★★★★★
FIVE STAR
OILFIELD INTELLIGENCE NETWORK
★★★★★·DEMO 01·V.L. PHASE 7·FLUID-END DIFFERENTIAL·CPU-ONLY
DEMO 01 · V.L. APPLICATION LAYER

Fluid-end differential.
The fuel shift should not fool you.

Four frac pumps on a shared fuel skid. A fuel-quality drop hits all four simultaneously — the kind of shared shift that fools every per-pump threshold and every single-asset ML model. Pump-0 has a slow drivetrain degradation starting at t=600. Three methods, one question: which one finds the real failure without crying wolf?

RUN IT LIVE · ON OUR SERVER

No cd commands.
No clone. Just click.

LIVE DEMO · FLUID-END SCENARIO BUILDER

Dial in a fleet shape that looks like yours. The bench re-runs against your scenario on our server. Real outcomes, fresh on every click.

Build your pad

All three methods (per-pump threshold, statistical-ML anomaly detection, V.L. peer comparison) run against your scenario. Watch how V.L. holds up across the dial.

THE SCENARIO

A pad. Four pumps. One real failure hiding under a fuel-quality shift.

Each pump emits one composite RPM-stability metric per timestep. Higher = worse. Fuel quality drops at t=300, recovers at t=500, drops again at t=700 — all four pumps see this identically. Pump-0 starts a slow drivetrain degradation at t=600 that diverges from its three peers. The other three pumps stay healthy throughout. We compare three detection methods on the same data.

REAL RESULT

From the live Validiti binary run.

Method Detection time Lead (pre-deg) False alarms Verdict
Per-pump threshold (μ+3σ) t=300 −300 205 FALSE ALARM — triggered on the fuel-quality confound, not the real degradation
Per-pump statistical-ML anomaly detection t=95 −505 291 FALSE ALARM — the canonical statistical-ML baseline fooled by shared shift
V.L. — Validated Learning t=631 +31 2 EARLY DETECTION — lead time within 50 timesteps of ground truth
CPU-ONLY AMD Ryzen 5 7640HS · 6C/12T · single thread · no GPU · ~5 seconds end-to-end on a single core.

What this means on the pad

Today's per-pump threshold methods would have raised 205 false alarms before the real degradation even started. Operators silence those alarms within a week. When the real failure shows up at t=600, nobody is listening.

Per-pump statistical-ML anomaly detection — the gold-standard statistical-ML approach used by Halliburton OCTIV, ChampionX Windrock, Bently System 1 and the rest of the incumbent stack — is worse. 291 false alarms. It has no way to know the fuel-quality shift is a shared effect shared by every peer pump.

V.L. cross-pump peer comparison detects pump-0's real degradation at t=631 with just 2 false alarms. The fuel-quality shift cancels in the cross-pump residual — by construction. This is ~100× fewer false alarms with correct early detection, on the same data, on a single CPU core, in seconds.

METHODOLOGY

Why single-channel methods structurally cannot do this.

Per-pump methods see one pump at a time. The fuel-quality shift looks anomalous because the pump's RPM stability dropped — that's a real signal. The method has no way to know all the other pumps' RPM stability dropped at the same instant because it doesn't look across pumps.

V.L. stores the peer comparison across pumps as a native operation. The cross-pump median absorbs the shared shift. A divergence from peers shows up as a structural anomaly in the joint — not as a per-pump threshold trip. The substrate makes this kind of detection free.

DATA SHAPE & PATTERN SOURCE

The pad-level signature in this bench is a synthetic stand-in for the multi-pump fleet-behavior pattern documented in field-monitoring vendor literature:

Synthetic data with deterministic seed lets anyone reproduce the exact numbers above. Real-pump telemetry available under license to Wave-1 Five Star customers for benchmarking against their own fleet.

WHO BENEFITS FROM THIS

Five roles, five wins.

Frac crew supervisor

Real alarms again. The fuel-quality wobble that everybody silences stops triggering; the real bore drift comes through clean.

Reliability engineer

Catch a fluid-end drift weeks before any per-pump threshold trips. Plan the swap during a window, not on the pad.

Service-company operator

Fewer surprise pulls on a pad. Lower variance in stage execution. Better customer math at the end of the job.

HSE / regulatory

Court-defensible record of what every pump saw at every crank angle, signed at write time. The chain answers what memory cannot.

CFO / commercial

One avoided fluid-end swap on a pad covers the substrate license for the year. Math is short.

WHAT CHANGES FOR YOU

In numbers.

~100× fewer false alarms

Compared to per-pump anomaly detection on the same data. Real bench output, not a claim.

Earlier detection

Drift surfaces inside ~30 cycles of the real degradation start, by construction. Industry baseline either misses it or triggers on the wrong thing.

Every alarm signed

When V.L. flags a bore, the record is signed at the moment it’s written. Replay it in court, in five years, byte-for-byte.

← ALL DEMOS NEXT · MATHS KERNEL RACE →