Prove a backtest in zero knowledge. The deterministic Wickra engine runs inside a RISC Zero zkVM guest, and the receipt proves the report's hash and headline metrics without revealing the candles or the strategy.

▶ Live demos: the backtester compiled to WebAssembly, an equity curve building bar by bar — backtest-live.wickra.org; one StrategySpec side by side in Python, Rust, JS and Go — playground.wickra.org; all 514 indicators of the core over a real Binance feed — live.wickra.org. Zero backend, all of them.

Part of the Wickra ecosystem: the same data-driven core also powers wickra-backtest, wickra-proof, wickra-verify and 20 more — see the full list.

wickra-zk runs a deterministic Wickra backtest as a guest program inside the risc0 zkVM and produces a succinct zero-knowledge proof over it. Anyone holding the proof -- a server, an auditor, a browser -- checks it against the pinned guest and can trust the reported performance metrics without ever seeing the price data or the strategy internals. The proof is a constant-size STARK receipt; wrapping it for an on-chain verifier is on the roadmap.

This works only because Wickra computes byte-deterministically: the report the guest proves is exactly the one wickra-backtest produces natively and wickra-proof canonicalizes and hashes.

use wickra_zk_host::{commit_dataset, prove, verify, ProveOptions, ZkSpec};

// The strategy and the candles are the private inputs; the proof is bound to
// the candles through their canonical hash, which the guest recomputes.
let spec = ZkSpec { strategy, dataset_commitment: commit_dataset(&candles)? };
let proof = prove(&spec, &candles, ProveOptions::default())?;

// Anyone verifies the receipt against the pinned guest and reads the journal
// from it -- the report hash, the commitment, sharpe, pnl, the trade count.
let journal = verify(&proof)?;
println!("report_hash: {}", journal.report_hash);

The same envelope -- prove, commit, verify, version as JSON -- is what the CLI and every binding speak: Python, Node.js, C, C++, C#, Go, Java and R prove and verify natively, and a WebAssembly build verifies in the browser.

Status

0.2.0 — the current release. Early development (0.1.0); See ROADMAP.md.

Documentation

• ARCHITECTURE.md — host/guest split, zkVM choice, determinism chain
• docs/ZK.md — receipt, journal, image ID; what is and is not proved
• docs/DETERMINISM.md — why the journal hash equals the native hash
• docs/PROVING.md — dev-mode vs. production proving, timing, memory
• docs/Cookbook.md — prove/verify recipes
• THREAT_MODEL.md, SECURITY.md

What is proved

• A public report_hash — the canonical hash of the full BacktestReport.
• A small set of public metrics (e.g. sharpe, pnl, n_trades).
• Bound to a specific GUEST_ID (the program that ran) so a verifier knows the honest engine produced the result.

What stays private: the OHLCV candles and the strategy internals — they are guest inputs, never revealed by the receipt.

Quickstart

cargo install wickra-zk
wickra-zk prove  --spec examples/specs/momentum.json --data examples/data/BTCUSDT.csv --out momentum.proof.json
wickra-zk verify --proof momentum.proof.json

The prover is in-process and the compiled guest ships inside the crate, so nothing else is installed. Add --dev (or RISC0_DEV_MODE=1) for a fast, unsound receipt while developing; a real proof takes minutes. One runnable example per language lives under examples/.

Use in any language

The same Prover handle — construct, drive with command(json) -> json, read version — is reachable from every binding, and prove / verify return the same bytes in each of them:

import json
import wickra_zk

prover = wickra_zk.Prover()
print(json.loads(prover.command(json.dumps({"cmd": "version"}))))
proof = json.loads(prover.command(json.dumps({"cmd": "prove", "spec": {"strategy": strategy}, "candles": candles})))
journal = json.loads(prover.command(json.dumps({"cmd": "verify", "proof": proof})))

prove blocks for as long as the zkVM runs — seconds with RISC0_DEV_MODE=1, minutes for a real receipt — in every binding alike; the command envelope is documented once, in docs/ZK.md. The C ABI hub (bindings/c) backs C, C++, C#, Go, Java and R; Rust, Python, Node.js and WASM are native (the WASM package verifies only: a browser checks a proof, it does not produce one). See each bindings/<lang>/README.md and the runnable examples/.

Project layout

crates/wickra-zk-host/     the host: prove, verify, the ZkSpec/PublicOutputs model
crates/wickra-zk-cli/      the `wickra-zk` command line
crates/wickra-zk-bench/    criterion benchmarks over the proving path
guest/methods/guest/       the guest program -- the code that runs inside the
                           zkVM and whose honest execution the receipt attests
guest/methods/             the compiled guest, committed: its ELF and image id
guest/builder/             compiles the guest through risc0-build and writes
                           the two files above; the only risc0 consumer
bindings/                  c, python, node, wasm (verify only), go, csharp,
                           java, r -- one JSON envelope over the host
examples/                  one runnable prove/verify per language
golden/                    frozen (spec, data) -> expected journals, one real
                           receipt, the case-to-dataset map
fuzz/                      libfuzzer targets over the JSON boundary

Building everything from source

git clone https://github.com/wickra-lib/wickra-zk && cd wickra-zk
cargo build --release                     # host, CLI, C ABI, Python and Node crates
RISC0_DEV_MODE=1 cargo test --workspace   # dev-mode receipts: seconds, unsound

The compiled guest is committed, so this needs no risc0 toolchain. Changing the guest does -- see CONTRIBUTING.md for the builder and the reproducible rebuild CI holds it to.

Testing

Run the suites with the commands in Building everything from source.

• wickra-zk-host — the determinism chain, end to end: the native path's report hash, the dev-mode guest's journal, and the blessed fixtures must all agree. That equality is the product; if the guest and the native engine ever disagree, the proof attests to something other than what the engine computes.
• The guest is rebuilt for riscv32im-risc0-zkvm-elf in risc0's build container on every change and compared byte for byte with the committed ELF and image id, and linted for that target separately, because a guest that builds on the host proves nothing about the one that runs in the circuit.
• Every binding proves the golden cases through the envelope in dev-mode and holds the journal to the blessed report hash and metrics; the WebAssembly verifier and the host verify a committed real receipt, so the sound path is exercised on every push without a prover.
• fuzz/ — libfuzzer targets over the spec and journal parsers, run as a time-boxed smoke in CI.
• Nightly — prove.yml runs the real proving path rather than dev mode. Dev mode produces an unsound receipt quickly, which is right for CI and wrong as the only thing ever exercised.

Requirements

• Linux or macOS. risc0 has no Windows host, so neither the crates nor any binding build there; on Windows use WSL.
• Rust 1.90+. The prover is in-process and the compiled guest is committed, so building, proving and verifying need no risc0 toolchain; changing the guest does (see CONTRIBUTING.md).
• On macOS, Xcode's Metal toolchain: risc0 compiles its Metal kernels whenever the prover is built there. xcodebuild -downloadComponent MetalToolchain once; the crate reports cannot execute tool 'metal' until then.
• Per binding: Python 3.9+, Node.js 20+, a C toolchain and CMake, .NET 8 SDK, JDK 22+, Go 1.23+, R 4.1+ -- the floors the manifests declare.
• See CONTRIBUTING.md for the full verify workflow.

Benchmarks

crates/wickra-zk-bench measures the proving path with criterion, and CodSpeed reports instruction counts on every pull request. Absolute proving times depend on the machine and on whether the receipt is real or dev-mode; the numbers worth watching are relative, which is what CodSpeed reports.

cargo bench -p wickra-zk-bench

Ecosystem

Part of the Wickra family — each one a data-driven core with a CLI and the same binding surface:

• wickra — main library (Rust core + Python / Node.js / WASM bindings + a C ABI for C / C++ / C# / Go / Java / R)
• wickra-playground — a polyglot strategy playground: one StrategySpec live side by side in Python, Rust, JS and Go, entirely in the browser
• wickra-exchange — unified market-data + execution across ten crypto exchanges
• wickra-backtest — event-driven backtester over the Wickra core
• wickra-terminal — the trading terminal: a TUI and a browser renderer over the stack
• wickra-screener — parallel multi-symbol screening over 514 streaming indicators
• wickra-radar — perp-universe alert radar: OI delta, funding flip, book imbalance, liquidation clusters, OI/price divergence
• wickra-copilot — local market copilot grounded in real order-book, liquidation and funding microstructure
• wickra-shazam — match an asset's current microstructure fingerprint against its entire history
• wickra-benchmark — reproducible, golden-verified benchmark suite — recompute any (strategy, dataset, report) in ten languages and confirm it byte-for-byte
• wickra-strategy-ci — Jest for trading strategies: golden-pin the report, catch regressions in CI, property-test against fuzzed data
• wickra-verify — confirm or refute a claimed backtest report against its strategy and data, in ten languages
• wickra-proof — Proof-of-Backtest: deterministic (spec, data) → report + blake3 hash, recomputable byte-for-byte in ten languages
• wickra-zk — this repository: prove a backtest zero-knowledge, verifiable anywhere without the data or the strategy
• wickra-impact — the backtester that knows you would have moved the market: agent-based fills on the real historical L2 order book
• wickra-darwin — evolutionary strategy search at millions of backtests per second, mutating and crossing JSON specs across the 514-indicator space
• wickra-gym — a Gymnasium-compatible, microstructure-aware backtest environment with O(1) steps for deterministic RL rollouts
• wickra-feature-store — OHLCV and microstructure streams into ML-ready feature matrices over 514 O(1) streaming indicators

Contributing

Pull requests welcome — see CONTRIBUTING.md. By participating you agree to the Code of Conduct.

Security

Report vulnerabilities privately — see SECURITY.md.

License

Licensed under either of

• Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option. Use it, fork it, modify it, redistribute it — commercially or not — file issues, send pull requests; all welcome.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.

Disclaimer

This software is provided for research and educational purposes. It is not financial advice. A zero-knowledge proof attests only to the honest execution of the pinned guest program over the prover's inputs; it makes no claim about the quality, provenance, or future performance of a trading strategy.