Address PR #440 R3 review (1 P1 — same class, different §4 site)

R2 fixed two §4/§5 sites that misframed event-study mechanics. R3
reviewer caught a third site at the §4 cell that introduces the
per-horizon SE-ratio table (`L422-426`). It said the post-period
horizons "aggregate across all post-period observations" and the
variance is "read off the full panel" — wrong. HAD event-study
fits each horizon as a SEPARATE local-linear on that horizon's
first differences (`ΔY_{g,t} = Y_{g,t} - Y_{g,F-1}`) against the
common `D_{g,F}` regressor (paper Appendix B.2;
`diff_diff/had.py:4298-4451`); the pointwise per-horizon SE reads
each horizon's own residual variance, not a panel aggregate.

Also softened the "per-horizon SE ratio should be larger than the
overall ratio" claim — the empirical per-horizon ratios on the
locked seed are mixed ([1.087, 1.0, 0.816, 1.203, 1.127, 1.085,
1.126]), some larger and some smaller than the overall ~1.10x. The
revised text frames this as an empirical observation of how PSU
correlation interacts with each horizon's `ΔY` distribution, NOT a
methodological guarantee.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: igerber

docs/tutorials/22_had_survey_design.ipynb

@@ -419,11 +419,15 @@
     "and only a small share of states near the boundary, the within-PSU\n",
     "correlation has a small lever to act on.\n",
     "\n",
-    "Contrast with the event-study path: the post-period horizons aggregate\n",
-    "across **all post-period observations**, so the variance is read off\n",
-    "the full panel and the survey-design machinery has more cluster\n",
-    "correlation to amplify. The per-horizon SE ratio should be larger\n",
-    "than the overall SE ratio:\n"
+    "Contrast with the event-study path: each event-time horizon is a\n",
+    "**separate** local-linear fit on that horizon's first differences\n",
+    "`ΔY_{g,t} = Y_{g,t} - Y_{g,F-1}` against the common regressor\n",
+    "`D_{g,F}` (paper Appendix B.2). The pointwise per-horizon SE is\n",
+    "computed from each horizon's own residual variance — not aggregated\n",
+    "across post-period observations. So the per-horizon survey-vs-naive\n",
+    "ratio is an empirical property of how PSU correlation interacts with\n",
+    "each horizon's `ΔY` distribution on this seeded DGP, not a\n",
+    "method-contract guarantee. We compute it below to inspect:\n"
    ]
   },
   {