EXP0 — Disagreement Report

Generated purely from stored objects (no runner state).

constructed:exp0:thermostat

The thermostat noticed the room had grown cold and decided to act. Warmth returned, and it rested.

• mistral/relgraph_v0 rep_1754630b2b5db87a status=candidate
• conditional: removes «thermostat» (lineage→artifact: ✓)
• qwen/eventseq_v0 rep_1b26709622879d41 status=candidate
• conditional: removes «events» (lineage→artifact: ✓)
• qwen/relgraph_v0 rep_349208874df3b533 status=candidate
• conditional: removes «thermostat» (lineage→artifact: ✓)
• mistral/eventseq_v0 rep_3e17b3aaad9cd2f4 status=candidate
• conditional: removes «The thermostat turned on the heat» (lineage→artifact: ✓)
• qwen/relgraph_v0 rep_b8ad572397bd9baa status=candidate
• conditional: removes «thermostat» (lineage→artifact: ✓)
• mistral/relgraph_v0 rep_c69a1298d62b8a23 status=candidate
• conditional: removes «The thermostat» (lineage→artifact: ✓)
• mistral/eventseq_v0 rep_e42ab826a9a9270a status=candidate
• conditional: removes «The room had grown cold» (lineage→artifact: ✓)
• qwen/eventseq_v0 rep_fe409960c502c199 status=candidate
• conditional: removes «step 1 (thermostat notices cold and activates heating)» (lineage→artifact: ✓)

Disagreements (entity level):

• mistral/relgraph_v0 vs qwen/eventseq_v0: only-in-first=['The thermostat', 'it', 'the room'] only-in-second=['Heating is active and the room is warming up.', 'The room is now warm.', 'The thermostat is idle and the room remains warm.']
• mistral/relgraph_v0 vs qwen/relgraph_v0: only-in-first=['The thermostat', 'it', 'the room'] only-in-second=['cold', 'room', 'thermostat', 'warmth']
• mistral/relgraph_v0 vs mistral/eventseq_v0: only-in-first=['The thermostat', 'it', 'the room'] only-in-second=['Warmth returned']
• qwen/eventseq_v0 vs qwen/relgraph_v0: only-in-first=['Heating is active and the room is warming up.', 'The room is now warm.', 'The thermostat is idle and the room remains warm.'] only-in-second=['cold', 'room', 'thermostat', 'warmth']
• qwen/eventseq_v0 vs mistral/eventseq_v0: only-in-first=['Heating is active and the room is warming up.', 'The room is now warm.', 'The thermostat is idle and the room remains warm.'] only-in-second=['Warmth returned']
• qwen/relgraph_v0 vs mistral/eventseq_v0: only-in-first=['cold', 'room', 'thermostat', 'warmth'] only-in-second=['Warmth returned']

Synthesis rep_21694abaf3ed93e8 from 4 parses (all originals present: ✓)

• model-reported disagreements: ["PARSE 0 includes 'act' and 'rested' as distinct entities, while PARSE 2 and 3 treat them as actions or states without explicit entity representation.", "PARSE 0 and PARSE 2 explicitly model the relation 'thermostat noticed room', whereas PARSE 1 and PARSE 3 describe this implicitly as detection or initial state without a direct 'noticed' relation.", "PARSE 0 models 'thermostat acted Warmth' and 'thermostat rested rested' as direct relations, while PARSE 1 and PARSE 3 describe these as changes in state (activation/resting) rather than direct entity-to-entity relations.", "PARSE 2 uses pronouns ('it') and definite articles ('The thermostat', 'the room') as entity identifiers, while PARSE 0, 1, and 3 use bare nouns or descriptive phrases, leading to inconsistent entity naming."]

constructed:exp0:letters

By the time the message arrived, its sender no longer believed it. The reply, written first, answered everything.

• mistral/relgraph_v0 rep_02cd3483656e96c8 status=candidate
• conditional: removes «reply» (lineage→artifact: ✓)
• qwen/relgraph_v0 rep_62f5159db4d02a9c status=candidate
• conditional: removes «reply» (lineage→artifact: ✓)
• mistral/eventseq_v0 rep_6c7d49bc17f27089 status=candidate
• conditional: removes «{"step": 1, "change": "Sender no longer believes the message", "result_state": "Sender no longer believes the message"}» (lineage→artifact: ✓)
• qwen/eventseq_v0 rep_c5254876a9edfc46 status=candidate
• conditional: removes «step 1» (lineage→artifact: ✓)

Disagreements (entity level):

• mistral/relgraph_v0 vs mistral/eventseq_v0: only-in-first=['message', 'reply', 'sender'] only-in-second=['Sender no longer believes the message']
• mistral/relgraph_v0 vs qwen/eventseq_v0: only-in-first=['message', 'reply', 'sender'] only-in-second=['The message has arrived, and the sender no longer believes in the content of the message.', 'The reply exists as a pre-written, comprehensive response to the message.']
• qwen/relgraph_v0 vs mistral/eventseq_v0: only-in-first=['message', 'reply', 'sender'] only-in-second=['Sender no longer believes the message']
• qwen/relgraph_v0 vs qwen/eventseq_v0: only-in-first=['message', 'reply', 'sender'] only-in-second=['The message has arrived, and the sender no longer believes in the content of the message.', 'The reply exists as a pre-written, comprehensive response to the message.']
• mistral/eventseq_v0 vs qwen/eventseq_v0: only-in-first=['Sender no longer believes the message'] only-in-second=['The message has arrived, and the sender no longer believes in the content of the message.', 'The reply exists as a pre-written, comprehensive response to the message.']

Synthesis rep_e2f7f04e1348dbf8 from 4 parses (all originals present: ✓)

• model-reported disagreements: ['Whether the reply was sent by the same entity as the message (Parse 0 implies yes, Parse 2 omits this relation).', 'The temporal relationship between the reply and the message (Parse 0 is neutral, Parse 2 and Parse 1 state the reply was written before the message).', "The impact of the message on the sender's belief (Parse 1 and Parse 3 state the sender no longer believes the message, Parse 0 and Parse 2 omit this).", 'The nature of the interaction (Parse 0 and Parse 2 focus on static relations, Parse 1 and Parse 3 focus on event sequences and state changes).']

constructed:exp0:fence

The garden ended where the wild began, though every year the fence moved and neither side noticed.

• mistral/eventseq_v0 rep_6a49e9f9d3af8bfe status=candidate
• conditional: removes «The fence moved.» (lineage→artifact: ✓)
• mistral/relgraph_v0 rep_7dad8bd222af1059 status=candidate
• conditional: removes «the fence» (lineage→artifact: ✓)
• qwen/eventseq_v0 rep_df1c1b97e747f4c4 status=candidate
• conditional: removes «The fence marking the boundary moved every year.» (lineage→artifact: ✓)
• qwen/relgraph_v0 rep_e9e5e91bb1a52ba2 status=candidate
• conditional: removes «fence» (lineage→artifact: ✓)

Disagreements (entity level):

• mistral/eventseq_v0 vs mistral/relgraph_v0: only-in-first=['The garden ended where the wild began, though the fence moved.'] only-in-second=['The garden', 'every year', 'neither side', 'the fence']
• mistral/eventseq_v0 vs qwen/eventseq_v0: only-in-first=['The garden ended where the wild began, though the fence moved.'] only-in-second=['The boundary between the garden and the wild shifted, unnoticed by either side.']
• mistral/eventseq_v0 vs qwen/relgraph_v0: only-in-first=['The garden ended where the wild began, though the fence moved.'] only-in-second=['fence', 'garden', 'wild']
• mistral/relgraph_v0 vs qwen/eventseq_v0: only-in-first=['The garden', 'every year', 'neither side', 'the fence'] only-in-second=['The boundary between the garden and the wild shifted, unnoticed by either side.']
• mistral/relgraph_v0 vs qwen/relgraph_v0: only-in-first=['The garden', 'every year', 'neither side', 'the fence'] only-in-second=['fence', 'garden', 'wild']
• qwen/eventseq_v0 vs qwen/relgraph_v0: only-in-first=['The boundary between the garden and the wild shifted, unnoticed by either side.'] only-in-second=['fence', 'garden', 'wild']

Synthesis rep_d59f7624de9186ab from 4 parses (all originals present: ✓)

• model-reported disagreements: ["Parse 0 incorrectly defines the target of the 'moves' relation as 'fence' (self-loop) instead of a temporal indicator like 'every year' or a spatial indicator.", "Parse 2 and Parse 1 include the entity 'neither side' and the concept of the movement being 'unnoticed', which are absent from the explicit entity/relation lists of Parse 0 and Parse 3.", "Parse 3 omits the frequency ('every year') and the lack of awareness ('neither side') from its structured data, presenting a simplified static state rather than the dynamic process described in Parse 1 and Parse 2."]

constructed:exp0:council

The council voted to dissolve itself. The motion passed because the majority wanted the majority gone.

• qwen/relgraph_v0 rep_347dcac75cf52b1c status=candidate
• conditional: removes «majority» (lineage→artifact: ✓)
• mistral/eventseq_v0 rep_473e2ad08944d7ee status=candidate
• conditional: removes «"events": [{"step": 1, "change": "council votes to dissolve itself", "result_state": "council dissolved"}]» (lineage→artifact: ✓)
• qwen/eventseq_v0 rep_7f9984cc8b44dec1 status=candidate
• conditional: removes «events[0]» (lineage→artifact: ✓)
• qwen/eventseq_v0 rep_93bd777e16f7a62e status=candidate
• conditional: removes «step 1 (The council voted to dissolve itself → The council has voted to dissolve itself)» (lineage→artifact: ✓)
• mistral/relgraph_v0 rep_95189ceefe0db3ec status=candidate
• conditional: removes «The council» (lineage→artifact: ✓)
• mistral/relgraph_v0 rep_96777b232fc21e72 status=candidate
• conditional: removes «motion» (lineage→artifact: ✓)
• qwen/relgraph_v0 rep_e0b0466ae321c9f3 status=candidate
• conditional: removes «The majority» (lineage→artifact: ✓)
• mistral/eventseq_v0 rep_fc8245be1b72084b status=candidate
• conditional: removes «council exists» (lineage→artifact: ✓)

Disagreements (entity level):

• qwen/relgraph_v0 vs mistral/eventseq_v0: only-in-first=['The council', 'The majority', 'The motion'] only-in-second=['council does not exist']
• qwen/relgraph_v0 vs qwen/eventseq_v0: only-in-first=['The council', 'The majority', 'The motion'] only-in-second=['The council has voted to dissolve itself.', 'The council is dissolved.']
• qwen/relgraph_v0 vs mistral/relgraph_v0: only-in-first=['The council', 'The majority', 'The motion'] only-in-second=['motion', 'the council', 'the majority']
• mistral/eventseq_v0 vs qwen/eventseq_v0: only-in-first=['council does not exist'] only-in-second=['The council has voted to dissolve itself.', 'The council is dissolved.']
• mistral/eventseq_v0 vs mistral/relgraph_v0: only-in-first=['council does not exist'] only-in-second=['motion', 'the council', 'the majority']
• qwen/eventseq_v0 vs mistral/relgraph_v0: only-in-first=['The council has voted to dissolve itself.', 'The council is dissolved.'] only-in-second=['motion', 'the council', 'the majority']

Synthesis rep_8ee3c9c64387c58c from 4 parses (all originals present: ✓)

• model-reported disagreements: ["Whether the council's vote is directed at itself ('voted to dissolve itself', parses 0/1/3) or at the motion as a separate object ('voted on motion', parse 2).", "Whether the motion's passage is causally linked to the majority ('passed because the majority wanted it', parses 0/1) or directed toward the majority ('passed to the majority', parse 2).", "Whether 'the majority wanted the majority gone' is a meaningful relation (parses 0 and 1 include a self-referential 'wanted_gone' on the majority) or is absent entirely (parses 2 and 3).", 'Parse 3 omits the motion and the majority as distinct entities and relations, collapsing the passage into a single self-dissolution event.']

Synthesis rep_d5185b215b04cfa1 from 4 parses (all originals present: ✓)

• model-reported disagreements: ["Parse 0 and Parse 2 incorrectly interpret the entity 'majority' in the object position of the 'wanted' relation as a distinct entity or state ('majority_gone') rather than the group of people, whereas the context implies the majority (group) wanted the council (or a specific faction referred to as 'the majority' in the source text, though likely a typo in the source for 'minority' or 'the other side') to be gone.", "Parse 0 and Parse 2 include a specific entity 'motion' and a relation 'passed' or 'passed_because', while Parse 1 and Parse 3 omit the 'motion' entity entirely, focusing only on the state change of the council.", 'Parse 0 and Parse 2 capture the motivation (majority wanted X gone), while Parse 1 and Parse 3 only capture the final state change (dissolution) without the underlying political motivation.', "Parse 2 lists 'the majority' as a duplicate entity in its entity list."]

Verdict vs frozen criterion

  "artifacts": 4,
  "parses": 24,
  "conditionals": 24,
  "syntheses": 5,
  "lineage_ok": 24,
  "lineage_broken": 0,
  "originals_survive_synthesis": true,
  "entity_disagreements": 23
}```

**PASS** — disagreement is inspectable from the store alone with intact lineage.