From ea19b1ce7f54c75cad7f970d55c383036d0acdf0 Mon Sep 17 00:00:00 2001
From: Lanre Shittu <136805224+Shizoqua@users.noreply.github.com>
Date: Sat, 7 Nov 2026 11:14:06 +0000
Subject: [PATCH] fix(officeqa): ignore sentence-ending periods when
 normalizing answers

normalize_answer keeps "." so decimals such as 1234.5 survive, but that
also kept periods that end a sentence. A correct answer written as "42."
did not match the gold "42", and as a single token it also got zero F1.
hard is the exact-match result and soft is F1, so these answers received
no reward during training and counted as misses in reported scores.

Drop a period unless a digit follows it. Decimal points, signs and
percent signs still affect the match.
---
 CHANGELOG.md                        |  3 +++
 skillopt/envs/officeqa/evaluator.py |  2 ++
 tests/test_officeqa_evaluator.py    | 35 +++++++++++++++++++++++++++++
 3 files changed, 40 insertions(+)
 create mode 100644 tests/test_officeqa_evaluator.py

diff --git a/CHANGELOG.md b/CHANGELOG.md
index a3babc82d..5b0317f68 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -147,6 +147,9 @@ All notable changes to SkillOpt are documented here. This project adheres to
 - Skip step-level edits whose target runs into a protected slow-update or
   appendix region, so a `replace` or `delete` can no longer remove the start
   marker and expose the protected content (thanks @Shizoqua).
+- Ignore sentence-ending periods when normalizing OfficeQA answers, so a
+  correct answer such as `42.` matches the gold `42` instead of scoring zero
+  exact match and F1 (thanks @Shizoqua).
 
 ### Security
 - Prevent managed-identity credentials from being sent to non-Azure or
diff --git a/skillopt/envs/officeqa/evaluator.py b/skillopt/envs/officeqa/evaluator.py
index 124d25d57..37b7aa219 100644
--- a/skillopt/envs/officeqa/evaluator.py
+++ b/skillopt/envs/officeqa/evaluator.py
@@ -12,6 +12,8 @@ def normalize_answer(text: str) -> str:
     text = text.lower().strip()
     text = text.replace(",", "")
     text = "".join(ch for ch in text if ch not in string.punctuation or ch in _NUMERIC_CHARS or ch == "%")
+    # "." is kept for decimals; a period not followed by a digit is sentence punctuation.
+    text = re.sub(r"\.(?!\d)", "", text)
     text = re.sub(r"\b(million|millions|billion|billions|dollars|dollar|nominal)\b", " ", text)
     text = " ".join(text.split())
     return text
diff --git a/tests/test_officeqa_evaluator.py b/tests/test_officeqa_evaluator.py
new file mode 100644
index 000000000..a4bfa9e3b
--- /dev/null
+++ b/tests/test_officeqa_evaluator.py
@@ -0,0 +1,35 @@
+import pytest
+
+from skillopt.envs.officeqa.evaluator import evaluate
+
+
+@pytest.mark.parametrize(
+    ("prediction", "gold"),
+    [
+        ("42.", "42"),
+        ("-3.5.", "-3.5"),
+        ("$1,234.5 million.", "1234.5"),
+        ("12.5%.", "12.5%"),
+        ("U.S.", "US"),
+    ],
+)
+def test_sentence_period_does_not_break_the_match(prediction, gold):
+    result = evaluate(prediction, gold)
+
+    assert result["em"] == 1.0
+    assert result["f1"] == 1.0
+
+
+@pytest.mark.parametrize(
+    ("prediction", "gold"),
+    [("1.5", "15"), ("0.5", "5"), ("-3", "3"), ("42", "43")],
+)
+def test_decimal_point_and_sign_still_matter(prediction, gold):
+    assert evaluate(prediction, gold)["em"] == 0.0
+
+
+def test_partial_answer_with_trailing_period_gets_token_credit():
+    result = evaluate("revenue was 42.", "42")
+
+    assert result["em"] == 0.0
+    assert result["f1"] == pytest.approx(0.5)