Data Signals Compute Use Cases Docs
Log in Get access

The canonical market data truth

Blizzard captures markets at the source — inside exchange data centers — with nanosecond fidelity. Everything downstream — replay, features, signals, theos, analytics — is built on that foundation.

Talk to us Request documentation
The Stack

Each layer enables the next

You can't have good signals without good features. You can't have good features without canonical events. You can't have canonical events without source-level capture.

1

Capture Topology

Colocation presence inside exchange data centers. Hardware timestamps at the source. Raw PCAP capture. This is why everything else works.

25+ venues Nanosecond precision Raw PCAP
2

PCAP → Parquet

Decode raw packets. Parse exchange-native protocols. Normalize into structured Parquet. This is where most vendors fail.

Protocol decoding Normalization Structured Parquet
3

Data Products

Standardized datasets. TAQ, exchange-constructed consolidated tape, L2/L3 books, order flow. Venue rules, auctions, halts encoded.

TAQ Consolidated tape L2/L3 books Order flow
4

Canonical Events

Universal atoms — best_bid, best_ask, trade, order events. Deterministic replay. The ground truth layer.

Universal schema Deterministic replay 20PB history
5

Features & Compute

DAG engine composing atoms into features. Pre-computed catalog. Custom definitions via DSL. Operates on Parquet.

C++ DAG engine Feature catalog Parquet output
6

Signals & Theos

Proprietary derivations. Fair values, Greeks, signed flow. Only possible because of layers 1–5.

Proprietary signals Greeks derivation Real-time + historical
7

LLM & Orchestration

Natural language interface to the entire stack. Semantic schema. The interface to everything above.

Natural language queries Semantic layer Context-aware
Coverage

Global markets, normalized

US Equities

Exchange-constructed consolidated tape. A higher-performance complement to SIP.

History
10+ years
Venues
15+ exchanges

EU Equities

Consolidated tape that actually works. MiFID II compliant. Same format as US.

History
8+ years
Venues
20+ venues

Asian Equities

Major APAC markets. Same format, same semantics, same API.

History
5+ years
Venues
10+ exchanges
View equities coverage →

US Options

Full OPRA feed, compressed 10:1. Every strike, every expiry, every exchange.

History
10+ years
Contracts
1.5M+ daily

Greeks & Theo

Pre-computed Greeks at every tick. Delta, gamma, vega, theta. Implied vol surfaces.

Latency
Sub-ms
Models
Black-Scholes, BAW

Chain Reconstruction

Indexed by moneyness and delta, not just symbol. Replayable book state.

Indexing
Delta buckets
Format
Parquet
View options coverage →

CME Group

ES, NQ, CL, GC, ZB, ZN and full complex. Globex depth, implied quotes, spreads.

History
15+ years
Products
All listed

Eurex & ICE

European derivatives. Same normalized format. Same API as CME.

History
10+ years
Coverage
Full order book

Asia-Pacific

SGX, OSE, HKFE. One codebase for all venues, all asset classes.

History
8+ years
Venues
5+ exchanges
View futures coverage →

Signed Volume

Aggressor-tagged transaction flow. Know who's lifting vs hitting.

Latency
Real-time
Coverage
All venues

Fair Values

ETF iNAV, ADR fair value with FX adjustment, options theo. Sub-second updates.

Instruments
ETFs, ADRs, Options
Update
Continuous

Benchmarks

VWAP, TWAP, auction estimates. Computed from our own flow, not delayed prints.

Types
VWAP, TWAP, POV
Source
Proprietary
View all signals →
20PB
Normalized history
25+
Global venues
10+
Years of data
ns
Timestamp precision
Capabilities

What you get

Blizzard replaces bespoke TAQ normalization, replay tooling, feature pipelines, and large parts of internal market-data infrastructure.

Normalized TAQ

Exchange-constructed consolidated view. A higher-performance complement to SIP.

Deterministic replay

Reproduces market state exactly, every time

Market semantics

Auctions, halts, venue rules built in

Security master

Golden source symbology you can trust

Research = Production

Same logic, same outputs, same results

Multi-point timestamps

Same data captured with precision at multiple colos via fiber and microwave. Recreate low-latency strategy opportunities.

Live streaming

History to live with the same API

Raw PCAPs available

Full packet captures when you need protocol-level

Compute

Features, Not Data

No need to buy the data. Define features, run them against the archive, Parquet lands in your bucket.

Pre-computed feature catalog — instant, no playback
Custom features via DSL or JSON
Universal atoms: best_bid, best_ask, trade, order events
Run against full history — 20PB
Output to your S3 / GCS bucket
Managed compute and scheduling
# Define a feature, run it on managed infra from blizzard.dag import Feature, book @Feature("imbalance_5lvl") def imbalance(ctx): bid = book.bid_size(5).sum() ask = book.ask_size(5).sum() return (bid - ask) / (bid + ask) # Submit — Parquet lands in your bucket imbalance.compute( symbols=["AAPL", "MSFT"], start="2014-01-01", end="2024-01-01", output="s3://your-bucket/" )
Vantage Signals

Derived data you can't get anywhere else

Computed from our own trading infrastructure. Not repackaged feeds — signals that require being inside the market to generate.

Signed Volume

Aggressor-tagged transaction flow. Know who's lifting vs hitting.

ETF iNAV

Real-time intraday NAV for ETFs. Sub-second fair value.

ADR Fair Value

Cross-listed arbitrage signals with FX adjustment.

Options Theo

Theoretical values computed in real-time. Not delayed Greeks.

VWAP / TWAP

Benchmark prices computed continuously, not EOD.

Auction Volume Modeling

Predicted auction volume and participation before the cross.

Additional signals — gamma exposure, dark pool estimates, spread decomposition — available under NDA.

Platform

Beyond historical data

A complete market data platform, not just an archive. Live feeds, raw captures, feature compute, and post-trade analytics.

Live Feeds

L1/L2/L3 redistribution. Normalized or raw. Live bars at any interval.

Raw Captures

PCAPs with nanosecond hardware timestamps. FPGA symbol splits available.

Feature DAG

Compute features at scale. Base nodes + DSL. Same graph in research and production.

Real-Time TCA

Cloud dashboard with PB drop copy ingestion. Markouts, venue analysis, market impact — as it happens.

Infrastructure

Deployed where markets trade

Blizzard operates infrastructure in the same colocation facilities as the world's major exchanges.

All locations shown represent active exchange colocation deployments.

Global datacenter presence map
North America
NY4 · NY5 · Carteret · Mahwah · Cermak · Aurora
Europe
LD4 / LSE · FR2 · ZRH · MIL · OSL
Asia Pacific
TY3 · HK1 · SG1 · BOM · SYD
Data Pipeline

Every layer of the stack

Access data at any level — from raw packet captures to derived signals. Most vendors only offer one layer. We captured it, we normalized it, we computed on it.

Raw PCAPs
Unmodified packet captures with hardware timestamps. Native protocol.
Use case
Latency analysis, protocol research
Normalized PCAPs
Merged, deduplicated, symbol-split. FPGA-filtered by instrument.
Use case
Feed handler testing, execution analysis
Normalized TAQ
Trades and quotes in Parquet. Venue rules, auctions, halts encoded.
Use case
Alpha research, backtesting, replay
Derived Signals
Computed features. Signed volume, fair values, Greeks, benchmarks.
Use case
Signal generation, TCA, real-time trading
Increasing abstraction — access any layer, or all of them
Infrastructure

Built different

Capabilities from operating market infrastructure, not reselling feeds.

Multi-path capture

Same tick via fiber and microwave across colos

Options replay

Compressed OPRA, contracts by moneyness and delta

FPGA infrastructure

Hardware timestamping and symbol filtering at line rate

20PB depth

10+ years continuous, nanosecond precision

Additional capabilities — TRF decode, ICE direct API, ML compute — available under NDA.

For institutions

Why firms adopt Blizzard

Research matches production

Same data, same semantics, same results. No translation layer, no drift.

Answers can be defended

Full provenance. Auditors get reproducible results. Compliance gets answers.

Complexity drops materially

Replace fragmented pipelines with a single, governed source of truth.

Team

Built by practitioners

Infrastructure that already runs markets — now available externally.

Head of Infrastructure
Head of Market Data
Head of Execution
Head of Quant Research

This is how the market data layer gets built now.

If this matters to you, you already know why.

Talk to us Request documentation