generated: '2026-08-28'
method: searched
source: https://www.speedata.io/technology
note: >-
  Speedata publishes no API contract, so none of the API-layer standards (OAuth 2.0, OIDC,
  RFC 9457, JSON:API, OData, SCIM, pagination, idempotency) can be asserted — there is no
  spec location to point evidence at, and the pipeline does not assert conformance from
  marketing prose. What Speedata does conform to is a DATA-ENGINE standard set: it is a
  drop-in accelerator, so its whole value proposition is fidelity to interfaces someone
  else defined.
standards:
- id: apache-spark-plugin-api
  name: Apache Spark plugin interface (spark.plugins)
  conforms: true
  evidence: >-
    The documented integration is a stock Spark plugin registration —
    `--conf spark.plugins=com.speedata.spark.DashPlugin` with
    `--jars speedata-dash-0.8.1-spark_2.12.jar` — which is the Spark 3.x plugin API, not a
    proprietary launcher. Source: https://www.speedata.io/product
- id: apache-spark-sql
  name: Apache Spark SQL / DataFrame semantics
  conforms: true
  evidence: >-
    Dash "transparently plugs into the Spark Catalyst optimizer to automatically identify
    and offload compute-intensive work to the APU" with "zero code changes required" —
    i.e. the accelerator must preserve Catalyst plan semantics exactly. Supported on
    Spark 3.x under Kubernetes, YARN and standalone. Source:
    https://www.speedata.io/technology
- id: apache-parquet
  name: Apache Parquet columnar format
  conforms: true
  evidence: >-
    The APU accelerates the full Parquet pipeline in silicon — decompression, decoding,
    columnar filtering, projections, row assembly and nested-data flattening (EXPLODE) —
    which requires bit-level conformance to the Parquet specification. Source:
    https://www.speedata.io/technology
- id: tpc-ds
  name: TPC-DS decision-support benchmark
  conforms: true
  evidence: >-
    The Workload Analyzer offers a TPC-DS comparison mode
    (https://www.speedata.io/workload-analyzer), and the company publishes a first-party
    MIT-licensed generator at https://github.com/Speedata-io/speedata-tpc-generator.
- id: substrait
  name: Substrait cross-platform query plan representation
  conforms: unknown
  evidence: >-
    github.com/Speedata-io maintains forks of `substrait` and `substrait-java` (both last
    pushed 2025-09-18), which is a strong signal the plan-offload path speaks Substrait —
    but no Speedata document states it and the forks carry no Speedata-authored change we
    could cite, so this is recorded as unknown rather than asserted.
- id: pcie-gen5
  name: PCI Express Gen 5 x16
  conforms: true
  evidence: >-
    "C200 PCIe Gen5 x16 interface, server-agnostic" — the card is a standard-form
    accelerator rather than a bespoke appliance. Source: https://www.speedata.io/product
not_applicable:
- id: oauth2
  reason: No API and no authorization surface published.
- id: rfc9457
  reason: No API and no error envelope published.
- id: openapi
  reason: No machine-readable API contract published anywhere on the domain.
- id: fhir / fapi / psd2 / scim / odata
  reason: >-
    Semiconductor and data-infrastructure vendor; none of these regulated-sector standards
    applies to its market. Reward-only check, so no penalty and nothing invented to fill it.
compliance:
  certifications: []
  trust_center: null
  note: >-
    No SOC 2, ISO 27001, PCI, HIPAA or FedRAMP claim found, no trust center
    (trust.speedata.io does not resolve), and probe-security-programs.py returned
    vdp=none trust=none. No Compliance or TrustCenter pointer is emitted.
evidence:
- url: https://www.speedata.io/product
  status: 200
- url: https://www.speedata.io/technology
  status: 200
- url: https://www.speedata.io/workload-analyzer
  status: 200
- url: https://github.com/Speedata-io
  status: 200