Drilling. The kick before the pit.

Same engine family as Wavepin, moved from the pipe to the annulus: transient flow, signature bank, frozen-threshold tournament. Gives the rig twin real physics; the rig twin stays rehearsal.

pipeintel (Wavepin engines): the line solver (a probe ran it unmodified as a vertical U-tube with a signed influx and exact mass balance), sensor noise, delay, bias and freeze, signature-bank fit, posterior with UNEXPLAINED, frozen-threshold tournament, ledger; about 30–35 %. The rig twin: rig modes as rehearsal context only. The playbook engine: shut-in and kill sequence as a checked contract, never executed (harness, not moat). New: positive-displacement pump boundary, a multi-section grid with drillstring, annulus and bit junction, non-Newtonian friction, pore-pressure influx and threshold-loss terms, ballooning with hysteresis, open-flowline return with lag and drain-back, a paddle sensor, pit and trip-tank integrator, per-mode re-baselining, a rig-state classifier, an independent second truth model. Phase 2: gas kick, shared with Tieback.

Four to six weeks, one person. Single-phase, water-based mud, simulation only. A vertical U-tube with positive-displacement pump, open flowline and pit; 2,000 scenarios with connections, pump ramps, ballooning and pit transfers; real-time channel noise from released drilling data if channels and licence check out. Pit-gain and delta-flow thresholds, flow-back fingerprinting and a lumped observer against the estimator. Kick detection volume and response time at a frozen false-flow-check rate; truth from an independent second implementation; ballooning-against-influx confusion stated.

Truth: a wellbore simulator (multi-section U-tube, pump, open flowline with lag and drain-back, paddle sensor, pit and trip tank) with liquid influx and loss of 0.2–20 bbl, ballooning, connections, pump ramps, pit transfers, heave-like noise, sensor drift and freeze; an independent second implementation as truth for a subset. Estimator on a coarser grid with nominal compressibility and friction. Against: pit-gain threshold, delta-flow threshold, trend rule, flow-back fingerprinting, a lumped observer, the signature-bank engine, constrained moving-horizon, a learned model. Thresholds frozen at a stated false-flow-check rate. Out of distribution: rheology, geometry, ballooning held out, heave, noise, hidden gas solubility. Scored: kick detection volume, response time, false flow checks per 100 rig-hours, ballooning misread rate, influx-rate error, interval coverage, minimum detectable volume per section.

No gain over flow-back fingerprinting and a lumped observer at the frozen false-flow-check rate: Annulus is dropped.