A yield-curve result is not a bond return

Name the target precisely

Suppose the 10-year Treasury par yield changes from 4.0% to 4.5%. That is a rise of 0.5 percentage points or 50 basis points. It is not a 0.5% bond return. Price response depends on duration, convexity, coupon, horizon and execution assumptions.

In the yield curve lesson, calculate the 10Y–2Y spread in basis points. Then use the Treasury research project to ask whether an inverted spread predicts a later change in the 10-year quote. The project compares a conditional forecast with an unconditional mean baseline.

Audit before modelling

Check date order, duplicates, missing quotes, unit conventions, large gaps and extreme values with the CSV checker. Save the exact source and its SHA-256 digest. A historical archive may lack release timestamps or revision history, so do not assert that every value was available at a historical decision time without further evidence.

Make the failure visible

Choose a horizon using the development interval. Inspect a later holdout once, report the baseline error, and show sensitivity to other horizons. If the conditional forecast is worse, keep that result in the report. Overlapping forecast windows also mean errors are dependent. State those limits when you defend the project.