The flagship · Predict

RASQI Line Simulator

"How much will this line actually make?"

A discrete-event simulator for serial and assembly lines. It propagates the variability a real line has: breakdowns, setups, scrap, blocking, starvation. So the answer is what the line will deliver, not what the cycle times promise.

This is the product you buy. The other three come with it, because on its own it answers one question, and a line decision has four.

The engine is checked, not claimed

1.65% measured error against M/M/1

At the integration step it ships with. The test rejects anything above 4%, so the number is a build condition rather than a marketing line.

There is also a suite of model invariants: series adds, parallel is the maximum, the operator limits, variability never creates capacity. A number can be right by accident. A law cannot.

What it models

Four layers, and no decoration

The line itself. What gets in its way. The statistical discipline that keeps a run from being an anecdote. And the conversion into money.

The line

the physical topology

  • Up to 49 stations, plus a line feeder you configure as a station
  • Parallel identical machines, dispatched the way a real PLC would: by committed load, not blind rotation
  • Synchronized assembly with multi-BOM feeders. The assembly waits when a component line cannot keep up
  • Finite buffers with real blocking and starvation: the Factory Physics coupling a spreadsheet does not have
  • Conveyors with transit time, accumulating belts, fork and merge
  • Returnable carts: a finite fleet, transfer batch size, reorder point
  • Line-level WIP cap (CONWIP). A part is released only while WIP sits below the cap

What gets in the way

why the line does not deliver what the arithmetic promises

  • Seven distributions fitted to your data, set per product and not only per station. In a real line it is the product that governs variability
  • Breakdowns by MTBF and MTTR, per cycle or per time. Failures accrue with usage, not with idle time
  • Scrap and at-station rework, re-running the cycle where the defect happens
  • Setups and multi-product campaigns, with a sequence-dependent changeover matrix
  • Shared operators who walk between stations. The machine waits for them to arrive, and walking counts as waste, so tightening the layout shows a real gain
  • Shift calendar and break stops: real availability, kept separate from labour cost
  • Optional stochastic reliability: Weibull time-to-failure, micro-stops, imperfect repair, wear curves. Off by default

The discipline

a run is not an anecdote

  • Replications with 95% confidence intervals, common random numbers, discarded warm-up
  • Paired A→B scenarios. Your change is isolated from luck, and it prints its own payback
  • Full-factorial DOE on up to three factors: main effects, significance Pareto, recommended optimum
  • An automatic improvement study on three targets, three levels, eight runs, at the line's configured horizon. A short horizon overstates the main lever by about 25%, in the direction that recommends overspending
  • The report states that the gains do not add up. Fix one, then run it again

The money

a line decision becomes a financial one

  • A loss waterfall in $/year, with availability, performance and quality separated. You fix the expensive loss first
  • Automatic diagnosis: findings ranked by value, each with its cause and a TOC, Lean, 6σ or TPM action
  • Payback, ROI, ROIC, NPV, IRR and EVA against the cost of capital
  • A one-page executive summary for print, with CSV and Excel round-trip
  • Your financial configuration never leaves the browser. Prices, margins and demand are not sent

You rewind the run and watch the same jam again

The recording travels as events, not as frames. The browser rebuilds the line at any instant: 1× to 300×, pause, drag back. None of that touches the server. And before recording, the app tells you how many hours fit your connection.

There is also a flow view over a photo of your plant: your stations laid on the real drawing, with per-product markers, buffer queues and the live bottleneck.

The boundary, before you pay

Five things it does not do, and why

Chasing every feature of a general-purpose platform would make it complex and slow, and it still would not out-power the specialists. So some things are deliberately out.

Re-entrant flows and job-shop routing

A part revisiting an earlier station: semiconductor, PCB, plating, heat-treat. Supporting it would complicate the router for every user who does not need it.

Non-FIFO queue priority

It does not move throughput, OEE or $/year. By Little's Law any work-conserving discipline gives the same. It changes lateness per order, which is scheduling, and that is a different job.

Times below one second

The integrator steps at a quarter of the shortest cycle, so one sub-second field makes the whole line step that finely. Past a point it saturates and loses accuracy silently. A whole batch per cycle is fully supported.

AGV fleets and tooling

We model the material handling that drives flow: conveyor transit and parts moving in batches. Fleet dispatch and secondary constrained resources stay out.

Cumulative gain of elevating two constraints

Removed in August 2026, and we would rather say so than leave it implied. It cost up to 186 runs and only fitted a short horizon, which overstates the main lever by about 25%.

The full list is open

Every limit, with its reason, on the capabilities page. Is your line somewhere in between? Tell us about it and we will say honestly whether it fits as it is.

Where it stops

It gives you the number. Three others make the number survive.

The simulator says so on its own screen. Its financial page is an engineering estimate for screening: no working capital, no financing, no sales taxes, no inflation, no FX, no ramp-up, no residual value. When the number has to survive an investment committee, that work belongs to another product. It is in the same licence.

Before · 6Sigma Studio

It fits the distribution to your data instead of assuming normal, which is exactly the variability input the simulator asks for. And Gage R&R answers first: can the measurement system see the variation you are trying to control?

After · Investment Analyzer

It takes the gain the simulator predicted and builds the full case: two complete states subtracted line by line, with tax, working capital and FX. That is what survives a committee.

After · Project Manager

It stamps the success criterion before the measurement window opens, then checks in the data whether the predicted gain showed up. "Not confirmed" is a legitimate way to close.

See the three in detail →

Prove it on your own line first

One free month with the real software. If it does not find money, do not buy it.