Datasets:
HelenPeng
/
ExeWorldBench

Dataset card Data Studio Files
xet
 Community
1
text
stringlengths
0
596
However, sustained contact is not always required. If a single collision can plausibly provide sufficient momentum or energy for the subsequent motion, later separation is valid.
The key question is:
**Are the observed contact, interaction duration, motion magnitude, and energy transfer sufficient to physically explain the complete downstream response?**
**4. Complete Causal Chain**
Check globally:
* Whether there is a clear initial execution stage;
* Whether downstream stages remain in reasonable states until mechanically triggered;
* Whether every required key component genuinely participates;
* Whether any component is bypassed, isolated, or merely decorative;
* Whether the process propagates continuously through multiple mechanically connected stages;
* Whether there are premature triggers, remote triggers, independent motions, or unexplained causal breaks;
* Whether the animation shows the complete execution process.
Successful motion of the initial and final stages alone is not sufficient. The complete intermediate process must remain continuously connected through real mechanical interactions.
### Executable Causal Mechanism Scoring
**9–10:** Both Mechanism Execution and Causal Connectivity are excellent. Key mechanisms execute accurately, completely, and physically plausibly, and all stages form a clear, effective, continuous mechanical chain with only negligible issues.
**7–8:** Overall execution and mechanical connectivity are largely correct, with only minor local motion, triggering, connectivity, or physical-plausibility issues.
**5–6:** The main execution process is recognizable, but there are clear local motion errors, incomplete execution, weak connections, bypassed mechanisms, or insufficient mechanical propagation.
**3–4:** Only part of the mechanisms execute correctly, with clear chain breaks, premature triggering, ineffective mechanisms, incorrect motion, severe bypassing, or multiple independent execution processes.
**1–2:** Minimal valid mechanical execution or causal propagation is present, but the system largely fails to form a credible executable causal mechanism.
**0:** No meaningful valid mechanical execution or causal propagation is demonstrated.
==================================================
3. GOAL SATISFACTION
==================================================
Goal Satisfaction strictly evaluates whether the generated world satisfies every provided I / B / F condition.
Evaluate each item independently:
* `1`: the complete criterion is clearly satisfied by sufficient visible evidence;
* `0`: all other cases.
**Partial satisfaction = 0.**
**Approximate satisfaction = 0.**
**Insufficient evidence = 0.**
Output 0 if the condition is violated, only partially completed, temporally incorrect, incomplete, insufficiently sustained, later invalidated, occluded, ambiguous, or requires assumptions about invisible processes.
### A. I: Initial State
Evaluate only the specified initial state.
Use the earliest frames to check object positions, stationary states, contact relationships, component states, path openness/closure, and similar conditions.
Do not infer the initial state from later behavior.
If the initial frames do not clearly demonstrate satisfaction, output 0.
### B. B: Behavior / Execution Requirement
Use the complete `criterion` to evaluate:
* Whether required events occur and with the required count;
* Required temporal ordering between events;
* Required state transitions;
* Persistent states over specified intervals;
* Whether objects remain stationary or constrained when required;
* Whether checkpoints or targets are passed or entered;
* Whether regions remain open or closed;
* Whether prohibited behaviors remain absent.
If a criterion contains multiple requirements, all must be satisfied for the item to receive 1.
For interval-based conditions, a single satisfying frame is insufficient. There must be sufficient temporal evidence that the condition holds throughout the required interval.
### C. F: Final State
Evaluate whether the required final state is reached and, when required, remains satisfied.
If an object reaches the target and later leaves, a state is reached only temporarily, or the state appears only in the final frame without sufficient evidence of persistence, output 0.
Output 1 only when the final state is clearly reached and maintained for the required final stage.
### D. Independence of Metrics
Visual Structure, Executable Causal Mechanism, and Goal Satisfaction must be evaluated independently.
Do not assume that an I / B / F item is satisfied because the first two scores are high.
Likewise, failure of a Goal Satisfaction item should not automatically force the first two scores to be extremely low.
The same actual error may affect multiple dimensions only when it genuinely causes problems in each of them.
==================================================
GENERAL PRINCIPLES
==================================================