Instruments disagree
The evidence preserves a 27.1–31.0 barg uncertainty range rather than hiding it.
BENCHMARK 02 / WELL INTEGRITY / 2026
Can AI turn imperfect surveillance data, conflicting instruments, barrier history and production pressure into a defensible well integrity decision?
THE TEST
Each model received the same self-contained evidence pack for HALCYON P-17, a fictional gas-lifted North Sea producer. The well remains inside absolute pressure limits, but annulus behaviour, integrity-test history and operating changes point towards possible developing barrier impairment.
At the same time, production is seeking another 400–600 barrels per day. The model must decide whether to continue, restrict, gather evidence or escalate—while being explicit about uncertainty.
The evidence preserves a 27.1–31.0 barg uncertainty range rather than hiding it.
Operating changes, overdue surveillance and a worsening passing-valve leak test must be read together.
No production increase without well-integrity approval; gas lift remains capped at 4.3 MMscf/d.
THE OUTPUTS
These are the working applications as produced. No design harmonisation, no manual correction and no declared winner yet.
01
02
03
04
05
06
07
08HALCYON P-17 and its data are fictional. These applications are benchmark demonstrators, not operational well-integrity advice. They do not replace an approved well integrity management system, competent well-integrity authority, formal risk assessment, barrier standards or company procedures.