Does information from other markets improve what we can predict about the next move?
Compare an instrument’s own history with earlier information from connected markets. Keep the target and evaluation dates identical.
Research brief / EXP-001
Explore. Question. Test.
Our first direction: relationships across markets.
Sources inventoried—Capabilities, not live feeds
Research-ready sources—History and rights still needed
Hypotheses defined04All remain untested
Completed experiments00No performance claims yet
Next decisions
Make the question runnable.
01
Choose the instruments
A bounded universe with reliable historical coverage.
02
Set the decision clock
When we observe, when we trade, and how far we predict.
03
Establish the data history
Availability times, revisions, rights and trading costs.
The world model begins with traceable observations. A geographic connection or persuasive explanation becomes evidence only after a reproducible test.
Operational context
Explore sites and layers.
Use ORB’s existing spatial tools alongside the research workflow.
Client and project
This scope controls operational queries. It does not select the market experiment’s instrument universe.
Map and field context
Location and voice activate only when requested. Operational records are not automatically sent to AI.
Spatial query
Claude or RAiNBot can generate globe views using the selected project. Generated output requires review before use as evidence.
Choose Claude or RAiNBot in Connections to use spatial generation. Ask AI remains the research conversation.
Relationships and prediction
The earlier Graph and Predict specifications describe intended capabilities. A validated relationship engine and risk model are not connected here yet.
Question register
What are we testing?
Four directions. One first experiment. Every claim remains open to rejection.
Markets inform other markets.
Earlier cross-market information adds predictive value beyond an instrument’s own history.
What would challenge it?
The improvement disappears on unseen dates or with achievable timing and costs.
Context changes the relationship.
Market conditions identified beforehand help explain when a predictor works.
What would challenge it?
The effect requires hindsight to label conditions or vanishes when periods change.
Physical events leave market traces.
Geographic events add information for instruments with documented exposure.
What would challenge it?
Correct publication times or exposure controls remove the apparent advantage.
Different errors can work together.
Combining modest predictors improves forecasts when their errors differ.
What would challenge it?
The gain comes from selecting lucky models, shared risk, or excessive turnover.
EXP-001 / HYP-001
Design the first comparison.
Capture choices here before they become assumptions in code.
Experiment brief
Saved in this browser only. Export the draft to retain it outside this browser; it does not update the repository specification.
Three models. Same test.
- Historical mean baseline
- Regularized model using own history
- The same model with earlier cross-market information
Time-based evaluation, training-only transformations and a complete trial ledger. Trading costs are evaluated separately.
No results yet
A frozen dataset and evaluation dates are required before a model can run.
Ideation / AI + you
Keep the reasoning.
Observations, questions, objections, and the next thing to test.
These notes belong to the current draft. Save the investigation to keep them after reload or close, or export Markdown.
Configured research assistant
Think with AI.
AI assistance · verify claims against evidence. This is a separate conversation with our project brief.
Only this chat and the curated project brief are sent. Your notebook and operational records are not attached. Answers can update a validated map layer or propose tables in interactive mode. They do not execute arbitrary code or run experiments.