Paganini is the Rust quantlib I use for market making, microstructure, and options dealer flow. Flow Master is the commercial packaging of it — the buy-side half of Quant Masters. This post is about what’s actually in the library, not the product page.

The shape, straight from the README:

ticks ─► LOB ─► features ─► quoter ─► OMS ─► fills ─► PnL
           │       │            │
           ▼       ▼            ▼
        Microprice  AS / GLFT    Avellaneda-Stoikov
        OFI         skew         (per-strategy)
        Welford RV  GPU dot
        Hawkes
        VPIN / Lee-Ready / Kyle's λ

Microstructure features

Eight streaming features over a price-level LOB (O(log K) update, O(1) best): microprice, order-flow imbalance, depth imbalance, Welford realized variance, Hawkes intensity, VPIN, Lee-Ready, Kyle’s λ. Streaming and batch APIs share one algorithm set.

The pricing crate

paganini-pricing is pure Rust: forward Black-Scholes price plus a 12-Greek set, three IV inverters —

• Schadner variance-space Halley (3 iterations ≈ 78 ns, ~2.2× a Jäckel inverter),
• safeguarded-Newton Jäckel,
• Corrado-Miller direct,

and five surface parameterizations (SABR / Wing / SVI / SSVI / eSSVI), each with Nelder-Mead, BFGS, and ridge-regularized calibration variants. On top of the standard no-arb checks there’s PAV isotonic + iterative-local butterfly repair for the cases where the raw surface isn’t monotone.

The vol-trading toolkit

The post-v0.3 work is the variance/skew layer:

• Variance swaps — Demeterfi-Derman-Kamal-Zou fair-strike, forward variance with a term-structure walker, and a VIX-style 30-day constant-maturity vol.
• Five OHLC realized-vol estimators — close-to-close, Parkinson, Garman-Klass, Rogers-Satchell, Yang-Zhang.
• Delta-quoted skew — 25-Δ risk reversal / butterfly with a strike-at-delta inverter.
• Per-Greek P&L attribution — decompose realized P&L back onto the Greeks.

There’s also a position-aggregating Portfolio with per-strike and per-tenor Greek buckets, eight OptionStrategy templates (verticals, straddle, strangle, iron condor, risk reversal, butterfly, calendar / diagonal), an IncrementalGreeks Taylor-update pricer, and inverse-quote (Deribit-style coin-margined) converters.

Market making

For the perp / linear desk: Avellaneda-Stoikov, GLFT, and inventory + OFI skew quotes. For the options desk (paganini-mm::options_mm): a vol-AS quoter with vega-bucket caps and a delta-band auto-hedge.

Numbers, with the caveats attached

From the Criterion suite, single core: bs_price ~13 ns, iv_corrado_miller ~5 ns, quote_chain(5 strikes) ~210 ns. SABR calibration round-trips to <1 bp RMSE. Broad test + doctest coverage across the workspace, with cargo-fuzz targets running clean.

These are self-benchmarked microbenchmarks, single core — useful for relative comparison, not a throughput SLA. The GPU backends (Metal, OpenCL, CUDA) are real kernels validated against the CPU reference. The Lean4 specs are sorry-free; the one non-theorem is an explicit axiom in the Avellaneda–Stoikov spec (first_order_optimal) standing in for the not-yet-formalized HJB optimality — the rest (Welford, microprice, OFI, Almgren–Chriss schedule properties) are closed by tactic.

What this library deliberately isn’t: a live trading system. Quoting, routing, and holding inventory against real venues is Trade Master / Vivaldi; the sell-side pricing core is Price Master / Sibelius.