predome.Workspace · sign-in
AI FOR SCIENCEResearch prototype

Forecast the outcome.
Keep the record.

An experiment in scientific forecasting. Compare how AI models judge a clinical question before the answer is known — then test those judgments against the result.

Explore the questions How it works
INITIAL RESEARCH FOCUS

Gastric & colorectal cancer

Model judgments are research hypotheses, not treatment advice.
Conceptual illustration of a cellular microenvironment
Concept illustration · not microscopy
PREDOME / RECORD FORMAT

Will the study meet its predefined primary endpoint?

Evidence cutoffSet before submission
Model judgmentsIndependent first round
Official resultAwait disclosure

Format preview. No prediction is submitted by this illustration.

The question desk

A forecast starts with
a question worth resolving.

Explore published trial-registry references. Review the question, endpoint and decision rule before a model makes its first prediction. These are question drafts; no predictions are submitted here.

ClinicalTrials.gov / published public referencesPublished snapshot · not live

Collected . This static website does not collect data while you browse. Updates require a new collection and publication.

NCT07822399Registry reference

A Study of MGC030 in Participants With Advanced Solid Tumors

Urothelial Cancer · GastroEsophageal Cancer · Colorectal Cancer · Esophageal Cancer · Gastric Cancer (GC) · Pancreatic Cancer

Phase 1Registry status at collection: recruitingSource update: 2026-10-30
NCT07715747Registry reference

A Study to Evaluate the Safety,Pharmacokinetics,Pharmacodynamics,and Efficacy of TJ037130 in Adult Participants With Advanced Solid Tumors

Solid Tumor · Metastatic Ovary Cancer · Metastatic Gastric Cancer · Metastatic Endometrial Carcinoma

Phase 1Registry status at collection: recruitingSource update: 2026-10-30
NCT07780305Registry reference

ABCSG R07 Rectal Cancer Registry

Rectal Cancer · Adenocarcinoma of Rectum

Phase not suppliedRegistry status at collection: recruitingSource update: 2026-10-30

A sample of 5 registry records, not a complete trial list. Source-update dates and collection time are different. Registry metadata is not a treatment result.

Explore all published references Download mapped snapshot Download official response

THE FORECAST PROTOCOL

A judgment becomes a record.
A result becomes a test.

The intended prospective workflow. Current research tooling and the forecast protocol are identified separately.

  1. Define the question

    One trial, one specified endpoint and a rule for what will count as an answer.

    Protocol

  2. Date the evidence

    Keep a source-linked snapshot of what was available before the result.

    Research records available

  3. Submit independently

    Models make their first judgment without seeing the others. Abstention stays visible.

    Planned forecast workflow

  4. Preserve the forecast

    Retain the original submission; a revision becomes a separate version.

    Planned forecast workflow

  5. Resolve against the source

    Use the predefined rule and official result. Unresolved questions keep waiting.

    Planned adjudication

  6. Evaluate over time

    Compare probability scores over prospective outcomes, including the misses.

    Scoring explainer available
Read the evaluation rules
ROLES WITH A PURPOSE

Every Agent has
a question to answer.

A division of research work, with visible inputs, deliverables and limits.

EVIDENCE AGENT / PROTOCOL ROLE

Establish what is known.

Receives
Official registries, published research and dated source material.
Produces
A source-linked evidence packet, with missing and conflicting information kept visible.

Public-source collection is available in the authorized workspace. Literature review and forecast-specific evidence selection require further integration.

Inspect research runs · sign-in
INTERACTIVE SCORING SIMULATION

Confidence needs
a consequence.

Try a binary Brier score. Choose two probabilities, reveal an illustrative outcome and see how the score changes.

Illustrative inputs · not model forecasts

Illustrative observed outcome
BINARY BRIER LOSSLower is better
Forecast A0.1225
Forecast B0.3600

LOSS = (PROBABILITY − OUTCOME)²Forecast A receives the lower loss for this outcome.

Scores are displayed to four decimal places. A single result illustrates the scoring rule. Predictive ability requires many resolved, prospective questions; unresolved cases do not count as failures.

INSIDE THE RESEARCH WORKSPACE

Work you can
inspect.

The separate research workspace retains source data, collection tasks and attached files. Access requires authorization; uploads and analysis remain there.

Open workspace · sign-in

Published reference data and computational work. This platform does not operate a wet laboratory.

Published GSE90944 RNA-seq data (2017) · recorded data visualization, not experiment footage.
THE STANDARD WE ARE BUILDING TOWARD

The misses stay
on the record, too.

The goal is a transparent prospective record: original judgments, independent review and outcomes that can be checked. Dated public registry references can be inspected here, with research tooling in the authorized workspace; a validated forecasting track record has not yet been established.

Read the protocol and current boundaries