Running the analysis

# --- Pre-flight data check (flexible, warn-only) --------------------------
# Run AFTER your data is loaded (`data` exists) and the config cells have run.
import difflib
import pandas as pd

try:
    data
except NameError:
    raise NameError("Load your data into `data` first (run the data-loading cell), then re-run this check.")

cols = list(data.columns)

def _resolve(name, role, required=False):
    if not name:
        return None
    if name in cols:
        return name
    near = difflib.get_close_matches(name, cols, n=3, cutoff=0.6)
    flag = "[X]" if required else "[!]"
    print(f"{flag} {role} '{name}' not found in your data.")
    if near:
        print(f"     closest columns: {near}")
    print("     -> update this name in your config, or remove it.")
    return None

# Names you configured (fall back gracefully if a list isn't defined)
treatment = globals().get('TREATMENT_VAR')
outcome   = globals().get('OUTCOME_VAR')
person    = globals().get('PERSON_ID_VAR', 'PersonId')
controls  = []
for listname in ('SUBGROUP_VARS', 'NETWORK_VARS', 'COLLABORATION_VARS'):
    controls += list(globals().get(listname, []) or [])
controls = list(dict.fromkeys(controls))

print(f"Rows: {len(data):,} | Columns: {len(cols)}")
if person in cols:
    print(f"Unique persons ({person}): {data[person].nunique():,}")
date_col = next((c for c in ('MetricDate', 'Date') if c in cols), None)
if date_col:
    print(f"Distinct dates ({date_col}): {data[date_col].nunique():,}")

# Essentials the run truly needs
t = _resolve(treatment, "Treatment", required=True)
y = _resolve(outcome,   "Outcome",   required=True)
p = person if person in cols else _resolve(person, "Person ID", required=True)

# Treatment variation & spread
if t:
    s = pd.to_numeric(data[t], errors='coerce')
    print(f"[ok] Treatment '{t}': min {s.min():g}, median {s.median():g}, max {s.max():g}")
    top_share = s.value_counts(normalize=True, dropna=True).max() if s.notna().any() else 1.0
    if s.nunique(dropna=True) <= 1:
        print("     [!] no variation in treatment - DML cannot estimate an effect.")
    elif top_share > 0.95:
        print(f"     [!] {top_share:.0%} of rows share one value - limited spread; "
              "check you have both low and high Copilot users.")

# Outcome sanity
if y:
    s = pd.to_numeric(data[y], errors='coerce')
    print(f"[ok] Outcome '{y}': {s.isna().mean()*100:.1f}% missing/non-numeric")

# Confounders / subgroup vars you configured
for name in controls:
    _resolve(name, "Configured variable")

# Missing-value summary for configured columns that exist
present = [c for c in [treatment, outcome, person, *controls] if c and c in cols]
miss = (data[present].isna().mean() * 100).round(1)
miss = miss[miss > 0]
if len(miss):
    print("[!] Missing values (these rows are dropped later):")
    print(miss.to_string())

# Available columns, to help you correct any name
print(f"\nAvailable columns ({len(cols)}): {', '.join(cols)}")
print("\n[i] Advisory only. Fix any [X] before running; [!] are safe to review and ignore.")