The samples are copied without alteration from the embedded JSON in the two supplied HTML demonstrations. Displayed threshold crossings are calculated from those samples.
The original event logs and narrative did not consistently match the embedded time-series thresholds and values. Revised pages distinguish sample crossings from authored event timing.
Scenario 2 stated an event on September 22 and overdue maintenance due September 24. The overdue assertion has been removed.
The referenced manuals, SOP, maintenance records, and standalone alarm CSVs were not provided. No source-document retrieval is claimed.
Absolute fault exclusions have been replaced with evidence-limited hypotheses.
The original multi-unit overlay charts have been replaced with separate labeled charts and downloadable data.
The broken scenario link has been replaced with working navigation.
What this demonstrates
A proposed way to present industrial evidence, compare two patterns, and explain uncertainty. It does not validate an AI model, demonstrate local inference or RAG, or establish actual equipment diagnoses.
Next validation step
Supply a consistent synthetic document set and event logs, define reference answers and failure cases, then run a real retrieval-and-evaluation pipeline. Report its measured results separately from these scripted pages.