Reporting the Successful Results of the MANTIS Solver Testing Period

0xbrainjar, Dzmitry Lahoda


The solver testing period for MANTIS has been completed. In this testing period, a small number of solvers worked to perform order matching along on the Picasso Cosmos Chain using our solution optimization algorithm. The solvers were able to successfully match orders according to Coincidence of Wants (CoWs), providing a proof of concept for Composable’s solution optimization algorithm and MANTIS intent settlement framework.

Experimental Design

For the last month and a half or so, Composable has been running an experiment to test the solution optimization algorithm that Composable co-created with research collaborator Bruno Mazorra. This is a novel algorithm for determining the optimal pathway for intent settlement across multiple blockchains.

In the present experiment, this solution optimization algorithm was leveraged by solvers, a term for entities that compete to determine an optimal solution for a given transaction. All solvers used the provided algorithm, though in the future we envision that solvers will customize this algorithm with the aim of making it as optimized as possible.

In this experiment, the transactions were orders that were generated by Composable. It is important to note that these orders are a simplified version of intents. Intents can be defined as users’ desires for a given transaction or other outcome. Intents include desired parameters (such as to swap X amount of A token for B token), but leave some room for flexibility (such as where this swap occurs) in the solution that solvers provide.

An example of an intent (taken beyond basic blockchain swaps and into a real-world scenario) is:

I live at coordinates X,Y on Earth, I need 20 MB internet; please make it happen for A amount of token T.

Intents in general thus necessitate:

  • Some sort of AI input (in order to find a solution)
  • Time delayed verification, with verification systems bound together (e.g. verification system must be composable)
  • Emulated atomicity (including with fiat/real world assets)

In the future, MANTIS will be compatible with crypto transaction intents, which, compared to orders, possess the following characteristics/capabilities:

  • Can post from any chain
  • Can route results to the final chain if it was settled via CoW
  • Can be compatible with NFTs
  • Can cover real assets
  • Can allow for staking/unstaking (i.e. transactions beyond simple swaps)

For the scope of this experiment, more simplified orders were used. These were transactions with cross-chain atomicity and swaps.

The tokens involved in the orders in this experiment were:

  • Picasso (PICA)
  • Osmosis (OSMO)
  • Neutron (NTRN)

Orders were generated on all sides with random limits, having a variability around 20%. Order prices were kept extremely low in order to run the experiment for an extended period of time to collect as much data as possible. Prices for assets were hardcoded.

Solvers were able to access these orders as they were created on the Picasso Cosmos chain. Then, the solution algorithm was applied by solvers to match orders along the principle of Coincidence of Wants. Along this principle, orders can coincidentally be the opposite of other orders (i.e. one intent to swap A for B and another to swap B for A form a CoW) and can thus be matched and used to settle each other. Solvers were able to provide their own liquidity to settle the CoWs they found.

It is important to note that solvers in this testing period did not use constant function market makers (CFMMs) such as DEXes as a part of any solution; instead, solutions were limited to CoWs and solvers’ own liquidity for the scope of this experiment. In the future, MANTIS will allow for solutions that comprise any combination of CoWs, solvers’ own liquidity, and CFMMs.

As all solvers were using the same algorithm, they would have been able to create the same matches/solutions. Thus, it was the solver that created the solution first that “won” and was able to complete the order. As mentioned, in the future, we imagine that solvers will modify the provided solution algorithm in the hopes of providing better (and not just faster) solutions.

Additional resources on the design of this experiment are as follows:

  • The solution optimization algorithm was previously detailed in our forum post here
  • We detail this algorithm in full in our research paper published here
  • The algorithm itself can be found here


A total of 3 solvers participated in this experiment. Two were run by Composable, with a third run by an external party. The solver run by the third party performed the greatest amount of order matches during this period (6741).

On average, solvers participated in the experiment for 42 ⅓ days (range: 34 to 49). These solvers matched an average of 4276 ⅓ orders (range: 2760 to 6741) during the experimental period. The average volume cleared by solvers during this period was 419088567 (range: 257641773 to 671478241), with the units for this value being the raw volume of aggregated tokens (e.g. a multiplication of raw token representations, which we termed “volume_amount”). Complete data for the performance of each solver can be found in Table 1 below.

Table 1: Operational data for solvers participating in the testing period.

Solver # Number of Days Solvers Operated Orders Settled volume_amount
1 44 6741 671478241
2 49 3418 328145687
3 34 2670 257641773
Average 42.33 4276.33 419088567

A total of 73,640 orders were created for this testing period. 17.06% of these were filled (13.95% fully filled and 3.11% partially filled). Complete data for orders in the testing period is found in Table 2 below.

Table 2: Data for orders created and filled during the testing period.

Order Status Total Orders Fitting Status Percent of Orders Fitting Status
Created 73640 100.00%
Fully Filled 10274 13.95%
Partially Filled 2289 3.11%
Fully + Partially Filled 12563 17.06%

Note: A fourth solver was created but only participated for a very brief period of time and was therefore excluded from the above data.


This experiment shows the utility of the solution optimization algorithm that Composable co-created with research collaborator Bruno Mazorra. Specifically, solvers were able to use this algorithm in realistic scenarios to match various orders along the principle of CoWs.

The next steps of developing and deploying solvers on MANTIS are as follows:

  • Expand the use of the solution algorithm to order settlement using not only CoWs but also constant function market makers (CFMMs), additional chains, and additional asset types
  • Onboard/deploy additional solvers
  • Have solvers solve intents, not just orders, including intents for actual users that solvers must then settle
  • Perform additional round(s) of solver testing