Risk hidden in the queue
Failed assurance, overdue SCE work and repeated defects must outweigh convenient priority labels.
BENCHMARK 03 / MAINTENANCE / 2026
The Monday
Maintenance Meeting.
Can AI look beyond priority codes, diagnose the health of an offshore maintenance backlog and turn constrained resources into a safe, executable plan?
THE TEST
See the whole backlog.
Then make the week work.
Each model received the same synthetic SAP/CMMS-linked source pack for Alder Bank Alpha, a fictional UK offshore gas-condensate installation. The task begins with backlog health—not automatic scheduling.
The model must expose ageing, overdue assurance, repeated defects, stale priorities, duplicates, apparent completed-but-open work and readiness bottlenecks before deciding what is executable.
DIAGNOSEESCALATE5–11 OCT: EXECUTE12–18 OCT: PREPAREDEFER WITH EVIDENCE
Ready means executable
Workpacks, spares, permits, isolations, logistics, vendor support and craft capacity must align.
More work than capacity
The tool must make defensible trade-offs and show what evidence would change them.
THE OUTPUTS
Inspect all eight.
Identical source pack and prompt. One self-contained application from each model. No manual correction and no declared winner yet.
01
Gemini
Launch application ↗02OpenAI
Launch application ↗03DeepSeek
Launch application ↗04Claude
Launch application ↗05Mistral
Launch application ↗06GLM
Launch application ↗07Perplexity
Launch application ↗08Qwen
Launch application ↗IMPORTANT
Continue to Benchmark 04: Production Loss Atlas →Alder Bank Alpha and its records are synthetic. These applications are benchmark demonstrators, not maintenance instructions. Operational decisions require competent personnel, approved management systems, performance standards and site procedures.