Audit round 3: aggte() cryptic KeyError on non-CS input

Third-round audit focused on boundary conditions, data-flow edge
cases, and cross-module API consistency.  Only one genuine new
issue surfaced; everything else (Mammen math, RCS IF normalisation,
breakdown_m formula, ggdid ordering, save_to path handling, BJS
Optional import) verified correct on close re-reading.

Issue: `aggte(sa_result)` crashed with `KeyError: 'group'` deep
inside `_weights_dynamic`, for the same reason as the round-2
`cs_report(sa_result)` bug — Sun-Abraham / BJS / dCDH event-study
frames carry `relative_time` but no `group` / `time` columns.

Round 2 fixed the cs_report entry point but aggte is also public
API, so the same error can be hit directly:

    sa = sp.sun_abraham(...)
    sp.aggte(sa, type='dynamic')   # <-- cryptic KeyError

Fix: identical upfront guard in `aggte()` — require the CS detail
schema {group, time, att, se, relative_time} and raise a clear
ValueError pointing to honest_did() for non-CS event studies.

Test: `tests/test_aggte.py::test_aggte_rejects_non_cs_result`.

Full DiD suite: 170/170 (was 169).

Items verified clean on this pass (kept notes in case a future
refactor touches them):
  * Mammen bootstrap: E[V]=0, Var(V)=1 confirmed algebraically;
    IQR/iqr_norm equals σ for normal bootstrap draws.
  * SE scaling: psi = inf @ W.T, boot = V @ centered / n, SE from
    IQR gives the correct √(Var(ψ)/n).
  * RCS IF: 4-cell ratio estimator IF is proportional to
    1{i ∈ cell}(Y_i − Ȳ_cell)/p_cell — correct for sample means.
  * breakdown_m: n_drift = max(e+1, 1) matches CS2021 smoothness
    bias bound M · (e + 1) for post-treatment e ≥ 0.
  * aggte-with-anticipation: `_get_gt_pairs` correctly skips cohorts
    lacking a valid pre-period; downstream `_cohort_weight_series`
    only uses cohorts actually present in detail, so the weights
    renormalise rather than double-count.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: brycewang-stanford

CONTRIBUTING.md

@@ -16,8 +16,8 @@ We welcome contributions to StatsPAI! This document provides guidelines for cont
 
 1. **Fork the Repository**
    ```bash
-   git clone https://github.com/brycewang-stanford/pyEconometrics.git
-   cd pyEconometrics
+   git clone https://github.com/brycewang-stanford/StatsPAI.git
+   cd StatsPAI
    ```
 
 2. **Set Up Development Environment**

src/statspai/did/aggte.py

@@ -138,6 +138,17 @@ def aggte(
             "result has no ATT(g,t) detail to aggregate — was this produced "
             "by callaway_santanna()?"
         )
+    # aggte needs the CS (g, t) grid; SA / BJS / dCDH event studies don't
+    # carry it and would otherwise crash deep inside the weight builders.
+    _required_cols = {'group', 'time', 'att', 'se', 'relative_time'}
+    _have = set(detail.columns)
+    if not _required_cols.issubset(_have):
+        raise ValueError(
+            "aggte() requires a Callaway–Sant'Anna result (detail frame "
+            f"with columns {sorted(_required_cols)}).  Got a result of "
+            f"method {result.method!r} with columns {sorted(_have)}.  For "
+            "Sun–Abraham or BJS event studies use honest_did() directly."
+        )
     if inf_matrix is None:
         # Analytic fallback still works but only gives pointwise intervals.
         bstrap = False

tests/test_aggte.py

@@ -206,3 +206,23 @@ def test_dynamic_overall_averages_post_event_only(cs_result):
     # Sanity: on this DGP the two differ enough that a silent regression
     # would be caught.
     assert not np.isclose(post_mean, all_mean, atol=1e-3)
+
+
+def test_aggte_rejects_non_cs_result():
+    """Passing SA / BJS output must raise a clear error instead of KeyError."""
+    import numpy as np
+    import statspai as sp
+
+    rng = np.random.default_rng(0)
+    rows = []
+    for u in range(60):
+        g_val = [3, 5, 7, 0][u // 15]
+        ui = rng.normal(scale=0.3)
+        for t_ in range(1, 9):
+            te = max(0, t_ - g_val + 1) * 0.5 if g_val > 0 else 0
+            rows.append({'i': u, 't': t_, 'g': g_val,
+                         'y': ui + 0.2 * t_ + te + rng.normal()})
+    df = pd.DataFrame(rows)
+    sa = sp.sun_abraham(df, y='y', g='g', t='t', i='i')
+    with pytest.raises(ValueError, match='Callaway'):
+        aggte(sa, type='dynamic', n_boot=50, random_state=0)