Optimized Snarkmaker

Post by **dvgrn** » April 13th, 2026, 8:28 am

On a different thread I made some unsolicited wild guesses about how close this new Snarkmaker recipe is to optimal. Here is some uneducated hand-waving about what might reduce the cost a bit more.

russellsprouts wrote: April 12th, 2026, 10:17 pm - Separate the elbow from the target (there's a bunch of cleanup that happens for things around the target, and I separate them by about 30 hd, but the ash quickly expands back into it. I decided I didn't care too much to keep them strictly separated)

1) On a fresh Snarkmaker compilation run, it would probably make sense to skip that "separate the elbow from the target" step.

2) I'd also guess that it's not necessary to create the offset block far off to the side, nearly as early. Could probably bubble up some junk into that area randomly at any time, even during the cleanup phase. There are all kinds of cool options for doing cleanup and leaving an elbow somewhere out there. In an emergency, random junk in that 90-degree target area can be adjusted with a 0-degree glider at a high offset -- but there are lots of other lucky possibilities that might come up naturally in a search, like this thing that actually happened:

russellsprouts wrote: April 9th, 2026, 4:44 pm After sending two gliders through the snark, there happened to be a 90 degree glider crossing the lane, so I was able to recover the elbow by intercepting it.

3) I'm very curious as to whether it's possible to expand this search technique by having the active "zero-degree elbow" area drop simple targets like blocks, blinkers and beehives directly into the near edge of the construction zone, instead of making painful slow-salvo modifications of existing targets.

E.g., is it too unlikely to hope to run a search until the active elbow area drops the second pond directly into position at the near edge of the two-ponds-and-blinker constellation? If that's too unlikely, can we drop a few common objects nearby to make a seed for a pond?

Long story short, there are ways to hugely expand the handy chart that russellsprouts used to navigate this construction. If a search can be arranged to take more advantage of more possible sneaky tricks, the Snarkmaker should keep getting cheaper.

Post by **dvgrn** » April 13th, 2026, 9:46 am

Hippo.69 wrote: April 13th, 2026, 2:30 am There are seeds of destructions there as well, what could complicate the maters.

Is it possible that we might be able to add seeds of destruction to a Snark, in a way that allows a whole chain of Snarks to be cleaned up with the same destruct signal? I'm not sure if that would really be convenient in practice, or if there's usually just a few Snarks at a time connecting other pieces of circuitry.

One fun way to make a chainable destruct-ready Snark would be to have the destruct signal be two gliders on the input lane, too close together. The first glider would escape normally, and the explosion from the second glider plus the "decorations" would produce a new second glider with the same relative timing. It sounds unlikely, but there are absolutely ways to do that -- just not sure how many decorations it would take.

Anyway, probably it would be better to have the destruct trigger glider be a glider returning exactly along the output lane; the added seeds of destruction would cause a single glider to escape down the input lane. It looks plausible that that could be arranged without placing any bait objects behind the Snark heart... which would mean it would be possible to borrow all of the difficult part of russellsprouts' Snarkmaker, and just run some replacement grey-goo searches to place baits and do a different cleanup.

NOTE: If anyone has a better short name than "grey goo" for "single-channel searches using an active elbow object", please suggest it -- or better yet, just start using it plausibly enough that other people start following along. I'm just using "grey goo" because it's a short distinctive existing term, not because I particularly like it ... then again, if we're getting to the point where people understand what I mean when I say "grey-goo searches", then we're done already!

----------------------------

Technically simsim314's conception of "grey goo" was a little different from russellsprouts' active elbow objects, I think: the idea was not to have an active elbow at all, but to just direct the active reaction so that it incrementally built constellations directly -- no 0-degree gliders necessary. But it does seem surprisingly efficient to use gliders the way the new Snarkmaker does it, especially with the occasional shortcuts that russellsprouts has mentioned.

So maybe there's room for a new term here, for the "active glider-producing elbow" subset of grey goo construction. (?)

Hippo.69 · Post by **Hippo.69** » April 13th, 2026, 10:24 am

dvgrn wrote: April 13th, 2026, 9:46 am Technically simsim314's conception of "grey goo" was a little different from russellsprouts' active elbow objects, I think: the idea was not to have an active elbow at all, but to just direct the active reaction so that it incrementally built constellations directly -- no 0-degree gliders necessary. But it does seem surprisingly efficient to use gliders the way the new Snarkmaker does it, especially with the occasional shortcuts that russellsprouts has mentioned.

So maybe there's room for a new term here, for the "active glider-producing elbow" subset of grey goo construction. (?)

Yes, simsim314 had p30 phase limitation which significantly reduced the search space (what made decomposition to individual emissions very expensive).

I would like something like "single lane in envelope search" (what is not simsim314 case).

Established versions of single lane recepies could be characterized as "speboe anchored recipes", both these "grey go" methods decided to accept the wide range of inbetween configurations, but simsim314 used them just to emit gliders on prescribed lane-phases, while russellsprout tries the inbetween configurations to evolve to the desired pattern and sometimes uses gliders inside the envelope to do the job.

So there are incentives promoting desired configurations known from slow salvos doing the job ... the whole cutting/motivating alchemy...

dvgrn wrote: April 13th, 2026, 9:46 am Anyway, probably it would be better to have the destruct trigger glider be a glider returning exactly along the output lane; the added seeds of destruction would cause a single glider to escape down the input lane. It looks plausible that that could be arranged without placing any bait objects behind the Snark heart... which would mean it would be possible to borrow all of the difficult part of russellsprouts' Snarkmaker, and just run some replacement grey-goo searches to place baits and do a different cleanup.

Unfortunately both suggestions are not aplicable to 0e0p metacell easily, the gen travels several times through the loop so (the trigger should be inserting cancer nearby glider to the gen), the opposite direction is even more complicated, it would require stop the glider stream and after a very long time (wait entire gen loop?) to send the glider in the opposite direction.

The cancer glider solution would require reaction sending glider in nearby lane directed as the normal loop, make time adjustments by 0ott's with last of them insserting it back to the stream. Sending perpendicullar glider out of the snarkmaker envelope and turn it would generate such huge delay, that making all the speedup 0deg ott's would have extreme cost. Of course if there is a phase distance where the glider goes through and the snark explodes behind, it would be optimal ... let me check this possibility ... no there is no such phase gap.

So at least there would be nontrivial change to the snark(maker) to prepare such SoD's.

I will still think about backward glider as even bigger then minimal seed of destruction could be worth doing by in envelope recipe if standardized then several cheap seeds of destruction which would require much longer single lane speboe recipes.
But even this would require 2 variants ... for first snark and others ... actually not for shell of 0e0p where the spiral is entered by scorbie reflector ... so there could be absorber or start of the SoD of the scorbie ... on the transparent lane.

speedydelete · Post by **speedydelete** » April 13th, 2026, 4:55 pm

For the past 1.5 months I've been working on an advanced search program for generating an optimized toolkit of single-channel recipes (all of the discussion about this has been on Discord). It supports different spaceships, different minimum timing gaps, and non-B01c INT rules. The recipes found by simeks are quite unoptimal due to having mandatory slow gliders every other glider, and only using one kind of elbow. I'm implementing a multi-elbow searcher along the lines of dvgrn's ideas, plus some optimizations of my own, especially an "elbow resolver" which is something like lookahead that figures out all arrangements of slow gliders that can follow an incomplete fast recipe to make a slow recipe (there can be multiple in the case of oscillators).

It's found these recipes so far (recipes were found by Sylvani):