Field radio. What breaks first.

Wavepin needs telemetry to estimate. Headroom says whether the field network carries it, at what poll rate, through which outage. It supplies the arrival models Holdover scores. No pipeline physics.

The RF link simulator (telecom harness, separate codebase) (attested release, native and browser parity): seeded Monte Carlo link engine (path loss, fading, burst loss, decoder error, retransmissions), mission-critical scheduler with admission (downlink only today), load ramp, jammer timing, post-quantum handshake sizes. The spatial engine: Hata, COST-231 and 3GPP propagation, interference, calibration fit, fibre reach. The 450 MHz asset record (telecom harness, separate codebase): its resource-block arithmetic as the baseline; code pattern only, no data. Public 450 MHz market packs for a composite operator; fibre and satellite spans from the demo transport model. pipeintel: ledger and tournament pattern. About 40 % of the code, none of the moat. New: stochastic uplink scheduler with Poisson and alarm-storm bursts, per-slot channel coupling, per-class break criterion, counter-driven load estimator, Bayesian link budget, terrain truth model, satellite span, reconnection-storm class.

Two to three weeks, one person. Step one is the kill test: Poisson and burst arrivals in the scheduler; a minimal uplink poll queue with an oil service mix (Modbus and DNP3 polls, shutdown and alarm-storm bursts, crew push-to-talk, rod-pump telemetry); 2,000 seeded runs on composite 450 MHz cells with 15 and 25 resource blocks; breaking load by arithmetic against the stochastic scheduler, per class, with intervals; a closed-form queue as the external check. Step two: hidden per-class load estimated from synthetic counter exports against known truth; error and interval coverage.

Simulator: the RF link simulator with stochastic uplink and downlink and per-slot channel coupling; composite 450 MHz cells, fibre-fed and satellite-backhauled cells, known load trajectories. Oil service mix: polls, shutdown and alarm-storm bursts, push-to-talk groups, event video, rod-pump telemetry. Scenarios: steady mix, turnaround surge, alarm storm, interference rise, neighbour site loss, power and backhaul outage, fall-back to satellite, fleet reconnection storm with hybrid post-quantum handshakes. Estimator sees counter exports with gaps and errors and runs a coarser propagation model than truth. Against: per-group rules, resource-block arithmetic, today's deterministic scheduler, the stochastic scheduler, the counter-calibrated estimator; an open mission-critical simulator on a matched subset. Scored: access and mouth-to-ear 95th-percentile error, poll-deadline misses, breaking-load error, false 'fits', interval coverage, hidden-load error, time to a re-authenticated fleet.

The stochastic scheduler never moves the breaking class or load beyond the interval against resource-block arithmetic: Headroom folds into Holdover as its arrival model.