Add JS/Python profiler parity tests

.github/workflows/audits.yml

@@ -57,6 +57,10 @@ jobs:
           python-version: '3.12'
           cache: 'pip'
 
+      - uses: actions/setup-node@v4
+        with:
+          node-version: '20'
+
       - name: Install faircode + pytest
         run: |
           python -m pip install --upgrade pip

assets/profiler-engine.js

@@ -52,7 +52,17 @@
   var SCORE_DROP_FLAG = 5;
 
   // Pandas-style missing tokens, so JS null-handling matches read_csv defaults.
-  var NA_TOKENS = { '': 1, 'na': 1, 'n/a': 1, 'nan': 1, 'null': 1, 'none': 1 };
+  var NA_TOKENS = {
+  '': 1,
+  'na': 1,
+  'n/a': 1,
+  'nan': 1,
+  'null': 1,
+  // Intentionally exclude "none" to match the Python profiler.
+  // In this project, pd.read_csv() preserves the literal string "none"
+  // as a categorical value, so treating it as missing breaks Python↔JS
+  // parity (see credit_customers.csv).
+};
 
   // ── Keyword lists - MUST mirror faircode/detect.py ─────────────────────
   var KEYWORDS = [
@@ -447,6 +457,12 @@
           flags.push(d.name + ": '" + g.label + "' is under-represented (" +
                      (g.share * 100).toFixed(1) + '%)');
         }
+        if (g.small_group) {
+          flags.push(
+            d.name + ": '" + g.label + "' has only " +
+            g.count + " rows; fairness metrics may be unreliable"
+          );
+        }
       });
       if (d.imbalance_ratio !== null && d.imbalance_ratio >= imbalanceFlag) {
         flags.push(d.name + ': imbalance ratio ' + d.imbalance_ratio.toFixed(1) +

scripts/profile-js.js

@@ -0,0 +1,22 @@
+#!/usr/bin/env node
+
+const fs = require("fs");
+const path = require("path");
+
+// Load the profiler engine. It registers itself on globalThis.FairCodeProfiler.
+require(path.join(__dirname, "..", "assets", "profiler-engine.js"));
+
+const E = globalThis.FairCodeProfiler;
+
+if (process.argv.length !== 3) {
+  console.error("Usage: node scripts/profile-js.js <dataset.csv>");
+  process.exit(1);
+}
+
+const csvPath = process.argv[2];
+const text = fs.readFileSync(csvPath, "utf8");
+
+const table = E.parseCSV(text);
+const result = E.profile(table);
+
+process.stdout.write(JSON.stringify(result));

tests/fixtures/small.csv

@@ -0,0 +1,5 @@
+sex
+M
+M
+M
+F

tests/test_js_parity.py

@@ -0,0 +1,52 @@
+"""Parity tests between the Python and JavaScript profiler implementations."""
+
+from pathlib import Path
+import json
+import subprocess
+
+import pandas as pd
+import pytest
+
+from faircode import profile
+
+FIXTURES = Path(__file__).resolve().parent / "fixtures"
+
+
+@pytest.mark.parametrize(
+    "csv_name",
+    [
+        "small.csv",
+        "adult.csv",
+        "compas-scores-raw.csv",
+        "credit_customers.csv",
+        "AI_Fair_Recruitment_Dataset.csv",
+    ],
+)
+def test_python_js_profiler_parity(csv_name):
+    """The Python and JavaScript profilers should produce equivalent structured JSON."""
+
+    csv = FIXTURES / csv_name
+
+    python_result = profile(pd.read_csv(csv))
+
+    completed = subprocess.run(
+        ["node", "scripts/profile-js.js", str(csv)],
+        capture_output=True,
+        text=True,
+        encoding="utf-8",
+        check=True,
+    )
+
+    javascript_result = json.loads(completed.stdout)
+
+    # Flags are human-readable messages. They duplicate information already
+    # present in the structured output and may differ because Python and
+    # JavaScript format floating-point values differently (e.g. 6.25 -> 6.2
+    # vs 6.3). Compare the structured data instead.
+    python_result = dict(python_result)
+    javascript_result = dict(javascript_result)
+
+    python_result.pop("flags", None)
+    javascript_result.pop("flags", None)
+
+    assert javascript_result == python_result