data-routing
Data Routing (Router)
The runtime data gateway owns acquisition, source health, queues and cache. This skill maps a research need to the existing gateway-backed tool; the model must not recreate a separate provider-selection or retry loop.
Use get_market_data, get_options_chain, and the research tools below first.
get_market_data with a task workspace exports the same returned rows to CSV
in Python by default (export_csv=false disables the sidecar) with source/currency/adjustment metadata and sampling flags. Its
_csv_file receipt provides the path, columns, row count and checksum. Sampling
still applies; this is a task artifact, not another acquisition or a full-history
download. _data_layout describes the saved JSON structure for custom calculations.
For Python bars use src.market_data.fetch_market_data; for additional vendor
fields use src.data_gateway.skill_sources or a managed_reader that creates,
reads and materializes its SDK result inside the bounded callback. Do not return
lazy SDK properties, iterators or network clients for later unbounded reads.
Credential-scoped data uses uncached acquisition and keeps existing authorization.
Large history requests use history_data; do not download whole years into an
unbounded script cache. These adapters share the existing resource limits.