all.sourceAllSourceEvent Store
Menu
Sign inStart 14-day trial
Market event analysis

Probability estimates from market event data

Store price and signal events, compute calibrated probability views, and query those results through dashboards, APIs, or an AI client.

Get Early AccessView Capabilities
11.9μs
Core indexed read p99
469K/s
Batch ingest reference
32
Default partitions
3
Analytics endpoints
Built for Performance & Scale

Architecture

Client Layer

  • Web App (React)
  • Python SDK
  • AI Query Engine

API Layer

  • REST API
  • WebSocket Streaming
  • Analytics Endpoints

Processing Layer

  • Rust Core (469K/s)
  • Elixir Query Service
  • Go Control Plane

Storage Layer

  • Parquet Columnar
  • WAL Durability
  • Point-in-Time Snapshots
What Core provides, and where your analysis begins

Technical capabilities

Where does the data live?

32 partitions

Accepted events are recoverable from a CRC32-checked write-ahead log and Snappy-compressed Parquet files. Entity IDs map to 32 partitions by default; the partition count is configurable.

  • By Symbol: entity_id = 'NQ' or 'BTC'
  • By Session: event_type tags for RTH/ETH
  • By Date: Microsecond timestamp precision

How fast is time slicing?

11.9μs p99

The published Core reference benchmark measured 11.9μs p99 indexed reads. Bar generation and range slicing scan or aggregate more data, so benchmark them with your event shape and hardware.

  • Indexed-read reference: 11.9μs p99
  • Batch-ingestion reference: 469K events/sec
  • No published end-to-end bar benchmark

How are corrections handled?

Append-only corrections

Immutable append-only event sourcing with correction events. Full audit trails with as_of temporal queries.

  • Original events are never modified
  • Corrections appended with references
  • as_of queries show pre-correction state

Can past analysis be reproduced?

Replayable inputs

AllSource preserves the ordered inputs needed for replay and point-in-time reconstruction. Reproducible analysis also requires deterministic projection code, versioned models, and captured configuration.

  • Snapshots: Every 100 events or 1 hour
  • as_of queries: Point-in-time state
  • Event Replay: Full deterministic replay

How easy is Python integration?

Multi-platform

Use the HTTP API from any Python client or the Python SDK in the AllSource repository. JSON responses can be loaded into pandas; WebSocket streaming supports live updates.

  • REST API: GET /api/v1/events/query
  • WebSocket: WS /api/v1/events/stream
  • Python client source in sdks/python-client

Does it support concurrent users?

Measure your route

Core uses a sharded concurrent map for hot reads. Capacity depends on filters, payload size, API hops, cache state, and hardware; load-test your end-to-end route before setting an SLO.

  • Concurrent in-memory read path
  • Prometheus latency and throughput metrics
  • No universal hosted QPS claim
Current endpoints in Core

Available analytics API

api-reference.ts
GET/api/v1/analytics/frequency// Event frequency bucketed by time window
GET/api/v1/analytics/summary// Statistical summary of events
GET/api/v1/analytics/correlation// Event correlation analysis
AllSource stores evidence; your model produces probabilities

Product boundary

Available now

Event ingestion, time-range queries, replay, snapshots, WebSocket streams, and three analytics endpoints.

You provide

Market-data licensing, bar construction, feature definitions, probability models, calibration, and validation.

Not claimed here

No built-in trading strategy, return guarantee, or shipped natural-language quant endpoint is implied.

Build on durable market events

Use AllSource for ordered history, replay, and current analytics endpoints. Keep your probability model explicit and independently validated.

Start 14-day trialView Pricing