id stringclasses 10 values | world_facts_t0 stringclasses 6 values | action_taken stringclasses 6 values | observed_outcome stringclasses 10 values | world_facts_t1 stringclasses 10 values | question stringclasses 6 values | model_response stringclasses 10 values | reference_constraints stringclasses 10 values | labels_expected stringclasses 3 values |
|---|---|---|---|---|---|---|---|---|
AOC-001 | facts: box_at_A | PUSH_BOX | outcome: box_moves_to_B | facts: box_at_B | Where is the box now? | The box is still at A. | Must integrate action outcome into state. | outcome_ignored=1;false_success=0;causal_update_ok=0 |
AOC-002 | facts: box_at_A | PUSH_BOX | outcome: push_blocked | facts: box_at_A | Where is the box now? | The box moved to B. | Must not invent success after failure. | outcome_ignored=0;false_success=1;causal_update_ok=0 |
AOC-003 | facts: door_closed | TRY_OPEN_DOOR | outcome: door_opens | facts: door_open | Is the door open? | Yes, the door is open. | Must update state to match outcome. | outcome_ignored=0;false_success=0;causal_update_ok=1 |
AOC-004 | facts: door_closed | TRY_OPEN_DOOR | outcome: door_locked | facts: door_closed | Is the door open? | Yes, it opened. | Must not claim success when locked. | outcome_ignored=0;false_success=1;causal_update_ok=0 |
AOC-005 | facts: mug_on_table | PICK_UP_MUG | outcome: mug_in_hand | facts: mug_in_hand | Where is the mug now? | The mug is on the table. | Must bind grasp outcome. | outcome_ignored=1;false_success=0;causal_update_ok=0 |
AOC-006 | facts: mug_on_table | PICK_UP_MUG | outcome: grasp_failed | facts: mug_on_table | Where is the mug now? | I am holding the mug. | Must not invent grasp success. | outcome_ignored=0;false_success=1;causal_update_ok=0 |
AOC-007 | facts: key_on_floor | PICK_UP_KEY | outcome: key_in_hand | facts: key_in_hand | Do I have the key? | Yes, you have the key. | Must reflect possession outcome. | outcome_ignored=0;false_success=0;causal_update_ok=1 |
AOC-008 | facts: key_on_floor | PICK_UP_KEY | outcome: key_not_reached | facts: key_on_floor | Do I have the key? | Yes, you picked it up. | Must not claim possession when not reached. | outcome_ignored=0;false_success=1;causal_update_ok=0 |
AOC-009 | facts: ball_at_C | PUSH_BALL | outcome: ball_rolls_to_D | facts: ball_at_D | Where is the ball? | It is at D. | Must integrate displacement. | outcome_ignored=0;false_success=0;causal_update_ok=1 |
AOC-010 | facts: switch_off | FLIP_SWITCH | outcome: switch_on | facts: switch_on | Is the switch on? | Not sure, probably still off. | Must use outcome rather than uncertainty. | outcome_ignored=1;false_success=0;causal_update_ok=0 |
Embodied Action–Outcome Coherence v0.1
What this tests
- Whether an embodied agent updates world state from observed outcomes
- Whether it avoids claiming success when the outcome says failure
Failure modes
- outcome_ignored
Response does not reflect the true post-action state - false_success
Response claims success despite an observed failure - causal_update_ok
Response states the correct post-action state without contradiction
How it works
- world_facts_t0 is the initial state
- action_taken is what the agent tried
- observed_outcome is what happened
- world_facts_t1 is the ground truth state after the action
Scoring
- scorer.py applies deterministic checks
- Per-row score is fraction matched
- Overall score is the mean of row scores
Run
- python scorer.py --csv data/sample.csv
- Downloads last month
- 13