Sidiora runs Paxeer’s internal price fabric as an aggregation engine rather than a single-feed oracle. Dozens of off-chain venues, brokers, and liquidity pools stream tick-level data into a vendor-partitioned storage layer, where our aggregation algorithms derive a canonical mid-price surface that powers risk, execution, and on-chain oracles.
The walkthrough below shows how a single symbol flows through the pipeline: from 50+ off-chain vendors, into ScyllaDB by vendor, through symbol-level aggregation and mid computation, into Postgres as the canonical surface, and finally out to 25+ on-chain oracles and more than 100 external APIs.
Step 1 · Off-chain ingress
More than 50 off-chain vendors — CEX, DEX, RFQ books, OTC streams — are continuously ingested as raw trades, orderbooks, and quotes. Sidiora normalizes symbol schemas, venues, and timestamps into a single ingress plane before any aggregation is applied.
Step 2 · ScyllaDB by vendor
Ingested ticks land in ScyllaDB, partitioned by vendor and symbol. This keeps hot history, depth, and microstructure statistics co-located with the originating venue, enabling low-latency replays and venue-specific health checks without corrupting the canonical surface.
Step 3 · Symbol aggregation & mid
A dedicated aggregation service pulls a symbol’s price surface from every vendor slice in ScyllaDB, applies venue-level filtering and weighting, and computes a stable mid across all eligible prices. Outliers are quarantined; only clean consensus prices move forward.
Step 4 · Canonical surface in Postgres
The resulting mid series is persisted to Postgres as Paxeer’s canonical price surface. Risk systems, execution engines, and funded wallets interact with this deterministic layer instead of with raw vendor feeds, simplifying observability and audits.
Step 5 · Oracles & APIs
From Postgres, dedicated fan-out workers stream the canonical surface into 25+ on-chain oracles and over 100 public and partner-facing APIs. Paxeer, agents, and external builders consume the same battle-tested feed, without having to maintain their own vendor mesh or aggregation stack.