Integrations

Every connector is read only by construction: it can pull, it cannot write. Credentials are encrypted at rest with a key that never leaves the service. Nothing below requires the engine to connect to a pricing system.

Ways to get data in

Option	What it is	Best for
File upload	CSV, Parquet or JSON lines. A column mapper proposes which of your columns is which and saves the mapping for every later pull.	First engagement, counsel with an export in hand
Multi-file engagement	A price log, an override log, a feature snapshot and a market feed, each from its own system, joined on SKU and nearest timestamp. The report records what joined and what did not.	Litigation-readiness packs
Signed drop URL	A one-shot, write-only PUT URL for one file. Your export job pushes; no credentials of ours are ever on your side, no browser upload limit.	Recurring exports, large files
In-VPC runner	The engine as a container inside your network. It reads your data where it lives and sends us only the report hash, the chain entry and the verdict. Raw data never leaves your perimeter.	Any company whose counsel objects to raw logs leaving

Connectors

Kind	Systems	Feeds
Warehouse view	Snowflake, Google BigQuery, Databricks SQL, Amazon Redshift, PostgreSQL, SQLite	Everything in Mode A from one read-only view; incremental pulls by timestamp
Object storage	Amazon S3 and S3-compatible, Google Cloud Storage, Azure Blob, a local drop directory	A manifest names the files and their roles; multi-file engagements on a schedule
Event stream	Apache Kafka, Amazon Kinesis, Google Pub/Sub, a JSON lines file	Continuous certification: a rolling window re-audited on a cadence, certificate re-issued
Lineage	OpenLineage and Marquez, Databricks Unity Catalog, Feast, Tecton, a CSV edge list	Feature names resolved to their upstream datasets, classified public, own, shared or cross-tenant against your source registry; the commingling module then works on facts instead of names
Approvals	ServiceNow, Jira, Slack approval channels, CSV	Recommendation-to-decision records with latency, so the delegation index knows who looked and how long they took
Market feeds	HTTP and SFTP CSV feeds from rate shoppers and price intelligence providers, Amazon Selling Partner API offers	Mode B competitor paths without touching a competitor, each row tagged with its public source

The canonical record

ts               ISO 8601 or epoch (unit configurable), UTC
sku              product, unit type, room type, rate plan, grade
seller           opaque label; your mapping to real identities stays with you
price            posted price; cents supported via a divisor
proposed_price   what the algorithm recommended            (Mode A)
override_status  accepted | rejected | modified              (Mode A)
decided_ts       when a person decided                       (Mode A)
features_json    the input vector at decision time           (Mode A)
source           which feed a market row came from           (Mode B)

What leaves your environment with the in-VPC runner

engagement id, verdict kind, report hash, chain entry hash, previous hash,
overall risk index, method version, flag codes

Not a row, not a seller label, not a feature name. The hashes let the certificate be verified against the report you keep.

Connect through the console with your API key, or through the contact form to start an engagement.