Spaces:
Sleeping
Sleeping
File size: 3,071 Bytes
92c59c2 2fccde8 c3a24db 2fccde8 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 | # P1 Manual Playtest Log
Scope:
- initial low-fidelity manual sanity check from the current default reset seed
- focus: can a human read the observation, choose a plausible move, and see a legible reward change?
Episode A: repair toward feasibility
Start state:
- seed: `0`
- params: `aspect_ratio=3.6`, `elongation=1.4`, `rotational_transform=1.5`, `triangularity_scale=0.55`
- low-fidelity feasibility: `0.050653`
- low-fidelity score: `0.0`
- constraints satisfied: `false`
Step 1:
- action: increase `rotational_transform` by `medium`
- expectation: improve iota without changing triangularity much
- result: feasibility stayed effectively flat at `0.050653`
- reward: `-0.1`
- interpretation: legible but weak; this move alone does not solve the boundary issue
Step 2:
- action: increase `triangularity_scale` by `medium`
- expectation: push the boundary over the triangularity threshold
- result: low-fidelity feasibility moved to `0.0`
- result: low-fidelity score moved to `0.291660`
- constraints satisfied: `true`
- reward: `+3.1533`
- interpretation: good reward behavior; the feasibility crossing was clearly positive and easy to understand
Episode B: move the wrong way
Start state:
- same default reset seed
Step 1:
- action: decrease `triangularity_scale` by `medium`
- expectation: worsen triangularity and move away from feasibility
- result: feasibility worsened to `0.107113`
- reward: `-0.3823`
- interpretation: good negative signal; the environment penalized an obviously bad move without needing a complicated reward term
Current conclusion:
- At the time of this initial playtest, Reward V0 was legible on the low-fidelity repair path around the default reset seed
- a real `submit` trace is now recorded; next manual validation is to extend beyond the initial 5-10 episode path and look for one clear exploit or ambiguity
Episode C: submit-side manual trace
Scope:
- same seed-0 start state as episode A
- actions: `rotational_transform increase medium`, `triangularity_scale increase medium`, `elongation decrease small`, `submit`
Step sequence:
- Step 1: `rotational_transform increase medium`
- low-fidelity feasibility changed by `0.000000` (still infeasible)
- reward: `-0.1000`
- Step 2: `triangularity_scale increase medium`
- crossed feasibility boundary
- low-fidelity feasibility moved from `0.050653` to `0.000000`
- reward: `+3.1533`
- Step 3: `elongation decrease small`
- low-fidelity feasibility moved to `0.000865`
- reward: `+0.2665`
- Step 4: `submit` (high-fidelity)
- final feasibility: `0.000865`
- final high-fidelity score: `0.296059`
- final reward: `+2.0098`
- final diagnostics `evaluation_fidelity`=`high`, `constraints`=`SATISFIED`, `best_high_fidelity_score`=`0.296059`
Artifact:
- [manual submit trace JSON](../baselines/submit_side_trace.json)
Note:
this is a historical submit-side artifact from the earlier Reward V0 / pre-telemetry contract surface. Keep it as supporting evidence for the old submit path, not as the current Reward V1 observation-format example.
|