How it works
1
Point-in-time investment universe
Each sector is defined as a curated dashboard of vendors, grouped by their
role in the supply chain. The universe is point-in-time: what the agent
knows about a company is anchored to a specific filing vintage, so theses
stay reproducible and auditable.
2
Ground truth from primary filings
Every vendor is routed to the ground-truth source for its jurisdiction.
US issuers come from SEC EDGAR (XBRL financials plus 10-K/10-Q links);
Korea, Taiwan, and Japan are currently served from consolidated Yahoo
Finance financials, with DART, TWSE/MOPS, and EDINET as the primary-filing
upgrade path. The model never invents numbers — figures come from the
ground-truth layer.
3
Structured company reasoning
An LLM produces a structured assessment of each company — where the
business stands and its near-term outlook — from that company’s
ground-truth financials, upcoming earnings, and recent news. Output is
schema-validated (Pydantic
json_schema structured output), and every call
is traced end-to-end with W&B Weave for observability.4
Dashboard payload
Ground truth, theses, bias signals, and option strategies ship as one
dashboard payload per sector — the same payload the web dashboards render
and the REST API serves.
What you can do with it
- Single-name deep-dives — financials, thesis, bias, and option strategies for any vendor in the universe.
- Earnings workflows — upcoming earnings events are part of the reasoning context, so theses reflect what is actually about to be reported.
- Sector research — compare vendors by supply-chain role within a sector dashboard instead of researching names in isolation.
Call the API
All research endpoints — market data, dashboards, reasoning refresh — with
verified request/response examples.