Introducing MechA, the Opus Magnum Metric of Your Nightmares

The 7th annual Opus Magnum tournament is going strong. As always, the tournament format has a single puzzle each week, designed in secret by the host and a team of playtesters. The puzzle goes public on the website each Friday, and is due just over a week later. At the deadline, players see how their designs stack up against the increasingly impressive community.

Haxton, this year’s host, delivered us a brand new type of challenge. In the just-now-concluded week 5, he unveiled Crystallized Air. It’s a reasonably simple puzzle, with a 3-atom input and a large 21-atom output:

For the first metric, we optimized area, a rather typical choice for large products with internal empty space. If players tie in area, we sort them by cycles, then cost. The abbreviation for this combo is ACG. Nothing too crazy here.

The crazy comes out with the second metric. You see, for the second metric there is a program on the website that takes a solution and removes all of the parts except arms and track. You have to input a machine that solves the puzzle, but with all of the pieces removed except arms it will only flail aimlessly. The second metric is ACG, for the modified solution.

Depending on how intuitively you understand the deep corners of Opus Magnum solution space, the previous statement may inspire anything between confusion and abject horror.

Now that it’s all in the open, I can share the horror and the story of what I built.

Mechanism-only Metrics

The parts that make up an Opus Magnum solution are the products, reagents, glyphs, and mechanisms. The products, reagents, and the molecules made in between, have no cost but do contribute area. Glyphs and mechanisms contribute both cost and area.

Back in 2020, I hosted the tournament and my week 7 puzzle High Gloss Finish had a “Mechanism Cost” metric. I designed it to require players to use only the simplest 20g arm, with as many glyphs as they wanted. I believed that Mechanism Cost was just “cost but easier”. This appears to have held true, as PentaPig made a proper winning solution within a week, while we didn’t see a true minimum cost solution until rebix a year and a half later.

2025 is a very different year. It may sound like Mechanism Area is area but easier. That is far from the truth. Mechanism Area is cost but harder.

With atoms and glyphs removed from the area calculation, you’re free to make enormous molecules and place glyphs in any place that they serve a purpose. In exchange, you are restricted to arm movements within the smallest possible footprint. What is the smallest possible footprint? To answer that question, we need to discuss access points.

Access Points

Access points refer to the different tiles on the board that an arm is able to grab. Borrowing a graphic from my previous cost optimization blog post, here are 3 different arms. I’ve highlighted their various access points with equilibrium glyphs.

Pistons, with their extension ability, have significantly better access, at 18 tiles without track. This typically makes pistons the MVP of area.

Relevant to MechA, a piston can always retract before swinging, always staying within this 19 tile footprint. That’s far better than the longer arms in the previous image, that swing through a lot of area whenever they move.

Of course, just because an arm has all these access points, doesn’t mean it needs to use them all. The ideal solution would have 1 MechA from the base, and 1 MechA per access point, and never move in a way that uses MechA outside of those access points. With that understanding, MechA asks the player to use one arm and as few access points as physically possible.

Zero Access Everything

Borrowing a second graphic from the cost optimization blog post:

Zero Access Everything is an impressively fear-inducing realization to an Opus Magnum player. By “everything”, I only mean glyphs here – the inputs and outputs require one access point apiece.

To Zorflax’s credit, he made a seriously difficult cost puzzle last year to get this reaction. He forced players to spend all access points on inputs and outputs. We could only use glyphs indirectly, by waving molecules around onto them.

The reason this worked at all was because the inputs and outputs all were multi-atom molecules. With structure available from the get-go, you can always boostrap up to larger and more powerful structures, and then break off an output such that one of its atoms is on an access point. Zero Access Everything is an insane engineering challenge. When I triumphed, I felt inspired to write an entire blog post.

Crystallized air has repeated this feat, in every way.

The input and output are multi-atom molecules. You can bootstrap, build structures, and then break off an output. And I’m writing right now. I built a zero-access everything solution for Crystallized Air.

The Implications

In a true cost puzzle, you can move the arm into 6 different positions. Even in Zorflax’s Latch-Hook Fireworks, you could manipulate your structures at any angle.

In this MechA puzzle, you get one input and one output. Each takes up one access point. To make a perfect solution, you can move the arm into only 2 positions. This would achieve a MechA score of 3. So certainly we have the same zero-access everything restriction that made last year’s cost monster so intimidating. But we also have an impressively severe movement restriction. Cost but harder.

The MechA 3 dream requires adding glyphs and programming to a layout like this, while keeping the arm in the triangle at all times.

Though I will confess, I didn’t start here. I suffered through less perfect designs, building an understanding with the intention of not losing my whole week in pursuit of something above my skill level.

First Design: 6 MechA

In order to not fall flat with nothing, I made a first solution to this puzzle using some cost principles. I splurged on 3 extra access points – two on a bonder and one on a debonder. The debonder let me take single atoms off of the input (while building the rest into a waste chain), and the bonder let me assemble those atoms into the output. I deliberately never moved the arm to the due-left position. A quick sketch of the MechA used here:

A reasonable person would note that this is slow and has an enormous area. Keep that in mind.

Second Design: 4 MechA, Incomplete

Uh. Allow me to explain.

The intention was to use the 4 access points shown, making a diamond shape.

I sacrificed the debonder access point, and one of the two bonder access points. My understanding was that I could build enough structure with this, and add glyphs as needed whenever that structure went somewhere. Eventually I would break off the output and deliver it. I spent 10 hours kind of just throwing my face at this design, and what happened was the horrifying abomination in the previous screenshot.

On cycle 167 it successfully made this:

This object comes from 9 inputs, and has a piece that resembles nearly half of the output. At the pictured cycle, it has separated from the held object. It also sits on a bonder, making it possible to reattach in a different way.

The principle of breaking things out of access and saving them with a bonder has become known as “lustrous syrup” tech. To quote the cost post:

“In order to have access to both parts of a molecule after debonding, you need a grab point on each.” This is not strictly true. The exception is when the molecule that is out of reach, sits atop a bonder. As long as you can reach the other side of the bonder, you can regain access to the otherwise-stranded molecule.

With moves like these, I was able to keep connecting things back to rescue stranded parts, building more and more of the product:

But I had so many out-and-back-again movements that I ultimately got lost. Every new glyph I added broke something necessary to get that far. It was a horrifying soup of glyphs and I couldn’t make it work. I went to bed on Tuesday genuinely upset at having lost 10 hours with nothing to show for it.

Third Design: 4 MechA, Complete

I understood that to make it work, I needed to be far more deliberate in my glyph placements. No soup of nightmares. I had also learned quite a few new nice features of the 4 MechA solution space. One example, which I was now prepared to take full advantage of, was a “behind-the-back” C shape.

This was what had me convinced that MechA 4 was possible. I could use the input to attach to an object which was sitting on the opposite access point, and move it around the arm. Doing this repeatedly built out a chain that extended with a unique offset vector, allowing me to put structures on new areas of the surrounding space.

In layman’s terms: If I just keep doing this, eventually my problems go away. They are replaced by new problems. Problems with really high cycle counts and area counts. But hey. I did it. Wednesday brought triumph.

Isn’t she horrible?

A reasonable person would note that this is both slower and higher area than the last one. A reasonable person probably also feels nauseous. Watch it to the end if you dare. The pieces move really far before they break off. Uncomfortably far.

I made a much better plan for separating the final product from the block, but it was still clunky. My vector math was also really annoying to implement, even though it led to a working solution. I titled it “Finally, half access arms” (as the arm had 3 of its 6 access points). My solution notes parodied the Sandia labs nuclear waste warning. “This is not a machine of honor. No highly esteemed achievement is commemorated here. Nothing valued is here. What is here was dangerous and repulsive to its maker. This metric is best shunned and left uninhabited.”

Broken Promises

I really did promise to put the puzzle away and not think about it until the deadline. The 4 I had built seemed to me like a limit, both of my own sanity and of the technical feasibility of MechA. With only the triangle of access points you lose that behind the back move. If an object is on the left access point, you cannot swap its position with an object on the right access point.

I didn’t really even need the bonder access point. The only thing I thought was preventing 3 MechA, was losing that movement.

I spent Thursday doing everything else, having a grand old time being a human being. And Friday was Valentine’s Day, I was certainly going to be busy all evening. But I mean, I could probably tinker with 3 MechA in the morning before work..

At 9:35 AM, I sent Haxton this message:

23 minutes later:

I think 3 was going to be too hard to bite off all at once. I’m glad I built a horrifying solution with 4 MechA before diving in at 3. But I genuinely surprised myself when I spent a little time on 3 and a working solution nearly fell into my lap. I learned so much from my 4s that I could reasonably quickly finish it out from here. I still did my work day, and I still did my Valentine’s Day. After hours and hours lost to the 2nd design, it felt almost surreal to have surpassed expectations in every way.

How It Works

To be clear, it still has a ton of glyphs and a ton of complexity. But it is far more efficient than the previous iteration, despite the increased restriction.

It first builds a large structure whose outer perimeter includes 12 of the 21 atoms of the output exactly. With only 3 MechA, it can only do this by appending new inputs onto input-shaped corners of the existing working region, so everything it builds is at least 3 atoms thick at all time.

It then uses input suppression to spin the whole thing around, breaking the useful shape off in a particular location in the lower left area of the board.

It then makes another 3 atom thick wormy thing, whose edge matches the remaining atoms needed:

And joins them:

Now, because of the glyphs already placed, it is quite annoying to separate this output from the brick in a position close to the arm. So I break it off sooner than that:

Then reattach it with syrup tech at only a single atom. This gives far fewer chances for errant bonds from necessary glyphs.

All that’s left is to reel it in, drop the product, and line up the huge waste block somewhere that lets me do another layer of syrup later down the line. Which works! Here’s the gif, which starts with the previous product’s waste already on the board:

Check those stats. This thing can run to completion in-game (hence the gif). It’s faster than the older ones, and the area isn’t that bad. It’s sub 100k. And it’s a 3 MechA solution! A triumph!

Community Impressions

Before the deadine, I kept looking at the chat every day. I saw that this puzzle had left a wide range of impressions. Some people were cursing Haxton. Some people were cursing Shadowcluster, the playtester who we all knew had a hand in this decision. A few people maybe didn’t make all of the logical connections from “optimize MechA” to “We must make a zero access everything solution with extremely limited movement” and felt like all of the doom talk was confusing.

SpiritualShampoo had probably the loudest stance. He repeatedly insisted that this was a metric that encouraged people to build solutions that would literally not run without mods (due to area calculations during movement of huge waste molecules, though he didn’t say this part out loud to avoid spoiling any details). This was bad for competition.

I have always thought of the tournaments as a content generator. Not so much a competition. The whole goal is to make memorable community moments, and stories to tell. I didn’t know how well the story of this puzzle would shake out, but as I worked through it and found a 3 of my own, I started to feel like this would go down as a high note of the 2025 tournament.

I could be wrong. It certainly stirred a lot of discussions from other players whose own extreme solutions were causing them to feel bad about the game and the format of the tournament. Haxton had regrets. People discussed mods as a necessity, and other tools like a bot made by Grimmy which would optimize solutions for you by removing unnecessary glyphs. In this particular puzzle, where knowing a glyph isn’t actually necessary could allow you to design a much more efficient late stage, that was a huge benefit.

How Everyone Placed

Watching the stream, the MechA pileup began at 7. These were the people who treated the puzzle like a cost puzzle, and used the full range of access of a length 1 arm. This lasted until Michira in 24th place, who had the fastest 7. We then spent a while looking at 6 MechA solves, which lasted until Mr Puzzel in 14th.

We then got to the 5s, which is where things required a lot of effort. Many people with 5s were convinced that lower was not actually practically possible. To copy Kazyan’s lovely notes from their 5 MechA solve:

You’re smoking that Mist of Hallucination if you think I’m doing this without access to both sides of a Glyph of Bonding, so mecha-five is what you get. I don’t care that you can use 500 of them; at some point you need to pivot atoms around and actually build the molecule instead of handwaving about things that scaffolding could theoretically do. This is like operating a ouija board with your feet, on stilts. This is playing with a ball-in-a-cup with your hands tied behind your back and drunk. This is like doing calligraphy in a tornado, with a beehive on the side of your face, and if you get stung you have to start all over.

SergyD spent over a dozen hours trying to make a 4, but was unable to complete it and shipped a 5. His solution notes were the lyrics to In The End by Linkin Park (tried so hard, and got so far.. in the end, it doesn’t even matter…) To SergyD’s credit, he did win the other metric.

There are lots more insights from the various participants in the solution notes, I advise watching the full stream. I love how the solution notes give a snapshot of the impressions of individual minds, before sharing all their notes.

Top 5

After albatross/who/you/all-of-us (the one with many names) in 6th place with the fastest 5, it was time to dig into the crazy people who did 4.

Just kidding. It jumped straight from 5 to 3. Nobody had a scoring 4. This blew my mind. There were 5 of us who rose to the challenge and succeeded. 5 crazy souls who made the zero access everything movement restricted insanity.

In 5th place was rebix. Known for achieving the nearly-impossible in the cost and cost-like universe, it came as no surprise to see rebix at 3 MechA. He used a piston! As did SpiritualShampoo in 4th. Both of these took a very similar dot-matrix-printer approach with enormous waste sticks. SpiritualShampoo had a far better cycle count, around 6000 compared to rebix’s 13000.

In 3rd, we saw CooCoo52. Known for braving esoteric metrics, he took an approach very similar to mine. However, it had a longer brick and led to 5000 cycles.

In 2nd, we saw Kaliuresis. He made a 3 that resembled the build process of my original failed 4, which had an internal piece that looked like the output. His solution had far less area, but still over 3000 cycles.

That meant I was the big winner! I’m being rewarded for obsessive building behavior.

Full Results Video

Thanks as always to Haxton and the entire crew running this tournament. It’s a lot of work to run a tournament, and you put a lot on the line delivering a challenge knowing it is going to become the obsession and timesink of a few dozen dedicated players. I look forward to the rest! To all readers, please feel free to join the tournament, there are no stakes besides seeing your solution alongside the ones built by others! I promise not all puzzles are like this one.