Quantitative research / Study frameworks

Methods first.
Findings when supported.

A research agenda for reproducible scientific AI and rigorous financial forecasting. Each study defines its question, comparison, evaluation approach, and limitations.

No measured findings published · Framework stage

Real Estate Forecasting / Research field

Follow the field. Inspect the forecast.

Explore 64 locations, proposed opening-round matchups, and a ten-measure comparison of historical home values, next-month forecasts, and model reliability. Datasets and measured findings remain pending.

Regional standingsFour regions. Sixteen seeds each. Supplied seeding with city search and regional filters.View standings ›Matchup scheduleThirty-two proposed opening matchups. Dates and pairing confirmation remain pending.View schedule ›Forecast box scoreCambridge vs Ithaca. Ten measures covering history, forecast uncertainty, and backtest error.Inspect the box score ›

Scientific AI

Reproducibility across repeated executions

Awaiting dataset and execution

Can a versioned workflow reproduce outputs under controlled execution conditions?

Dataset and sample: Not supplied. Source, license, sample size, inclusion criteria, and time range must be documented before execution.
Comparison: A documented notebook workflow using the same inputs and method.
Proposed protocol: Repeated runs across fixed seeds and declared environments; compare outputs within prespecified numerical tolerances.
Planned measures: Reproduction rate; output variance; artifact completeness; execution failure rate.

Limitations and interpretation

Environment differences, stochastic algorithms, and incomplete provenance may affect comparisons.

Results pending.

No effect size, confidence interval, or performance claim is available. Code release and data access are pending.

Financial forecasting

Forecast value beyond a naive baseline

Awaiting dataset and execution

Does the proposed model improve out-of-sample forecasts over a prespecified baseline?

Dataset and sample: Not supplied. Source, license, sample size, inclusion criteria, and time range must be documented before execution.
Comparison: Last-observation and seasonal naive forecasts, where appropriate.
Proposed protocol: Chronological training and rolling evaluation; isolate tuning from final testing and document data availability at each prediction time.
Planned measures: MAE; RMSE; directional accuracy; interval coverage; performance by evaluation period.

Limitations and interpretation

Regime shifts, limited samples, leakage, and transaction costs can change interpretation. Forecast accuracy does not establish investment returns.