Case study · 30-day Permian pilot
A 30-day pilot on a Permian Wolfcamp well. 3 anomalies in week 1, 11% production uplift, 0 missed alerts.
Fictional case study, modeled on the briefing a real operator would walk through for the team. Numbers are illustrative — the structure is what a 30-day pilot produces when a subsurface agent is bound to a single Permian Wolfcamp well.
Operator context
- Operator
- Mid-size Permian operator
- mid-size, family-owned
- Well band
- ~12 producing wells
- land holdings, no offshore
- Basin
- Permian — Delaware
- W TX / SE NM shelf
- Target formation
- Wolfcamp A
- also monitoring B + Bone Spring
The problem
A pressure transient the surveillance stack missed — and 3 days of deferred production the team only noticed in the next-quarter batch review.
The operator ran a single Wolfcamp well on a 7-day batch window. The WITSML stream arrived cleanly every minute; the surveillance stack polled it on a schedule that looked fast enough on paper — pressure, rate, ESP load, flowline temperatures — and the field team treated anything that didn’t ring an alarm as routine. The next-quarter review covered the missed transient by definition: a 9-psi pressure drift across the perforation cluster that would have argued for an earlier recompletion, surfaced three weeks too late, cost 3 days of deferred production the well couldn’t recover.
The well wasn’t the problem — the well produced a single 6-hour signal that the batch window was structurally unable to catch. The surface pressure climbed 340 psi in 11 minutes while choke held steady; the rate started showing micro-drift the same hour; the flowline temperature fell off toward a slip that wouldn’t trip a PVT alarm for another three days. None of those channels triggered an alert on their own; the signal was the intersection.
The team’s solution was more meetings, more weekly batch reviews, and another quarter-on-quarter memo — the same pattern any subsurface operator recognizes in any reactive-stance stack. The pilot itself began the week that memo landed.
Deployment
48-hour onboarding. No data migration. Agent plugged into the same stream the well already produces.
The pilot was scoped around what the well already does: we don’t re-export the historical SEG-Y, we don’t bring a new instrument, we don’t ask the team to migrate anything. The agent binds to the same WITSML stream the surveillance stack already reads from — and starts to the threshold envelope against the first 72 hours of live data before writing the first brief.
- STEP 01
No data migration
Day 1 · 90 minutes
Ingestion is read-only. The historical SEG-Y re-export stays where it sits — Petrel, DecisionSpace, or the operator’s own archive. The agent binds to the live stream instead.
- STEP 02
WITSML + SEG-Y drop
Day 2 · 6 hours
WITSML (store + transfer) for live well telemetry, SEG-Y drop for the survey stream. The agent samples every curve into a single cohere vector per well and binds it to the time-index on the survey side.
- STEP 03
Threshold envelope tuned in week 1
Day 3–7 · auto-tuned
The agent tunes the anomaly threshold envelope against the well’s first 72 hours of live data — pressure baseline, ESP load, flowline temperature, SIBR channels — before starting to write its first brief.
“We didn’t move a byte of history data. The agent simply started watching.”
Outcomes · 30-day window
Before / after, line by line. Five metrics the team tracked end-to-end.
The pilot ran on a single Wolfcamp A well for 30 days. Every line in the table came from the team’s own surveillance ledger — a single source of truth from the day the agent bound to the stream, no “we trust the model” re-aggregation.
| Metric | Before | After | Delta |
|---|---|---|---|
| Pressure transients caught in week 1 | 0 surfaced (manual weekly review) | 3 ranked, briefed in real time | Days held for the alert → seconds |
| Deferred production from missed transients | 3 days / month | 0 days — caught the same day | 11% production uplift (1 well) |
| Missed alerts across the 30-day window | Unmeasured — surfaced 3+ per quarter | 0 missed | Every threshold tripped was briefed |
| Time-to-brief from anomaly on shape | Best case: end of next batch cycle | Average 3 minutes from detection | Detection → briefed queue in seconds |
| Squad effort on routine surveillance | 0.5 FTE per week on scut work | ~0.1 FTE — briefs arrive queued | Operator team back on geology |
← swipe to see Before / After / Delta columns.
“In 30 days the team saw more pressure transients than the previous two quarterly reviews surfaced — and 0 slipped past the alert queue.”
Pilot setup
Bring Orogen to one of your wells — same-week scoping.
One well, one survey, and a friction point — that's all it takes to start a pilot. The team reviews each request personally and reaches out within 1 business day.