THE OPEN RESEARCH TERMINAL
Explore the possibilities. Find the disagreement. Keep the evidence.
02 / EXPLORE THE DISTRIBUTION
FOUR WAYS TO READ THE SAME EVENT
Each row shows DOWN / FLAT / UP. A different probability is not evidence of a better model.
MAKE THE THESIS FALSIFIABLE
PRESERVE THE DISAGREEMENT
Select two to four runs on the same frozen history. Differences in evidence and assumptions stay visible.
IDENTICAL EVENT. DIFFERENT MODELS.
Evaluate the price-only quant, historical bootstrap, EWMA and Student-t baselines on this run's frozen inputs.
Forecasts, abstentions and failures remain in the local experiment registry. Scores need real matured outcomes.
SOURCES → CLAIMS → OBJECTIONS
Project evidence edits do not rewrite frozen runs. Publication and availability are asserted inputs; use the CLI for separate timestamps and document anchors.
WHAT COULD YOU KNOW THEN?
Configure SEC contact and ALFRED credentials in Data hub or the local server environment. Sources are fetched only when requested.
READ THE CHANGE, NOT JUST THE DOCUMENT
Structured inputs for portfolio analysis, agent memory and evaluation. These tools operate on supplied research data.
YOUR MODELS. ONE FROZEN QUESTION.
Route the roles, review the context, then watch the reasoning unfold.
FOLLOW THE CONNECTIONS
Recorded activity and explicit references. Select a node to inspect the receipt behind it.
No nodes match these filters.
Drag the background to pan. Use + / − to zoom. Tab and Enter select nodes; the list offers the same details. Lines indicate recorded relationships, not causality.
Frozen inputs are available before any model calls.
Replay uses saved event checkpoints. Stream text is the last saved snapshot for each call; token-by-token history and private model reasoning are unavailable.
The cost preview uses your supplied prices and a conservative input reservation. It does not guarantee a provider invoice.
Sources and outputs remain text. Models can request only five read-only tools on this frozen context. No orders.
Connect a model and preview the loaded run to begin.
WHERE THE REASONING DIVERGES
Validated forecasts and model-supplied objections appear here after a session.
COMMIT BEFORE THE OUTCOME
Local commitments preserve validated predictions. Supply the actual trading-calendar settlement close. External publication and the outcome's authenticity require separate verification.
BRING YOUR OWN MODEL
Use a compatible cloud endpoint or a model running on your own computer.
AVAILABLE CONNECTIONS
No connections configured.
Tradefloor does not inherit an AI host's login or subscription. API billing and local model requirements belong to your chosen provider. A connection test sends a small fixed prompt, without research data.
KNOW WHERE THE INPUTS CAME FROM
Import daily history, preserve its provenance, and open a scenario on the supplied data.
Makes a provider request. Refreshed adjusted history is captured now and does not establish historical availability. SEC and ALFRED source metadata are available in Information timeline.
The built-in reaction model remains assumption driven. No synthetic settlement is supplied for imported history.
Configure SEC submissions and ALFRED vintages here, then fetch a selected source from Information timeline.
READ THE SOURCE. KEEP THE RECEIPT.
Capture a document, find the passage, and review its availability before freezing it into research.
Uses the SEC contact configured in Data hub. Discovery does not download filing contents.
Uses the title, rights and dates above. Public HTTPS only; no credentials, query strings, redirects or private network requests. Makes an explicit source request.
LOCAL, SEARCHABLE, IMMUTABLE
Select exact source passages to preview them here.
Only dated sources available by the loaded scenario cutoff are eligible. This creates a new receipt on unchanged prices and numerical assumptions. Models receive the sources only after a separate context preview and Start.
TAKE THE INVESTIGATION WITH YOU
Preserve the question, assumptions, forecasts, audit flags, unresolved observations and exact source appendix.
No model calls. Reports contain the selected research and source excerpts; review their redistribution rights before sharing. Synthetic settlement and future paths are omitted.