Smelter Throughput Guide: The Six Ratios That Actually Matter

Before the rework, a belt fed from two sides distributed input evenly regardless of upstream pressure. It no longer does. Since 3.4, a merge favours the side with the fuller lane, which means any design that relied on symmetrical merging now drifts under load. If your smelting column is running at ninety-four percent and you cannot find the missing six, this is almost certainly why.

Everything below was measured in-game using the throughput readout on a thirty-minute steady-state sample, repeated on three separate saves at different tech levels. We are publishing the method as well as the numbers so you can check them on your own build rather than trusting us.

The six ratios worth memorising

These are the only ratios that appear often enough in a normal playthrough to be worth learning. Everything else can be derived from them.

The most consequential of these is the alloy furnace to plate press figure. Running the old one-to-two ratio leaves your presses idle for roughly eleven seconds in every minute, which compounds badly across a large factory because the idle window drifts out of phase with the furnace cycle.

How to measure this yourself

You do not need a spreadsheet. The in-game readout is accurate if you use it correctly, and most people do not, because they sample too early.

1. Build the section in isolation with a dedicated input line. Shared inputs make every measurement useless.
2. Let it run untouched for at least eight minutes. Belt contents need to reach steady state, and the first four minutes are always optimistic.
3. Read the output counter over a full thirty minutes rather than a spot value. The cycle length on alloy furnaces is long enough that short samples swing by twelve percent.
4. Repeat once with the input line deliberately over-supplied. If the output changes, your bottleneck is upstream and the ratio you just measured is meaningless.

Step four is the one everybody skips. Half of the incorrect ratios circulating in the community come from measuring a starved line and treating the result as a design constraint.

Merge behaviour after the rework

The practical rule is simple: never rely on a passive merge to balance two lanes. If two inputs must be split evenly, use an explicit balancer, even though it costs three extra tiles. We tested a symmetrical merge under fifteen minutes of uneven load and measured a persistent fifty-eight to forty-two split that never self-corrected.

This has one useful consequence. You can now deliberately prioritise a lane by keeping it fuller, which makes it possible to build a soft priority system without any logic circuits at all. Feed your critical line from a buffered source and your overflow line from a direct one, and the buffer will win every merge.

What we got wrong last time

Our original version of this guide, published on patch 3.3.2, listed the coal feed ratio as one to eight. That figure came from a measurement taken on a line that was itself coal-starved, which is precisely the mistake described in step four above. The corrected figure is one to six. We are leaving this note here permanently rather than editing the history out, because a guide that has never been wrong is usually a guide that has never been checked.

Quick reference