Correct foundational estimands, randomization, and p-values

[AliFozooni]

Summary

Chapter 1

• distinguish when a difference in means identifies ATT versus ATE
• make the consistency step and the add-and-subtract bias decomposition explicit
• define unit-indexed and population potential-outcome notation and the conditional bar
• make the tablet example consistently school-level and fix two notation/text errors

Chapter 2

• distinguish randomized assignment from equality of the two potential outcomes
• state both potential-outcome balance equalities needed for the ATE result
• explain why assignment can be independent of potential outcomes while treatment still causes the observed outcome
• link the primary source for the online-learning experiment

Chapter 3

• correct the variance, funnel-plot, interval-tail, and standard-error explanations
• compute two-sided p-values as twice the upper-tail probability beyond the absolute z statistic
• regenerate the affected outputs and align the significance narrative with the corrected values

Why this matters

These are the first three chapters, where notation and identification claims become the foundation for every method that follows. The revised text no longer promotes an ATT decomposition into an ATE claim without the second comparability condition, and it makes the role of consistency visible. The inference chapter now uses genuine two-sided p-values for negative and positive test statistics and explains the corresponding confidence intervals consistently.

Validation

• executed Chapter 3 end to end and regenerated only the affected outputs
• built the full Jupyter Book successfully
• validated all three notebook files and checked the diff for whitespace errors

Closes #229
Closes #348
Closes #384
Closes #391
Closes #395
Closes #410
Closes #417
Closes #422
Closes #437
Closes #463
Closes #464
Closes #465
Closes #472
Closes #477
Closes #478
Closes #481
Closes #482