run": "codex-gpt6_astra_quick-high",
  "mode": "privileged",
- "trial_choice": "a counted GPT-6 Astra success",
- "poster": "frame 97: just after the release / strike, robot and object in view (the last frame no longer shows the robot)",
- "host": "gpu3 062523e6f067",
- "encode": "libx264 preset slow CRF 16, yuv420p, bt709",
- "created": "2026-10-18T07:23:33-0500"
+ "trial_choice": "GPT-6 Astra's privileged success in the quick check of this task version (no formal run on it yet)",
+ "poster": "frame 41 (same scene, camera and settings as the video)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:40-0500"
 }

masters/roboolympics-fun/discus/video.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:caf6cca967fa537c1ef73468aaa419ff5225b5d19cd67ecf7fc39ead80519913
-size 6489823
+oid sha256:898a064a6519084d2ce9a32d2cda3f4779613d68916e3ecbd401592011041d09
+size 6559398

masters/roboolympics-fun/dodgeball/meta.json

@@ -0,0 +1,103 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "dodgeball",
+ "family": "dodgeball",
+ "task_dir": "embodied-olympics-fun-dodgeball-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra_quick-high-embodied-olympics-fun-dodgeball-i00-privileged/embodied-olympics-fun-dodgeball__sZ5uVQa",
+ "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": 497,
+ "seconds": 16.57,
+ "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 (33db8589fff69e88)",
+ "end_check": {
+  "success": true,
+  "stages": [
+   "dodge 1",
+   "dodge 2",
+   "dodge 3",
+   "dodge 4",
+   "dodge 5"
+  ],
+  "stages_total": 5,
+  "progress": 1.0,
+  "metrics": {
+   "dodged": 5
+  },
+  "time_s": 13.06,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "0dca23c35f (files 0.3.3)",
+  "final_state_sha": "33db8589fff69e880eb1e6125ee601631f2b646ad4125da63a52e9fd967fcbe9",
+  "steps": 1306,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra_quick-high-embodied-olympics-fun-dodgeball-i00-privileged/embodied-olympics-fun-dodgeball__sZ5uVQa/artifacts/app/output/trajectory.npz",
+  "n_actions": 1306,
+  "control_steps_replayed": 1306,
+  "control_hz": 100.0,
+  "every_control_steps": 3,
+  "final_state_sha": "33db8589fff69e880eb1e6125ee601631f2b646ad4125da63a52e9fd967fcbe9",
+  "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": 25.39,
+  "az_rule": "robot-travel-70",
+  "el_deg": 22.0,
+  "fovy_deg": 34.0,
+  "fill": 0.72,
+  "distance_m": [
+   7.133,
+   19.085
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 2.383,
+  "focus": {
+   "robot+object": 277,
+   "robot": 1,
+   "object": 106,
+   "robot(incoming)": 53
+  },
+  "objects": {
+   "ball1": 20.24,
+   "turret1_yaw": 0.03,
+   "ball2": 20.02,
+   "turret2_yaw": 0.03,
+   "ball3": 20.18,
+   "turret3_yaw": 0.03,
+   "ball4": 18.31,
+   "turret4_yaw": 0.03,
+   "ball5": 9.92,
+   "turret5_yaw": 0.03
+  }
+ },
+ "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": 2.64,
+ "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), quick-check run",
+ "run": "codex-gpt6_astra_quick-high",
+ "mode": "privileged",
+ "trial_choice": "GPT-6 Astra's privileged success in the quick check of this task version (no formal run on it yet)",
+ "poster": "frame 84 (same scene, camera and settings as the video)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:40-0500"
+}

masters/roboolympics-fun/dodgeball/video.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9bb5f06833231633b5804c87b60983589cb09ea9e4f35e4d92772aff601f8639
+size 9637449

masters/roboolympics-fun/double_dutch/meta.json

@@ -0,0 +1,102 @@
+{
+ "suite": "roboolympics-fun",
+ "task": "double_dutch",
+ "family": "double-dutch",
+ "task_dir": "embodied-olympics-fun-double-dutch-i00-privileged",
+ "task_version": "JamesKrW/embodied-first-exam@4fde6ba91c",
+ "source": "agent-success:codex-gpt6_astra-high-embodied-olympics-fun-double-dutch-i00-privileged/embodied-olympics-fun-double-dut__uSMXSEu",
+ "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": 264,
+ "seconds": 8.8,
+ "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 (ffb43c97d9893f26)",
+ "end_check": {
+  "success": true,
+  "stages": [
+   "entry",
+   "jump 1",
+   "jump 2",
+   "jump 3",
+   "jump 4",
+   "jump 5",
+   "jump 6",
+   "jump 7",
+   "jump 8",
+   "jump 9",
+   "jump 10",
+   "landing"
+  ],
+  "stages_total": 12,
+  "progress": 1.0,
+  "metrics": {
+   "jumps": 10
+  },
+  "time_s": 6.07,
+  "failed": null
+ },
+ "replay_hash_check": {
+  "how": "the trial's trajectory replayed with the task's own `rcb_rfo.py replay`, twice in the new version's verifier image and once in the image it was graded on; all three end-state hashes equal, and the verdict, steps, progress, stages and metrics equal the recorded reward.json",
+  "graded_on": "6109cac419 (files 0.3.2)",
+  "final_state_sha": "ffb43c97d9893f263e1d5a5138d9fd154abe75f07feb95b43a1eebbf8cdb244d",
+  "steps": 607,
+  "pass": true
+ },
+ "replay": {
+  "trajectory": "codex-gpt6_astra-high-embodied-olympics-fun-double-dutch-i00-privileged/embodied-olympics-fun-double-dut__uSMXSEu/artifacts/app/output/trajectory.npz",
+  "n_actions": 607,
+  "control_steps_replayed": 607,
+  "control_hz": 100.0,
+  "every_control_steps": 3,
+  "final_state_sha": "ffb43c97d9893f263e1d5a5138d9fd154abe75f07feb95b43a1eebbf8cdb244d",
+  "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": [
+   5.949,
+   6.393
+  ],
+  "robot_root": "pelvis",
+  "travel_m": 1.448,
+  "focus": {
+   "robot": 1,
+   "robot+object": 203
+  },
+  "objects": {
+   "crank_a_s": 0.37,
+   "crank_a_n": 0.37,
+   "crank_b_s": 0.65,
+   "crank_b_n": 0.65
+  }
+ },
+ "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": 2.7,
+ "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": "last frame (solved state)",
+ "encode": "libx264 preset slow CRF 16, yuv420p, bt709, +faststart",
+ "created": "2026-10-24T13:13:40-0500"
+}