A live experiment in which StockPulse stores a morning KOSPI forecast, evaluates it after the market close, analyzes failure patterns, and applies changes to prompts, ML parameters, and data features for the next run.
A continuous experiment: prediction → evaluation → failure analysis → applied improvement.
Current finding
The loop is operating in production: forecast, evaluate, diagnose, apply a change, and run the next forecast with the changed environment.
Prediction and evaluation
Read-time evaluation from the existing predictions table. Accuracy is not copied into translated post content.
Latest forecast
보합
2026-08-27
Latest score
—
Actual: Not recorded
All evaluated runs
38%
10/26 correct
Translated Lab Notes in this pilot
The pilot includes one successful and one failed run so the English page preserves both outcomes rather than showing only a polished result.
Pilot boundary
StockPulse morning and market-close Feed posts are intentionally not translated in this pilot. The Project and selected Lab Notes are translated so the experiment can compare original-value content without turning a daily Feed into a translation maintenance queue.