Proven on our own system before it reaches yours.
Research governance, AI-agent operating frameworks and disclosure-based company analysis are available now. Two more lines are in preparation.
Three available now. Two in preparation.
Research governance design and build
We design pre-registration, point-in-time validation, self-scoring ledgers and rejected-result disclosure rules, then port them into code and documents.
- Who it is for
- Investment institutions, fintech companies, corporate research teams
- What you receive
- Governance rulebook, ledger schema, validation scripts, operating checklist
- Basis
- 563 pre-registrations and an append-only ledger, run on our own system since July 2026.
AI-agent operating framework
We design and build a framework that separates build and review roles, adds cross-review by a different model, pre-execution guards, budget and contract gates, and a work ledger.
- Who it is for
- Any development organization. It does not have to be in finance.
- What you receive
- Agent role definitions, guard rules, review procedure, ledger and dashboard
- Basis
- 227 research modules and 474 test files were built under this framework. A model cannot ship without a declared tuning budget.
Disclosure-based company analysis (B2B)
For business purposes such as competitor, counterparty, acquisition, inventory and cash-flow reviews, we extract and compare original regulatory filings and turn them into reports.
- Who it is for
- Corporate clients only. No buy or sell judgment on any security.
- What you receive
- Comparison tables, one-page briefs, scheduled updates
- Basis
- We extract down to footnote level (inventory detail, valuation allowances) from original XBRL filings on DART, Korea's electronic disclosure system.
Volatility interval data and reports
We are preparing 80% interval data and reports at 30- and 60-minute horizons as a direction-free risk measure.
- Who it is for
- Institutions
- What you receive
- To be finalized after legal and data-terms review.
- Basis
- Ten models racing live, 27 Korean symbols and 2.95M observations scored (as of 2026-09-23), working-paper draft in hand.
Market surveillance and compliance analytics
We are preparing analytics that detect abnormal moves and interval breaches.
- Who it is for
- Exchanges, issuers, institutions
- What you receive
- To be finalized after legal and data-terms review.
- Basis
- Built on the volatility layer's interval forecasts and the guards and ledgers of the agent framework.
Evidence instead of adjectives.
Models registered before scoring.
Rejected hypotheses stay beside the ones that passed.
The model that built the code never approves it.
A model cannot ship without a declared tuning budget.
Four steps from diagnosis to operation
- 01
Diagnose
We look at how research and development currently flow and are recorded, and find what is used without validation.
- 02
Design
We set pre-registration, validation, ledger and review rules in writing, fitted to the organization.
- 03
Build and port
We move the rules into code and forms, attach them to existing tools and run them.
- 04
Operate and check
We operate alongside you for a period, check against the checklist, and report bad results as they are.
The boundary comes first.
- ×We do not act as an investment adviser and do not recommend buying or selling any financial instrument.
- ×We take no discretionary mandates and accept no authority over client assets.
- ×We hold no client money. There is no trading desk and no order or account API.
- ×We do not sell paid stock information to individuals.
Frequently asked questions
No. Research governance and the AI-agent framework apply to any organization where unvalidated results reach decisions: development teams, data teams, corporate research groups. Disclosure-based analysis serves corporate clients in any industry.
Yes. We apply point-in-time validation and pre-registration rules to your data and record what passes and what is rejected in a ledger. Data stays inside your environment, and the ledger is never edited once written.
We report them as they are. Not hiding bad results is part of the service. We applied this framework to our own system before offering it to anyone, and the record includes failures. Rejected results stay in the ledger and go into the report.
Discuss a project or request materials.
Inquiries: weekdays 10:00–18:00 KST