settlemint

What Settlemint proves, and what it does not

Written so a reviewer does not have to find the gaps by reading source. Everything below is checkable in this repo.

What is real and measured

• The NEAR settlement core. contract/src/lib.rs is the whole milestone-1 mechanism in one file, with no dependency beyond near-sdk: the four-phase lifecycle (Open, Resolved, Settled, Voided), resolver-only resolution, pro-rata pari-mutuel settlement, a last-winner dust sweep, and a permissionless void escape hatch. cargo test runs 23 unit tests against it on the NEAR unit-testing harness, offline.
• A deployable artifact. cargo build --release --target wasm32-unknown-unknown produces a 205 KB WASM contract. That is the artifact a testnet deploy would upload. Building it is proof the contract compiles for the NEAR runtime, not just for host tests.
• A cross-language mirror, pinned. settlemint/model.py reimplements the same settlement arithmetic in Python. tests/test_parity.py asserts the exact payout literals the Rust unit tests assert (150/50 pro-rata, 1/1/3 dust sweep, 140/100 liability, and the refund and void cases), so the on-chain and off-chain math cannot drift without failing a test on one side. 32 Python tests pass.
• A recorded run, and the UI renders it. scripts/record_run.py drives a full market lifecycle through the offline client and writes docs/run.json. tests/test_run_fixture.py asserts the committed run conserves value to the last yocto, ends Settled with zero escrow, and reproduces exactly on a fresh run. ui/index.html renders that same data, so nothing on the page is a typed placeholder.
• The adapter seam is testnet-only by construction. tests/test_near_adapter.py proves LiveNearClient refuses a mainnet RPC URL and any non-testnet host, and that build_client() falls back to the keyless offline backend when no credentials are set.

The recorded run is an offline fixture, not a public testnet receipt

docs/run.json is produced by the offline NEAR client, which runs the same arithmetic as the contract but does not touch a network. It is a genuine execution of the settlement math, but there is no public block explorer link for it, because no funded testnet key is shipped in this repo. The live backend (settlemint/near.py) targets NEAR testnet given NEAR_RPC_URL and a signer; producing a public-testnet run with a real deploy transaction is the explicit next step, committed for milestone 1’s on-chain verification, not something already done here.

What is NOT built, deployed, or measured

• No mainnet, ever, in this scope. All NEAR work is testnet only. The live client rejects mainnet endpoints in code.
• No public testnet deploy transaction yet. The WASM builds; this repo does not ship a funded key, so it records the run offline and does not claim a deploy receipt it cannot show.
• Milestones 2 to 4 are not in this repo. The verifiable-resolution hash commitment (milestone 2), the operator agent driving NEAR over a scoped access key on vendored agent_core (milestone 3), and the forkable template (milestone 4) are scoped out. agent_core is vendored ready for milestone 3 but the milestone-1 core does not depend on it.
• No users, no revenue, no partnerships, no audit. The 55 passing tests across the two suites are our own, not a third-party audit. No real market has been run for anyone.
• The NEAR Rewards cohort window is not confirmed on an official page. The programme mechanics and tiers were verified; the live cohort number and deadline were not. An operator must confirm the open window on the Airtable form before applying.

We would rather a reviewer read this section first. The claim is narrow and true: the NEAR settlement core is written, compiles to a testnet WASM, and is tested to 55 passing checks with cross-language parity, today.

This site is open source. Improve this page.