fix(profiler): name a blank header cell Unnamed: N in the web engine (#1046)

assets/profiler-engine.js

@@ -269,8 +269,11 @@
   // Handles quoted fields, escaped quotes (""), and newlines inside quotes.
   // pandas renames a repeated header to name.1, name.2, ... (skipping any name
   // already taken); without this the second copy overwrote the first in every
-  // row object and both dimensions read the LAST column's values (#834).
+  // row object and both dimensions read the LAST column's values (#834). A blank
+  // header cell is named "Unnamed: N" (N = zero-based column index) first, as pandas
+  // does, so the column has the same name in both engines (#1046).
   function dedupeHeaders(columns) {
+    columns = columns.map(function (c, i) { return c === '' ? 'Unnamed: ' + i : c; });
     var used = Object.create(null), seen = Object.create(null);
     columns.forEach(function (c) { used[c] = (used[c] || 0) + 1; });
     var taken = Object.create(null);

faircode/SPEC.md

@@ -23,7 +23,7 @@ that is what the `unfair.py` / `fair.py` audits do. The Profiler answers a diffe
 
 ## 1. Column auto-detection
 
-Repeated header names are made unique the way pandas does (`name`, `name.1`, `name.2`, ... skipping names already taken) before anything else; the browser parser does the same so two same-named columns are two dimensions, never one overwriting the other (#834).
+A blank header cell is named `Unnamed: N` (N is the zero-based column index), as pandas does. Repeated header names are then made unique the same way (`name`, `name.1`, `name.2`, ... skipping names already taken) before anything else; the browser parser does the same so two same-named columns are two dimensions, never one overwriting the other (#834).
 
 **Tokenize** the column name: strip accents (Unicode NFD, then drop U+0300-U+036F, so `Género`
 becomes `Genero`), split on separators **and** camelCase boundaries, then lower-case.

tests/test_js_parity_profile_parsing.py

@@ -223,13 +223,17 @@ def test_python_js_sample_dataset_is_byte_identical():
 @pytest.mark.parametrize("name,text", [
     ("duplicate_headers", "sex,sex,race\nM,F,A\nM,F,B\nM,F,A\nF,M,B\n"),
     ("duplicate_headers_colliding_suffix", "a,a,a,a.1\n1,2,3,4\n"),
+    ("blank_header_cell", "sex,,race\nM,a,x\nF,b,y\nM,a,x\nF,b,y\n"),
+    ("all_blank_header_row", ",,\nM,a,x\nF,b,y\nM,a,x\nF,b,y\n"),
+    ("blank_header_next_to_unnamed_name", "Unnamed: 1,,race\nM,a,x\nF,b,y\nM,a,x\nF,b,y\n"),
     ("quoted_empty_single_column_row", 'region\r\n""\r\nN\r\nS\r\n'),
     ("quoted_empty_last_line", 'region\r\nN\r\n""'),
     ("non_ascii_digit_ages", "age\n٣٠\n25\n40\n٣٠\n"),
 ])
 def test_python_js_parity_on_parser_edge_cases(tmp_path, name, text):
     """#834 (pandas-style header de-duplication), #835 (a quoted-empty row is a
-    missing value, not a blank line) and #836 (only ASCII digits are ages):
+    missing value, not a blank line), #836 (only ASCII digits are ages) and #1046
+    (a blank header cell is "Unnamed: N", not an empty name):
     the CSV parsers and the profile of the result agree on each."""
     path = tmp_path / f"{name}.csv"
     path.write_text(text, encoding="utf-8", newline="")