Mathematically correct software for finance.
We build software for finance, where the numbers have to be right: reports for CFOs, smart contracts that hold money, and financial research.
“God made the integers; all else is the work of man.”
Leopold Kronecker · 1823–1891
Our products
Investment reporting, and tools for building safely on the Stellar blockchain.
Investment reporting
Reports a CFO can trace, figure by figure
Haruno brings a company's QuickBooks Online data together and drafts the statements, the management report and the investor report for a CFO to review. It explains what moved a line by naming the transactions behind it, tracks cash, burn and runway across a portfolio, and analyses actuals against budget and forecast. When an input is missing it names the input instead of estimating around it.
$300 per company, per month
Smart contracts
Money arithmetic that cannot round or wrap
Inference is a programming language for the logic that holds and moves funds. Amounts are integers, overflow stops the program instead of wrapping, and there is no floating point to round a balance. You write the rules a contract must keep, and they are checked against every possible input rather than a sample. It compiles to WebAssembly and deploys on Stellar.
v0.0.6 available now
On-chain finance security
Tools for reviewing the contracts that hold assets
We build and maintain the Stellar Security Portal, a public record of audit reports and vulnerabilities for smart contracts on Stellar. We publish a decompiler that recovers readable Rust from a deployed contract, and we have verified properties of Polkadot's balances module in public. The work is open, including the parts that did not go to plan.
Open source, in a live ecosystem
How we work
What “mathematically correct” means here
Say what must be true.
We start by writing down the rule: a balance can never go below zero, a total must equal its parts. If a rule is not written down, nobody is checking it.
Check every case, not some of them.
A test only tries a few examples. We check each rule against every possible input, and every number in a report traces back to its transactions.
Fail loudly.
When something is wrong, the software says so. A calculation stops when a number gets too large to store, instead of carrying on with a wrong one. A report tells you when data is missing instead of guessing.
Finance R&D
Some problems are not a product. Before capital goes into a market or a protocol, we do the research: how the mechanism behaves, where money is made and lost, and what it takes to operate there. Then we build the models and the tools.
- Market and protocol research
- Capital deployment models
- Custom analytics and tooling
- See the R&D work
Start a conversation
Tell us what you are working on and what has to be true about it. An engineer reads every message and replies.