What This Exercise Is About
Process over predictionYou are technology students, not finance students. You do not need to become traders. You need to become disciplined thinkers who can gather evidence, reason under uncertainty, compare AI outputs, and explain a conclusion clearly.
Set the prediction
Use last week's actuals, current market data, three agents, and four AI models to form a locked weekly prediction.
Record actuals
Record the final weekly numbers. Do not edit your locked prediction. The GitHub timestamp is evidence.
Score and learn
Compare prediction against actuals, analyse calibration, present mistakes, and improve the workflow.
Your Week 3 Mission — Sprint Active Now
2–6 June 2026Current sprint instruction: W2 is finished and must now be scored. Your immediate task is to use the W2 actuals, update your calibration log, run your W3 evidence workflow, and submit a locked W3 prediction before Sunday 7 June 2026, midnight SGT.
Score W2 Calibration
- Compare your W2 prediction against W2 actuals.
- Record whether each direction call was correct, wrong, or uncertain.
- Calculate calibration quality: did your confidence match the outcome?
- Identify which AI model was closest and which model was most overconfident.
- Commit your calibration update to GitHub as evidence.
Build the W3 Evidence Base
- Open Finviz Futures Performance and set the view to 1W.
- Open Yahoo Finance Sectors and set the view to 5D.
- Open TradingEconomics Calendar and identify the week-ahead events, especially the Friday NFP/jobs data.
- Write three separate agent outputs first: Almanac, Macro/News, and Technical.
- Save screenshots and raw notes in the GitHub evidence folder.
Run the Multi-LLM Synthesis
- Use the same shared prompt for Claude, ChatGPT, Gemini, and DeepSeek.
- Do not give extra information to one model that the other models do not receive.
- Save all four raw AI outputs in GitHub.
- Complete the Multi-LLM Comparison Table.
- Write a human judgment paragraph explaining where your team agrees with or overrides the AI models.
Lock the W3 Prediction
- Commit your final W3 prediction file before Sunday 7 June 2026, midnight SGT.
- Recommended filename:
prediction_2026-W23_[teamname].md. - Create a GitHub release tag named
vW23. - Teams 4 and 5: your release tag status is urgent. Fix it before the deadline.
- The GitHub timestamp is evidence. Do not edit the locked prediction after the deadline.
Live Presentation — W2 Calibration Review + W3 Prediction Defence
- Every role holder must speak. No silent role holders.
- Show GitHub live: prediction file, evidence folder, AI outputs, comparison table, and release tag.
- Present W2 calibration first, then defend the W3 prediction.
- Product Owner leads. Scrum Master manages time.
- Professor Dr. Tan may direct questions to specific individuals.
Plain-English summary: Score W2, learn from it, build W3 evidence, compare four AI models, write your team’s own judgment, lock the W3 prediction in GitHub, and come prepared to defend it on Monday 8 June.
How to Submit Your W3 Work
Due Sunday 7 June · before midnight SGTCommit Evidence and Prediction
- Commit all W3 evidence files to your team GitHub repository.
- Commit your final prediction file as
prediction_2026-W23_[teamname].md. - Create a release tag named
vW23. - Check that the commit timestamp is before Sunday 7 June midnight SGT.
Prepare the Monday 8 June Presentation
- Use one PowerPoint deck for the W2 calibration review and W3 prediction defence.
- Include speaking notes so every role holder knows exactly what to say.
- Every role holder must speak during the live presentation.
- Save as
presentation_2026-W23_[teamname].pptx.
Send Final Submission Reference
- Post your public GitHub repository URL in the correct CP3405 Discord channel.
- Send Professor Dr. Tan a Discord DM with the GitHub URL and PowerPoint attachment.
- Include a one-sentence summary of your SPX, NDX, and IWM direction and confidence.
Hard deadline: Sunday 7 June 2026, before midnight SGT. Late work risks not being accepted. The GitHub timestamp is your evidence.
The Nine Assets and Eleven Sectors
ReferenceEach team predicts the direction and approximate weekly range for SPX, NDX, RUT, Gold, Oil, 10-Year Yield, TLT, VIX, and Bitcoin. Teams also identify the S&P 500 sector expected to lead and lag.
| Asset | What it tells you | Common relationship |
|---|---|---|
| S&P 500 | Broad US market benchmark | Risk appetite |
| Nasdaq 100 | Growth, technology, AI, semiconductors | Very sensitive to yields |
| Russell 2000 | Small-cap domestic US economy | Weakness can warn of growth stress |
| Gold | Fear, inflation, currency distrust | Often rises when confidence falls |
| Oil | Inflation, energy, geopolitical stress | Boosts energy, pressures consumers |
| 10-Year Yield | Global finance gravity | Higher yields pressure stocks and bonds |
| TLT | Long-duration US Treasury bond ETF | Usually moves opposite yields |
| VIX | Fear index | Usually rises when stocks fall |
| Bitcoin | 24/7 risk appetite and liquidity signal | Often leads risk-on/risk-off mood |