Some sensor-selection scenarios expect failure modes instead of sensors

[lentil32]

Hi, I noticed two sensor-selection scenarios where the prompt asks the agent to list sensors, but characteristic_form describes the expected answer as one or more failure modes.

Because characteristic_form is used as expected behavior during evaluation, this could make a correct sensor-list answer look wrong to the judge.

Source checked:

• HF dataset: https://huggingface.co/datasets/ibm-research/AssetOpsBench
• Raw file: https://huggingface.co/datasets/ibm-research/AssetOpsBench/raw/main/data/scenarios/all_utterance.jsonl
• Related task config: data/task/failure_mapping_senarios.jsonl
• Related guideline: docs/guideline/case_study_industrial_asset_management.md

Affected rows:

id	type	Prompt asks for	characteristic_form currently expects
111	FMSA	sensors of Chiller 6 potentially relevant to Compressor Overheating	one or more failure modes of Chiller 6
609	multiagent	sensors of Chiller 6 at MAIN site potentially relevant to Compressor Overheating	one or more failure modes of Chiller 6

Nearby rows suggest these should use sensor-oriented expected behavior:

• id 112 asks which sensor should be prioritized for compressor overheating and expects sensors from the Chiller 6 sensor list.
• id 610 is the multiagent version of that same sensor-prioritization task.
• The project guideline also lists "List all sensors of Chiller 6 that are potentially relevant to Compressor Overheating." under Sensor-Failure Mode Mapping examples.

Possible fix:

• Update ids 111 and 609 so characteristic_form expects Chiller 6 sensor names, not failure modes.
• A replacement could use the same installed Chiller 6 sensor list already used by ids 112 and 610.

Minor adjacent cleanup: ids 112, 113, 610, and 611 already have sensor-oriented expected forms, but use the phrase one of more sensors; I think that should be one or more sensors.

While checking related task configs, I also noticed possible candidate-list cleanup items in data/task/failure_mapping_senarios.jsonl: ids 6 and 12 ask for failure modes, but some entries in their candidate failure-mode lists look like sensor/measurement terms (speed, power, pressure or vacuum, oil debris, etc.). That may be worth auditing separately.