# `of_analytics`
[](https://crates.io/crates/of_analytics)
[](https://docs.rs/of_analytics)
[](../../LICENSE)
`of_analytics` contains additive advanced market microstructure analytics for
Orderflow. The crate exists so heavier analytics can evolve outside `of_core`
without breaking existing `AnalyticsAccumulator`, runtime, C ABI, Python, or
Java APIs.
`of_analytics` starts at `0.1.0` in the broader Orderflow `0.5.0` development
line because it is a new public Rust surface.
The first foundation is dependency-light:
- market-quality/TCA primitives for quoted spread, effective spread, realized
spread, price improvement, quote freshness, and side-aware slippage,
- execution-quality/TCA primitives for implementation shortfall, arrival and
decision slippage, adverse selection, trade-through, and fill-quality score,
- liquidity/depth primitives for top-of-book depth, multi-level depth,
proportional imbalance, order-flow imbalance, depth slope, depth convexity,
book pressure, replenishment/depletion rates, drought detection, and
sweepability,
- market-impact primitives for Kyle-style lambda, Amihud-style illiquidity,
calibrated expected impact, square-root impact, temporary/permanent impact,
impact decay, and child-order attribution,
- VPIN-style fixed-bucket toxicity primitives plus post-trade markout,
adverse-selection, quote-fade, toxic-burst, and informed-flow proxy signals,
- fixed-window volatility/noise primitives with bipower variation, jump
variation, OHLC range estimators, signature-plot points, and intraday
seasonality buckets,
- threshold-based market regime classification plus composite trend/range/chop,
liquidity, spread, session, hidden-liquidity, and transition-confidence
labels,
- feed-quality primitives for sequence gaps, out-of-order events, duplicates,
stale events, locked/crossed books, timestamp skew, resets, health scoring,
replay usability, primary issue selection, and operator review reports,
- feature-vector primitives for stable feature ids, ordered schemas, schema
hashes, missing-value policy, quality labels, and reusable fixed-capacity
writers,
- resiliency primitives for threshold-based spread/depth shock detection,
recovery timing, and liquidity resiliency scoring,
- queue/fill primitives for passive queue position estimates, fill
probability, expected time-to-fill, amend queue loss, top-level survival,
cancel/replace cost, and maker/taker decision scoring,
- pattern-risk primitives for spoofing/layering, quote-stuffing, stop-run,
absorption, momentum-ignition, iceberg, hidden accumulation/distribution,
stacked-imbalance, and failed-breakout risk indicators,
- venue/route primitives for fill, reject, cancel, latency, route-health,
venue-liquidity, toxicity, fill-quality, reliability, route-quality, drift,
and degradation diagnostics,
- cross-asset primitives for rolling correlation, beta, pair divergence,
thresholded basis pressure, latency-adjusted correlation, cross-venue
divergence, ETF/component imbalance, and relationship-degradation
diagnostics,
- derivatives primitives for put/call pressure, volume/open-interest anomaly,
implied-volatility flow, gamma exposure, IV skew, term structure,
implied-versus-realized richness, gamma pressure, futures basis, roll
pressure, funding divergence, and aggregate derivatives stress,
- explicit feature profiles so users can opt into future impact, toxicity,
volatility, regime, data-quality, feature-vector, resiliency, queue-fill,
cross-asset, pattern, derivatives, institutional, and ML feature modules
without forcing all downstream users to compile them.
The crate does not submit orders, manage runtime state, own persistence, or
replace existing `of_core` APIs. It consumes normalized market data and returns
typed snapshots that hosts can wire into runtime, research, or execution
systems.
## Feature Profiles
Default features:
- `market-quality`
- `liquidity`
Reserved additive profiles:
- `impact`
- `toxicity`
- `volatility`
- `regime`
- `data-quality`
- `feature-vector`
- `resiliency`
- `queue-fill`
- `cross-asset`
- `patterns`
- `derivatives`
- `institutional`
- `ml-features`
- `all`
## Public API Meaning
The analytics types are typed calculators and snapshots, not hidden strategy
decisions. Context types such as quote, trade, depth, route, derivative, and
feature inputs carry normalized observations and timestamps. Tracker or
calculator types accumulate a bounded window; snapshot types expose the result
without mutating the input observation. Constructors validate units, capacities,
thresholds, and required metadata before a tracker is used.
The market-quality and execution-quality surfaces explain spread, slippage,
impact, adverse selection, and fill outcomes. Liquidity, queue, resiliency,
toxicity, pattern-risk, volatility, regime, cross-asset, and derivatives
surfaces answer distinct analytical questions rather than producing one
universal signal. Feature vectors and model metadata define stable input
ordering and quality/missing-value semantics for downstream models.
All profiles are additive feature boundaries. A feature flag controls compile
cost and dependency inclusion; it does not silently change the meaning of
existing `of_core` snapshots. These calculations do not place orders, own
persistence, or override `DataQualityFlags`; hosts remain responsible for
quality gating, replay validation, and execution risk.
## Market Quality Example
```rust
use of_analytics::{MarketQualityTracker, QuoteContext, TradeContext};
use of_core::Side;
let mut tracker = MarketQualityTracker::new(1_000_000);
tracker.on_quote(QuoteContext::new(499_975, 500_025, 100, 120, 1_000)?);
let snapshot = tracker.evaluate_trade(
TradeContext::new(500_050, 10, Side::Ask, 1_500)?,
Some(500_075),
)?;
assert_eq!(snapshot.quoted_spread(), 50);
assert!(snapshot.effective_spread_bps() > 0);
# Ok::<(), Box<dyn std::error::Error>>(())
```
## Execution Quality Example
```rust
use of_analytics::{ExecutionBenchmark, ExecutionQualityAnalyzer, TradeContext};
use of_core::Side;
let trade = TradeContext::new(101_000, 10, Side::Ask, 1)?;
let benchmark = ExecutionBenchmark::new(
100_000,
100_500,
99_950,
100_050,
Some(100_750),
)?;
let snapshot = ExecutionQualityAnalyzer::evaluate(trade, benchmark);
assert!(snapshot.implementation_shortfall_bps() > 0);
assert!(snapshot.trade_through());
# Ok::<(), Box<dyn std::error::Error>>(())
```
## Liquidity Example
```rust
use of_analytics::{
LiquidityDepthAnalyzer, LiquidityFlowConfig, LiquidityFlowEvent,
LiquidityFlowTracker,
};
use of_core::BookLevel;
use of_core::Side;
let bids = [
BookLevel { level: 0, price: 499_975, size: 100 },
BookLevel { level: 1, price: 499_950, size: 80 },
BookLevel { level: 2, price: 499_925, size: 70 },
];
let asks = [
BookLevel { level: 0, price: 500_025, size: 120 },
BookLevel { level: 1, price: 500_050, size: 90 },
BookLevel { level: 2, price: 500_075, size: 60 },
];
let snapshot = LiquidityDepthAnalyzer::new(3).analyze(&bids, &asks, 150)?;
assert_eq!(snapshot.bid_depth(), 250);
assert_eq!(snapshot.ask_depth(), 270);
assert!(snapshot.sweepable_buy_qty() >= 120);
assert!(snapshot.book_pressure_bps() < 0);
let mut flow = LiquidityFlowTracker::new(LiquidityFlowConfig::default());
flow.on_event(LiquidityFlowEvent::new(Side::Bid, 100, 0, 0, 1)?);
flow.on_event(LiquidityFlowEvent::new(Side::Ask, 0, 250, 25, 500_000_001)?);
assert!(flow.snapshot().order_flow_imbalance_bps() > 0);
# Ok::<(), Box<dyn std::error::Error>>(())
```
## Impact And Toxicity Example
```rust
use of_analytics::{
ExpectedImpactEstimator, ExpectedImpactInput, ImpactCalibration,
ImpactSample, ImpactTracker, ToxicityAnalyzer, ToxicityConfig,
ToxicityInput, TradeContext, VpinTracker,
};
use of_core::Side;
let mut impact = ImpactTracker::new();
impact.on_sample(ImpactSample::new(500_000, 501_000, 100, 50_000_000)?);
assert!(impact.snapshot().kyle_lambda_ppm() > 0);
let calibration = ImpactCalibration::new(
1_000_000,
200,
10_000,
500,
250,
1_000_000_000,
)?;
let estimate = ExpectedImpactEstimator::estimate(ExpectedImpactInput::new(
Side::Ask,
1_000,
10_000,
100_000,
1_000_000_000,
calibration,
)?);
assert!(estimate.expected_total_impact_bps() > 0);
let mut vpin = VpinTracker::<4>::new(100)?;
vpin.on_trade(TradeContext::new(500_000, 80, Side::Ask, 1)?);
vpin.on_trade(TradeContext::new(500_000, 20, Side::Bid, 2)?);
assert_eq!(vpin.snapshot().bucket_count(), 1);
let toxicity = ToxicityAnalyzer::new(ToxicityConfig::default()).evaluate(
ToxicityInput::new(
TradeContext::new(100_000, 10, Side::Ask, 1)?,
100_500,
1_000,
1_000,
1_000,
500,
8_000,
7_000,
)?,
);
assert!(toxicity.toxic_flow_burst());
# Ok::<(), Box<dyn std::error::Error>>(())
```
## Volatility And Regime Example
```rust
use of_analytics::{
CompositeRegimeClassifier, CompositeRegimeInput, OhlcVolatilityEstimator,
OhlcVolatilityInput, RegimeClassifier, RegimeInput, RegimeKind,
SessionRegimeKind, TrendRegimeKind, VolatilitySeasonalityTracker,
VolatilitySignatureEstimator, VolatilityTracker,
};
let mut vol = VolatilityTracker::<8>::new()?;
vol.on_price(100_000)?;
vol.on_price(101_000)?;
vol.on_price(100_500)?;
let snapshot = vol.snapshot();
assert!(snapshot.bipower_vol_bps() > 0);
let regime = RegimeClassifier::default().classify(RegimeInput::new(
5,
snapshot.realized_vol_bps(),
0,
0,
));
let composite = CompositeRegimeClassifier::default().classify(
CompositeRegimeInput::new(
8_000,
1_000,
1,
20,
2_000,
3_600_000_000_000,
3_600_000_000_000,
false,
0,
0,
)?,
);
assert_eq!(composite.trend(), TrendRegimeKind::Trend);
assert_eq!(composite.session(), SessionRegimeKind::Continuous);
let ohlc = OhlcVolatilityEstimator::estimate(OhlcVolatilityInput::new(
100_000,
102_000,
99_000,
101_000,
Some(99_500),
)?);
assert!(ohlc.parkinson_vol_bps() > 0);
let signature = VolatilitySignatureEstimator::estimate(1_000_000, &[10, -10, 20, -20])?;
assert_eq!(signature.noise_ratio_bps(), 10_000);
let mut seasonality = VolatilitySeasonalityTracker::<2>::new(25)?;
seasonality.on_return(0, 10)?;
seasonality.on_return(0, -30)?;
assert_eq!(seasonality.snapshot(0)?.jump_count(), 1);
# Ok::<(), Box<dyn std::error::Error>>(())
```
## Data Quality Example
```rust
use of_analytics::{
FeedQualityConfig, FeedQualityEvent, FeedQualityFlags, FeedQualityTracker,
ReplayQualityAnalyzer,
};
let config = FeedQualityConfig::new(10, 20, 1)?;
let mut tracker = FeedQualityTracker::new(config);
tracker.on_event(FeedQualityEvent::new(
Some(10),
100,
105,
Some(99),
Some(101),
)?);
let flags = tracker.on_event(FeedQualityEvent::new(
Some(12),
110,
140,
Some(100),
Some(100),
)?);
assert!(flags.contains(FeedQualityFlags::SEQUENCE_GAP));
assert!(flags.contains(FeedQualityFlags::LOCKED_BOOK));
assert!(tracker.snapshot().health_score_bps() < 10_000);
let report = ReplayQualityAnalyzer::default().evaluate(tracker.snapshot());
assert!(report.operator_review_required());
# Ok::<(), Box<dyn std::error::Error>>(())
```
## Feature Vector Example
```rust
use of_analytics::{
FeatureDefinition, FeatureId, FeatureQuality, FeatureSchema, FeatureUnit,
FeatureVectorWriter, MissingValuePolicy,
};
let mut schema = FeatureSchema::<2>::new()?;
let spread = schema.register(FeatureDefinition::new(
FeatureId::new(1)?,
"spread_bps",
FeatureUnit::BasisPoints,
1,
MissingValuePolicy::Sentinel(-1),
)?)?;
schema.register(FeatureDefinition::new(
FeatureId::new(2)?,
"quality_bps",
FeatureUnit::ScoreBasisPoints,
1,
MissingValuePolicy::Zero,
)?)?;
let mut writer = FeatureVectorWriter::new(&schema);
writer.set(spread, 25, FeatureQuality::Good)?;
let vector = writer.finish();
assert_eq!(vector.values(), &[25, 0]);
assert_eq!(vector.schema_hash(), schema.schema_hash());
# Ok::<(), Box<dyn std::error::Error>>(())
```
## Resiliency Example
```rust
use of_analytics::{ResiliencyConfig, ResiliencySample, ResiliencyTracker};
let config = ResiliencyConfig::new(5, 1_000, 25, 5_000, 8, 9_000)?;
let mut tracker = ResiliencyTracker::new(config);
tracker.on_sample(ResiliencySample::new(100, 5, 500, 500)?);
let shock = tracker.on_sample(ResiliencySample::new(200, 30, 400, 400)?);
let recovered = tracker.on_sample(ResiliencySample::new(1_000_200, 7, 500, 500)?);
assert!(shock.active_shock());
assert_eq!(recovered.recovery_count(), 1);
assert_eq!(recovered.last_recovery_time_ns(), 1_000_000);
# Ok::<(), Box<dyn std::error::Error>>(())
```
## Queue And Fill Example
```rust
use of_analytics::{
QueueDecisionAnalyzer, QueueDecisionConfig, QueueDecisionInput,
QueueFillConfig, QueueFillTracker, QueueFillUpdate, QueuePositionEstimate,
QueueUpdateKind,
};
let config = QueueFillConfig::new(5_000, 100, 1_000_000_000)?;
let estimate = QueuePositionEstimate::new(100, 50, 200, 1)?;
let mut tracker = QueueFillTracker::new(config, estimate);
let snapshot = tracker.on_update(QueueFillUpdate::new(
QueueUpdateKind::Trade,
40,
160,
2,
)?);
assert_eq!(snapshot.qty_ahead(), 60);
assert!(snapshot.fill_probability_bps() > 0);
let decision = QueueDecisionAnalyzer::new(
QueueDecisionConfig::new(5_000, 5_000, 10_000_000_000)?,
)
.evaluate(QueueDecisionInput::new(
snapshot,
10,
5,
1,
2,
1,
1_000,
0,
0,
)?);
assert!(decision.maker_taker_decision_score_bps() > 0);
# Ok::<(), Box<dyn std::error::Error>>(())
```
## Pattern Risk Example
```rust
use of_analytics::{
PatternDetailAnalyzer, PatternDetailInput, PatternRiskClassifier,
PatternRiskInput, PatternRiskLiquidity,
};
let classifier = PatternRiskClassifier::default();
let snapshot = classifier.classify(PatternRiskInput::new(
800,
900,
10,
8_000,
5,
PatternRiskLiquidity::new(10, 1_000)?,
1_000_000,
)?);
assert!(snapshot.spoofing_layering_risk_bps() > 0);
assert!(snapshot.quote_stuffing_risk_bps() > 0);
let detail = PatternDetailAnalyzer::default().evaluate(
PatternDetailInput::new(8, 4, 1_000, 100, 4, 8_000, 500, 2, 30, 35, 1_000)?,
);
assert!(detail.iceberg_risk_bps() > 0);
assert!(detail.failed_breakout_risk_bps() > 0);
# Ok::<(), Box<dyn std::error::Error>>(())
```
## Venue Route Example
```rust
use of_analytics::{
VenueRouteEvent, VenueRouteEventKind, VenueRouteQualityAnalyzer,
VenueRouteQualityInput, VenueRouteTracker,
};
let mut tracker = VenueRouteTracker::new();
tracker.on_event(VenueRouteEvent::new(VenueRouteEventKind::Sent, 100, 0, 10)?);
tracker.on_event(VenueRouteEvent::new(VenueRouteEventKind::Fill, 60, 100, 20)?);
let snapshot = tracker.snapshot();
let quality = VenueRouteQualityAnalyzer::default().evaluate(
VenueRouteQualityInput::new(snapshot, 9_000, 500, 9_500, 9_500)?,
);
assert_eq!(snapshot.sent(), 1);
assert_eq!(snapshot.fills(), 1);
assert_eq!(snapshot.avg_quote_to_fill_latency_ns(), 100);
assert!(!quality.route_degraded());
# Ok::<(), Box<dyn std::error::Error>>(())
```
## Cross-Asset Example
```rust
use of_analytics::{
CrossAssetConfig, CrossAssetDiagnosticAnalyzer, CrossAssetDiagnosticInput,
CrossAssetSample, CrossAssetTracker,
};
let mut tracker = CrossAssetTracker::<8>::new(CrossAssetConfig::default())?;
tracker.on_sample(CrossAssetSample::new(100_000, 200_000, 1)?);
tracker.on_sample(CrossAssetSample::new(101_000, 202_000, 2)?);
tracker.on_sample(CrossAssetSample::new(102_000, 204_000, 3)?);
let snapshot = tracker.snapshot();
let diagnostic = CrossAssetDiagnosticAnalyzer::default().evaluate(
CrossAssetDiagnosticInput::new(snapshot, 1_000_000, 1_100_000, 5, 100)?,
);
assert!(snapshot.correlation_bps() > 0);
assert!(snapshot.lead_lag_score_bps() > 0);
assert!(diagnostic.synchronization_quality_bps() > 0);
# Ok::<(), Box<dyn std::error::Error>>(())
```
## Derivatives Example
```rust
use of_analytics::{
DerivativesDiagnosticAnalyzer, DerivativesDiagnosticInput,
DerivativesVolatilitySurface, FuturesBasisAnalyzer, FuturesBasisInput,
OptionFlowSample, OptionFlowTracker, OptionKind,
};
let mut options = OptionFlowTracker::new();
options.on_sample(OptionFlowSample::new(
OptionKind::Call,
100,
1_000,
50_000,
2_000,
1_000,
)?);
options.on_sample(OptionFlowSample::new(
OptionKind::Put,
200,
1_500,
150_000,
3_000,
-2_000,
)?);
assert!(options.snapshot().put_call_pressure_bps() > 0);
let basis = FuturesBasisAnalyzer::analyze(FuturesBasisInput::new(
100_000,
101_000,
100_500,
101_000,
102_000,
25,
)?);
let diagnostic = DerivativesDiagnosticAnalyzer::default().evaluate(
DerivativesDiagnosticInput::new(
options.snapshot(),
basis,
DerivativesVolatilitySurface::new(3_000, 3_500, 2_500, 3_000, 3_250, 2_000)?,
),
);
assert!(basis.basis_bps() > 0);
assert!(diagnostic.iv_richness_bps() > 0);
# Ok::<(), Box<dyn std::error::Error>>(())
```
## Low-Latency Principles
- No async runtime dependency.
- No heap allocation in hot update/evaluate methods.
- Borrow existing `of_core` book levels instead of copying snapshots.
- Use integer arithmetic for prices, quantities, spreads, and basis points.
- Keep batch/research features separate from live hot-path trackers.