generated: '2026-09-05'
method: searched
source: >-
  https://app-static-7l.storage.googleapis.com/latest/7Learnings_initial_data_request.pdf,
  https://7learnings.com/docs/output-options/, https://7learnings.com/docs/pipeline-subscriptions/,
  https://7learnings.com/changelog/, https://7learnings.com/.well-known/oauth-authorization-server
docs: https://7learnings.com/docs/
scope: >-
  These are DATA-EXCHANGE conventions, not HTTP API conventions. 7Learnings ships no public
  REST API: integration is batch data transport in (warehouse, object store, SFTP/FTPS/HTTPS,
  or vendor connector) and optimized price/marketing changes out (table, file, or an API call
  into the customer's own commerce platform). The one HTTP protocol surface the company serves
  publicly is the OAuth-gated MCP endpoint on 7learnings.com, whose conventions are OAuth 2.0's.

authentication:
  style: >-
    No public API key. OAuth 2.0 authorization-code + PKCE (S256) on the MCP endpoint; SSO, MFA
    and passwordless login on the Steering App; per-customer transport credentials on every
    data connector.
  detail: see authentication/7learnings-authentication.yml

transport:
  input_formats: [database table, CSV, Parquet]
  input_targets: [BigQuery, Snowflake, Microsoft Fabric, Google Cloud Storage, AWS S3, Azure Blob Storage, SFTP, FTPS, HTTPS]
  input_connectors: [Tradebyte, Shopify, Google Shopping, Google Ads, Google Analytics, Amazon Selling Partner API, Amazon Ads]
  output_formats: [database table, CSV, Parquet, Salesforce Commerce Cloud Pricebook, vendor API call]
  output_targets: [BigQuery, Snowflake, user computer, Google Cloud Storage, AWS S3, Azure Blob Storage, SFTP server, HTTPS server, Salesforce Commerce Cloud, commercetools, Tradebyte, Plentymarkets, Shopify, Scayle Commerce Engine]
  csv_dialect:
    rfc4180: true
    quoting: RFC 4180 compliant
    timestamps: UTC
    booleans: lowercase
    since: '2026-08-11'
    source: https://7learnings.com/changelog/
  customization_note: >-
    "We always customize the data export format for each customer, so that you can decide which
    KPIs should be in these export files." The export schema is therefore per-tenant, not a
    published fixed contract.

pagination:
  applicability: na
  detail: No public HTTP collection endpoints exist; data moves as whole tables and files.

field_selection:
  applicability: na
  detail: >-
    There is no sparse-fieldset or expand parameter. The nearest equivalent is the per-customer
    choice of which KPIs appear in an export file, and Upload Filters in the Steering App.

metadata:
  applicability: na

request_id_tracing:
  applicability: na
  detail: >-
    No request-id header is published. The traceable unit is a pipeline RUN, surfaced in the
    Steering App's Pipeline Status view (step names and progress timeline, reworked 2026-08-31)
    and in Upload Evaluation.

versioning:
  scheme: none-published
  detail: see lifecycle/7learnings-lifecycle.yml

error_envelope:
  applicability: partial
  http_surface: >-
    The MCP endpoint returns a JSON error object shaped
    {"code": "...", "message": "...", "data": {"status": <int>}} - the WordPress REST error
    envelope, observed as {"code":"mcp_unauthorized","message":"MCP authentication required.",
    "data":{"status":401}}. This is not RFC 9457 problem+json.
  batch_surface: >-
    Data-pipeline failures are not returned to a caller; they are notified by EMAIL through
    Pipeline Subscriptions ("Forecast Failure - Data Issue") and shown in Pipeline Status.
  rfc9457: false

rate_limit_signaling:
  applicability: na
  detail: No rate-limit headers or documented limits. see rate-limits/7learnings-rate-limits.yml

idempotency:
  applicability: na
  coverage: none
  supported: false
  mechanisms: []
  detail: >-
    No replay-protection mechanism is documented, because there is no public mutating HTTP
    surface to protect: there is no Idempotency-Key header, no idempotency key field, and no
    documented at-most-once semantics. Batch transport is snapshot-shaped instead - the
    provider states its system "either handles daily updated snapshots of the full history, or
    receives the full history once and then daily snapshots", which makes re-delivery of the
    same snapshot naturally convergent rather than idempotency-keyed. That is a property of
    the design, not a published guarantee, so `coverage` is recorded as none and
    `applicability` as na rather than claiming a mechanism the provider does not document.
  source: https://app-static-7l.storage.googleapis.com/latest/7Learnings_initial_data_request.pdf

dry_run_mode:
  applicability: na
  supported: false
  detail: >-
    No dry-run flag on any public interface. The product-level analogue is human-in-the-loop:
    optimizations are reviewed in the Steering App, Optimization Comparison lets a user compare
    candidate optimizations before committing, and an upload can be configured with reviewers
    before the "Create Upload" step pushes changes out. None of this is exposed as a machine
    interface.
  source: https://7learnings.com/docs/output-options/

reversibility:
  applicability: na
  grade: na
  detail: >-
    There is no public write API, so there is no reversal operation to record. NOTE FOR ANY
    FUTURE PASS - the consequential write here is real but it happens in the CUSTOMER's system:
    7Learnings pushes price changes into commercetools, Tradebyte, Plentymarkets, Shopify,
    Scayle or a Salesforce Commerce Cloud pricebook. Whether such a push can be reversed, and
    inside what window, is governed by the receiving platform, and 7Learnings publishes no
    rollback operation, no undo endpoint and no stated reversal window of its own. No window is
    asserted here because none is stated anywhere in the public documentation.
  write_surfaces:
  - surface: Scheduled / manual price+marketing upload to the customer's commerce platform
    reversal_operation: null
    window: null
    docs: https://7learnings.com/docs/output-options/
  - surface: Optimization records held in the Steering App
    reversal_operation: null
    window: null
    note: >-
      Not a reversal - a retention rule. The 2026-08-13 changelog states unused optimizations
      are removed after one year.
    docs: https://7learnings.com/changelog/

cross_links:
  errors: null
  lifecycle: lifecycle/7learnings-lifecycle.yml
  authentication: authentication/7learnings-authentication.yml
  rate_limits: rate-limits/7learnings-rate-limits.yml
  data_model: data-model/7learnings-data-model.yml