---

Wickra Verify

Verify any backtest. Hand over a (strategy, data, claimed report) triple and get a deterministic verdict — confirmed or refuted — that anyone can recompute in ten languages.

Part of the Wickra ecosystem. Built on the same deterministic backtest engine and ten-language binding surface as wickra-backtest, wickra-proof and the rest.

wickra-verify takes a claim — a strategy spec, the candle data it was run over, and the BacktestReport someone says that run produced — and recomputes the report with the pinned wickra-backtest engine. It then compares the claimed report against the fresh one, field by field, within an explicit tolerance, and returns a verdict: matches: true if every metric agrees, or a stable, sorted list of the exact mismatches if not.

It is a free anti-fraud tool against doctored backtests — not a hosted service, not a SaaS, not a backend. A CLI plus ten language bindings, and an optional static in-browser WASM demo. Nothing you submit ever leaves your machine.

Determinism is the product

• Recompute, never trust — the verdict comes from re-running the backtest, not from comparing two supplied numbers. A fudged Sharpe, an inflated PnL, a quietly-changed parameter all surface as a concrete mismatch.
• Explicit tolerance — floats are compared within a fixed atol + rtol·max(|a|,|b|) tolerance (the numpy.allclose rule), never bitwise, so legitimate last-ULP noise never causes a false refutal.
• Stable, sorted mismatches — the mismatch list is ordered by field, so the verdict is byte-identical across all ten languages.
• Canonical hashes — every verdict carries the blake3 hashes of the claimed report, the recomputed report and the full inputs, under the same canonicalization wickra-proof uses; a verdict's inputs_hash equals the proof hash of the same inputs.

Status

Pre-release — functionally complete, CI-verified, not yet published. The core, the CLI, all ten language bindings, the byte-exact golden corpus, the property + fuzz suites, the benchmarks and one runnable example per language are built and green across Linux, macOS and Windows. Packages are not yet on the registries. Track progress in ROADMAP.md.

Documentation

• docs/ARCHITECTURE.md — how the pieces fit together.
• docs/CLAIM_FORMAT.md — the Claim and its embedded StrategySpec / BacktestReport.
• docs/VERDICT.md — the Verdict, mismatches and tolerance.
• docs/CANONICALIZATION.md — the hashing contract shared with wickra-proof.
• docs/DETERMINISM.md — why the verdict is identical everywhere.
• docs/Cookbook.md — recipes, including "gate a claim in CI: exit 2 = fraud".

Quickstart

# Verify a claim: recompute the report and compare it, field by field.
# --claim is a JSON/TOML Claim; --data a CSV or a directory of <SYMBOL>.csv files
# (omit --data when the claim carries its candles inline).
cargo run -p wickra-verify -- \
  --claim examples/data/claims/fudged.json \
  --data examples/data/candles

# Exit 0 = verified, 2 = refuted (CI-friendly), 1 = error.
# --explain renders the mismatches and forces exit 0.

The example claim inflates fees_paid, so the verdict is refuted and names the fees_paid mismatch — a fabricated number cannot pass a recomputation.

Claim format

A claim is the assertion to be checked — "this strategy on this data produced this report":

• strategy — the embedded wickra-backtest StrategySpec (indicators, entry/exit rules, sizing, costs, risk).
• dataset_ref — where the candles come from: inline (embedded per symbol, fully self-contained) or files (named symbols resolved out of band).
• claimed_report — the BacktestReport the claimant asserts this run produced. Untrusted: verification recomputes and compares, so a doctored claimed_report cannot pass.

Full schema in docs/CLAIM_FORMAT.md.

Verdict and tolerance

verify returns a Verdict:

• matches — true iff every compared field agrees within tolerance.
• mismatches — a field-sorted list of {field, claimed, actual, delta}.
• claimed_report_hash / actual_report_hash / inputs_hash — 64-hex blake3 digests over the canonical forms.
• engine_version — the wickra-backtest version the verdict was reached under.

Two floats count as equal when |a − b| ≤ atol + rtol·max(|a|,|b|), with tight defaults (atol = 1e-9, rtol = 1e-6) that absorb last-bit float noise and nothing more. See docs/VERDICT.md.

Canonicalization and hashing

The hashes are only as trustworthy as the serialization they run over, so canonicalization is byte-for-byte the contract wickra-proof uses (see crates/verify-core/src/canon.rs): keys sorted by code point, no structural whitespace, floats quantized to 1e-8 with trailing zeros trimmed and whole values collapsed to integers, no NaN/±inf. blake3 over that canonical string yields each 64-hex hash. A verdict's inputs_hash therefore equals the wickra-proof hash of the same inputs — the two products share one determinism moat. See docs/CANONICALIZATION.md.

Anti-fraud use in CI

Because a refuted claim exits 2, verification drops straight into a pipeline:

# Fails the job (exit 2) if the committed report does not match a fresh run.
wickra-verify --claim claim.json --data candles/

Wire this into a strategy repo's CI and a doctored claimed_report can never be merged — the recomputation is the gate. Recipes in docs/Cookbook.md.

Use in any language

The core is a JSON-over-C-ABI data API (Verifier::command) exposed natively in Rust, Python, Node.js and WASM, and over the C ABI hub in C, C++, C#, Go, Java and R. Every binding drives the same verify / explain / canonicalize / version commands and returns the core's canonical response verbatim; the cross-language golden tests assert byte-for-byte equality. One runnable example per language lives under examples/; per-binding quickstarts are in each bindings/<lang>/README.md.

Language	Binding	Package
Rust	verify-core (native)	crates.io
Python	PyO3 (native)	PyPI
Node.js	napi (native)	npm
WASM	wasm-bindgen (native)	npm
C / C++	C ABI	header + library
C#	C ABI (P/Invoke)	NuGet
Go	C ABI (cgo)	Go module
Java	C ABI (FFM/Panama)	Maven
R	C ABI (.Call)	R-universe

Project layout

crates/verify-core          the library: claim + compare + canonicalize + verify
crates/wickra-verify-cli    reference CLI (verify), binary `wickra-verify`
crates/verify-bench         Criterion benchmarks
bindings/{c,python,node,wasm,go,csharp,java,r}   ten-language surface
golden/                     fixed claims -> expected verdicts (byte-exact)
examples/                   runnable per-language demos + static web demo
fuzz/                       cargo-fuzz targets (claim parse, compare, canonicalize, verify)

Building from source

cargo build --workspace
cargo test --workspace --all-features

Each binding builds with its own toolchain; see bindings/<lang>/README.md. The C-ABI consumers (C/C++, C#, Go, Java, R) need the C ABI library first: cargo build --release -p wickra-verify-c.

Requirements

Rust 1.86 (workspace) / 1.88 (Node binding). Per-binding toolchains: Python 3.9+, Node.js 22+, .NET 8, JDK 22+, Go 1.23+, R release, and a C11/C++14 compiler with CMake for the C example.

Benchmarks

Criterion benchmarks for verify, compare and canonicalize live in crates/verify-bench; numbers and methodology are in BENCHMARKS.md.

Contributing

See CONTRIBUTING.md and the Code of Conduct. Every change runs the full CI matrix (all ten languages × three OSes) plus CodeQL, Scorecard and zizmor.

Security

Report vulnerabilities per SECURITY.md. The threat model is in THREAT_MODEL.md.

License

Dual-licensed under either MIT or Apache-2.0, at your option.

Disclaimer

wickra-verify is research and engineering tooling, not financial advice. A verdict attests only that a claimed report is (or is not) the deterministic result of a given strategy over given data — it makes no claim about the quality, profitability or future performance of any strategy, nor about whether the data itself is genuine. Trading carries risk; you are responsible for your own decisions. wickra-verify is free software you run yourself: no hosted service, no data collection, no warranty.