LedgerLoops
Experiments / show your work
Ideas need
something to run.
Follow the experiments from a small browser simulation to loop-search benchmarks and peer-to-peer coordination.
The browser workbench
The historical browser demo lets you act as a participant and send simulated IOUs. Open the developer console to follow the messages. It is an early experiment with its original limitations, not a place for real accounts or funds.
Strategy Pit: finding loops
Strategy Pit compares loop-search strategies. The website’s 2024 write-up describes experiments on a debt graph generated from the Sarafu transaction dataset: roughly 440,000 transactions between 55,000 accounts.
The graph is a constructed experiment: standard Sarafu transfers are treated as changes to bilateral balances, not as evidence of actual unpaid debts. A snake-like depth-first search follows a path until it meets its own tail. Strategy Pit compares this centralized DFS baseline with min-cost flow followed by DFS.
| Approach | Debt cleared |
|---|---|
| Min-cost flow + DFS | 67% |
| Snake-like depth-first search | 60% |
The same write-up reports about 1,000 loops per second for a roughly 350-line TypeScript implementation on a laptop. These are historical results from that experiment, not fresh measurements or guarantees for a live network.
Jerboa: coordination between peers
The website introduced Jerboa in October 2024 as a prototype demonstrating the feasibility of peer-to-peer obligation clearing.
The published prototype used a centralized semaphore service. Later work changed that: a June 2025 commit removed the semaphore, followed by experiments with randomized probe timers on individual nodes. The README still describes the earlier design in places, so read the code and history together. A search benchmark and a distributed implementation answer different questions.
Jaribu: another path through the graph
Jaribu began in June 2025 as another attempt at peer-to-peer cycle detection through depth-first-search messaging. Its history explores message queues, multiple search “worms,” and their interactions. The README calls it work in progress.
Useful questions to investigate
- What happens when messages arrive late or in a different order?
- How does a peer recover when another peer stops responding?
- Which information does a search reveal to its neighbors?
- How much debt can be cleared, and at what messaging cost?
- Which assumptions change when a simulation becomes a network?
Start with the repository’s own instructions and tests. Record the version, dataset, parameters, and limitations when sharing a result.
Pick an experiment →