[QTI] Perseus export publishes wrong answers for input questions that aren't plain numbers #6195

❌ This issue is not open for contribution. Visit Contributing guidelines to learn about the contributing process and how to find suitable issues.

Overview

parser.extract_value accepts an input question answer when a number appears anywhere in it, and publishes that number as the correct answer, so learners who enter the intended answer are marked wrong. This affects 22 items in 4 live channels.

Complexity: Low
Target branch: unstable

Context

• Every parse function in parser.py uses re.search, so it matches a number anywhere in the answer:

  studio/contentcuration/contentcuration/utils/parser.py

  Lines 103 to 141 in 8cd2daf

  	def parse_integer(text):
  	match = INTEGER.search(text)
  	return match and float(to_en(match.group(1)))
  	def parse_decimal(text):
  	match = DECIMAL.search(text)
  	return match and float(to_en(match.group(1)))
  	def parse_fraction(text):
  	match = FRACTION.search(text)
  	return match and float(parse_integer(match.group(2))) / float(
  	parse_integer(match.group(3))
  	)
  	def parse_mixed_number(text):
  	match = MIXED_NUMBER.search(text)
  	if match:
  	number = parse_integer(match.group(1))
  	return (abs(number) + parse_fraction(match.group(3))) * (
  	float(number) / float(abs(number))
  	)
  	return None
  	def parse_percentage(text):
  	match = PERCENTAGE.search(text)
  	return match and extract_value(match.group(1)) / float(100)
  	def parse_exponent(text):
  	eval_str = None
  	match = EXPONENT.search(text)
  	if match:
  	val1 = extract_value(match.group(2) or match.group(4))
  	val2 = extract_value(match.group(5))
  	eval_str = val1 and val2 and "{int}e{exp}".format(int=val1, exp=int(val2))

• The Perseus export replaces each input answer with the result of extract_value:

  studio/contentcuration/contentcuration/utils/assessment/perseus.py

  Lines 74 to 80 in 8cd2daf

  	def _process_input_answers(self, processed_data):
  	"""Extract input answer processing logic"""
  	non_empty_answers = []
  	for answer in processed_data["answers"]:
  	answer["answer"] = extract_value(answer["answer"])
  	if answer["answer"] or answer["answer"] == 0:
  	non_empty_answers.append(answer)

• Stored answers and what they publish as:

  4\.62   -> 0.62
  249\.6  -> 0.6
  8\.5    -> 0.5
  0\.0025 -> 0.0025
  1:3     -> 1.0
  

The Change

• extract_value should return a number only when the whole answer is one, and reject anything else (e.g. 1:3) instead of pulling a number out of part of it.
• A markdown-escaped decimal point should parse as a plain one: 4\.62 reads as 4.62.

How to Get There

• Create an exercise with an input question whose answer is 4\.62, publish the channel, and open the item's JSON in the published .perseus file: the numeric-input answer value is 0.62.
• Or call extract_value("4\\.62") directly.

Out of Scope

• Rewriting the stored \.-escaped answers.
• Finding which editor or import wrote the escapes.

Acceptance Criteria

General

• extract_value returns 4.62, 249.6, 8.5 and 0.0025 for 4\.62, 249\.6, 8\.5 and 0\.0025.
• extract_value returns None for 1:3, 5 apples, and the test_parser.py cases that only parse part of the answer: 1,00,00, 1.-2, 1.2,00, 1/2/123, 12/0, 1.1.23, 1,000e+-3.
• The other existing test_parser.py cases return the same values as today.

Testing

• test_parser.py covers escaped decimals and answers with text around a number.

Testing

• pytest contentcuration/contentcuration/tests/test_parser.py

AI usage

I directed a survey of input question answers on production and chose this issue's scope, including which formats parse and which are rejected. Claude Code wrote and ran the survey scripts, traced the wrong answers to extract_value, and drafted this issue section by section.