Remove internal episode wording from journey doc

pi0_so101/MULTITASK_INFERENCE_AND_COMPRESSION_JOURNEY.md

@@ -38,9 +38,9 @@ Representative demo artifacts:
 - Whistle to green bin: `eval_pi05_multitask_whistle_green/eval_demo.gif`
 - Lego/block to yellow-blue target: `eval_pi05_multitask_lego_yellow_bluen/eval_demo.gif`
 
-Representative successful whistle/green-bin rollout from episode 10 direct inference:
+Representative successful whistle/green-bin rollout from live direct inference:
 
-![Direct Pi0.5 whistle to green bin](../traces/trace_pi05_direct_whistle_green_bin_runpod_live_12_1200_from_ep10_start/eval_demo_once.gif)
+![Direct Pi0.5 whistle to green bin](docs_assets/direct_whistle_green_bin_rollout.gif)
 
 ## 2. Direct Remote Inference
 
@@ -59,7 +59,7 @@ uv run scripts/serve_policy.py --port 8000 \
   --policy.dir=/workspace/checkpoint/4999
 ```
 
-On the Mac, `experiments/policy_trace/record.py` records a full closed-loop trace while controlling the robot. For a clean run, we first move the robot to a known start pose:
+On the Mac, `experiments/policy_trace/record.py` records a full closed-loop trace while controlling the robot. For a clean run, we first move the robot to a fixed start pose:
 
 ```bash
 PYTHONPATH=. python experiments/policy_trace/move_to_start_pose.py \
@@ -85,7 +85,7 @@ PYTHONPATH=. python experiments/policy_trace/record.py \
   --max-relative-target 15 \
   --jpeg-quality 80 \
   --frame-quality 92 \
-  --start-pose "trace_pi05_whistle_to_bin_runpod_10_start"
+  --start-pose "saved_start_pose"
 ```
 
 This answered the first practical question:
@@ -163,9 +163,9 @@ This stage did produce useful robot behavior. The important lesson was more spec
 - jerk amplification,
 - and eventually real robot rollouts.
 
-The decoder-only episode 10 replay is a good example: it showed the idea could work, but it also exposed the high-jerk behavior that made action-level evaluation necessary.
+The decoder-only robot replay is a good example: it showed the idea could work, but it also exposed the high-jerk behavior that made action-level evaluation necessary.
 
-![Decoder-only episode 10 replay with high jerk](../experiments/trace_compression/runs/whistle10_ld64_smooth_320x240/predicted_action_trace/action_replays/trace_pi05_robot_replay_01/eval_demo.gif)
+![Decoder-only robot replay with high jerk](../experiments/trace_compression/runs/whistle10_ld64_smooth_320x240/predicted_action_trace/action_replays/trace_pi05_robot_replay_01/eval_demo.gif)
 
 ## 5. Real Encoder-Decoder: Global Latents
 
@@ -281,7 +281,7 @@ Visual inspection matched this:
 - bin edges and small object/gripper structure were softened,
 - and the policy likely lost important spatial cues.
 
-The comparison below shows local camera frames against the server-decoded latent frames from the live latent episode 10 run:
+The comparison below shows local camera frames against the server-decoded latent frames from a live latent rollout:
 
 ![Local frames vs server-decoded latent frames](docs_assets/latent10_local_vs_server_decoded.gif)