Muttersprache koppeln und russischen A1-Kurs ergänzen
GUI und neue Erklärungen verwenden gemeinsam native_language. Deutsch und Russisch sind die beiden Muttersprachen; Englisch bleibt die Lernsprache. Kompatibilitätsfelder werden atomar gekoppelt, widersprüchliche Eingaben abgewiesen und bestehende Profile sowie eingefrorene Inhalte erhalten. Ergänzt vollständige russische GUI-Kataloge und 840 geprüfte Aufgabenansichten in sieben Modulen einschließlich A1-Test, mit getrennten Sprachrollen für Ausgangs- und Lernmaterial. Äquivalente Ansichten teilen ihre Nachweisidentität. Stabile API-Fehlercodes, neue deterministische Hilfen und Trainingsbegründungen berücksichtigen die Muttersprache. Dokumentiert das Sprachmodell in AGENTS.md. Validierung: Go-Race-Suite mit eigener lokaler PostgreSQL-Testdatenbank, go vet, Web-Build, Offline-Bank-/Variantenaudit, Diff-Prüfung sowie lokale Browser-Smokes für Sprachen, Kurs, Administration, Aussprache und Aufnahme. Keine Live-Anbieter, Produktionsprofile, Deployment-Schritte oder historischen Neubewertungen.
This commit is contained in:
+111
-63
@@ -1,8 +1,9 @@
|
||||
"""Offline build/audit of explicitly reviewed explanation-language variants."""
|
||||
"""Offline compilation of reviewed Russian native-language presentations."""
|
||||
import copy
|
||||
import hashlib
|
||||
import json
|
||||
import pathlib
|
||||
import re
|
||||
|
||||
BASE = pathlib.Path(__file__).resolve().parent
|
||||
DEST = BASE.parents[1] / 'apps/api/internal/store/coursebank/variants'
|
||||
@@ -10,92 +11,139 @@ BANK = DEST.parent / 'en-US/a1-v2'
|
||||
|
||||
|
||||
def checked_sources():
|
||||
review_bytes = (BASE / 'variant_review.json').read_bytes()
|
||||
review = json.loads(review_bytes)
|
||||
assert review['reviewed'] and review['entry_count'] == 12
|
||||
data = (BASE / 'variant_review.json').read_bytes()
|
||||
review = json.loads(data)
|
||||
assert review['reviewed'] and review['entry_count'] == 840
|
||||
for name, digest in review['sources'].items():
|
||||
assert hashlib.sha256((BASE / name).read_bytes()).hexdigest() == digest, f'Variant review is stale: {name}'
|
||||
return json.loads((BASE / 'variant_sources.json').read_text()), review_bytes
|
||||
return json.loads((BASE / 'variant_sources.json').read_text()), data
|
||||
|
||||
|
||||
def dictionary(name):
|
||||
rows = [line.split('|') for line in (BASE / 'russian' / f'{name}.tsv').read_text().splitlines() if line]
|
||||
assert all(len(row) == 2 and all(row) for row in rows), name
|
||||
result = dict(rows)
|
||||
assert len(result) == len(rows), f'Duplicate source in {name}'
|
||||
return result
|
||||
|
||||
|
||||
def compile_variants():
|
||||
source, review_bytes = checked_sources()
|
||||
source, review = checked_sources()
|
||||
translations = {name: dictionary(name) for name in ('vocabulary', 'grammar', 'questions', 'writing', 'open')}
|
||||
entries = []
|
||||
for row in source['entries']:
|
||||
candidates = json.loads((BANK / f'introductions.{row["skill"]}.json').read_text())
|
||||
base = candidates[row['number'] - 1]
|
||||
task = copy.deepcopy(base['task'])
|
||||
variant_id = f'intro_en_explanations_v1_{row["skill"]}_{row["number"]:02d}'
|
||||
canonical = base['id'] if row['equivalence'] == 'translation' else variant_id
|
||||
task.update(instruction_language=source['instruction_language'], variant_id=variant_id,
|
||||
canonical_task_id=canonical, content_revision=source['revision'], instruction=row['instruction'])
|
||||
task['prompt'] = row['instruction']
|
||||
if 'stimulus' in row:
|
||||
task['stimulus'] = row['stimulus']
|
||||
task['prompt'] += ' ' + row['stimulus']['text']
|
||||
key = task['answer_key']
|
||||
key['explanation'] = row['explanation']
|
||||
key['rubric'] = row.get('rubric', ['Choose the one answer supported by the sentence, word meaning or supplied text.'])
|
||||
key['reference_answer_language'] = source['instruction_language'] if task['skill'] == 'vocabulary' else task['target_language']
|
||||
for option in task.get('options', []):
|
||||
option['role'] = 'meaning' if task['skill'] == 'vocabulary' else 'target'
|
||||
option['language'] = source['instruction_language'] if option['role'] == 'meaning' else task['target_language']
|
||||
if 'option_meanings' in row:
|
||||
option['label'] = row['option_meanings'][option['label']]
|
||||
if option['id'] == key.get('correct_option_id'):
|
||||
key['reference_answer'] = option['label']
|
||||
entries.append(dict(id=variant_id, revision=source['revision'], base_entry_id=base['id'],
|
||||
canonical_task_id=canonical, instruction_language=source['instruction_language'],
|
||||
module_title=source['module_title'], module_description=source['module_description'],
|
||||
equivalence=row['equivalence'], reviewed=True, task=task))
|
||||
return entries, review_bytes
|
||||
for path in sorted(BANK.glob('*.json')):
|
||||
if path.name == 'manifest.json':
|
||||
continue
|
||||
module = path.name.split('.')[0]
|
||||
for base in json.loads(path.read_text()):
|
||||
task = copy.deepcopy(base['task'])
|
||||
skill, kind = task['skill'], task['kind']
|
||||
variant_id = base['id'] + '_ru_v1'
|
||||
key = task['answer_key']
|
||||
instruction = ''
|
||||
stimulus = None
|
||||
if skill == 'grammar':
|
||||
instruction = 'Дополните предложение.'
|
||||
stimulus = dict(role='target', language='en-US', text=task['prompt'].removeprefix('Ergänze: '))
|
||||
key['explanation'] = translations['grammar'][key['explanation']]
|
||||
elif skill == 'vocabulary':
|
||||
word, example = re.fullmatch(r'Was bedeutet „(.*?)“ in „(.*?)“\?', task['prompt']).groups()
|
||||
instruction = f'Что означает «{word}» в этом предложении?'
|
||||
stimulus = dict(role='target', language='en-US', text=example)
|
||||
for option in task['options']:
|
||||
label = option['label']
|
||||
option['label'] = 'до' if label == 'vor' and word in ('before', 'after') else translations['vocabulary'][label]
|
||||
key['reference_answer'] = next(o['label'] for o in task['options'] if o['id'] == key['correct_option_id'])
|
||||
key['explanation'] = f'В этом предложении «{word}» означает «{key["reference_answer"]}».'
|
||||
elif skill in ('reading', 'listening'):
|
||||
instruction = translations['questions'][task['prompt']]
|
||||
key['explanation'] = ('В тексте указан ответ: ' if skill == 'reading' else 'В аудиозаписи указан ответ: ') + key['reference_answer'] + '.'
|
||||
elif kind == 'short_text':
|
||||
instruction = 'Напишите это предложение по-английски.'
|
||||
stimulus = dict(role='source', language='ru-RU', text=translations['writing'][task['prompt'].removeprefix('Schreibe diesen Satz auf Englisch: ')])
|
||||
elif kind == 'repeat_sentence':
|
||||
instruction = 'Повторите английское предложение вслух.'
|
||||
else:
|
||||
visible = task['prompt'].removeprefix('Antworte auf Englisch. ').split(' Du darfst eigene oder erfundene Angaben verwenden.')[0]
|
||||
instruction = 'Ответьте по-английски. ' + translations['open'][visible] + ' Можно использовать реальные или вымышленные сведения. '
|
||||
instruction += 'Произнесите два или три коротких предложения, не дольше 30 секунд.' if kind == 'free_speech' else 'Напишите два или три коротких предложения.'
|
||||
if kind == 'choice':
|
||||
key['rubric'] = ['Выберите единственный ответ, соответствующий предложению, значению слова или приведённому тексту.']
|
||||
else:
|
||||
key['rubric'] = [instruction,
|
||||
'Оценивайте только явно указанные требования и конкретные языковые ошибки с учётом уровня A1. Неясности задания не считайте ошибками учащегося.',
|
||||
'Допустимы другие правильные формулировки. Образец ответа не является обязательной формулировкой.']
|
||||
if kind in ('free_text', 'free_speech'):
|
||||
key['rubric'].append('Собственные или вымышленные имена, места, числа и личные сведения допустимы. Данные из образца не обязательны.')
|
||||
key['explanation'] = 'Образец — один из возможных ответов. Допустимы другие понятные формулировки уровня A1, выполняющие явно указанные требования.'
|
||||
if skill == 'speaking':
|
||||
key['rubric'].append('Содержание и английскую речь оценивайте отдельно от произношения Azure. Без реальных измерений не утверждайте наличие ошибок произношения или беглости.')
|
||||
key['explanation'] += ' Произношение оценивается отдельно и только по реальным измерениям.'
|
||||
key['reference_answer_language'] = 'ru-RU' if skill == 'vocabulary' else 'en-US'
|
||||
for option in task.get('options', []):
|
||||
option['role'] = 'meaning' if skill == 'vocabulary' else 'target'
|
||||
option['language'] = 'ru-RU' if skill == 'vocabulary' else 'en-US'
|
||||
task.update(instruction_language='ru-RU', variant_id=variant_id, canonical_task_id=base['id'], content_revision=source['revision'], instruction=instruction, prompt=instruction)
|
||||
if stimulus:
|
||||
task['stimulus'] = stimulus
|
||||
task['prompt'] += ' ' + stimulus['text']
|
||||
title, description = source['modules'][module]
|
||||
entries.append(dict(id=variant_id, revision=source['revision'], base_entry_id=base['id'], canonical_task_id=base['id'], instruction_language='ru-RU', module_title=title, module_description=description, equivalence='translation', reviewed=True, task=task))
|
||||
return entries, review
|
||||
|
||||
|
||||
def build():
|
||||
entries, review_bytes = compile_variants()
|
||||
entries, review = compile_variants()
|
||||
DEST.mkdir(parents=True, exist_ok=True)
|
||||
data = (json.dumps(entries, ensure_ascii=False, indent=2) + '\n').encode()
|
||||
name = 'introductions.en-US.v1.json'
|
||||
(DEST / name).write_bytes(data)
|
||||
manifest = dict(total=len(entries), editorial_review_sha256=hashlib.sha256(review_bytes).hexdigest(),
|
||||
files={name: hashlib.sha256(data).hexdigest()})
|
||||
files = {}
|
||||
for module in sorted({e['task']['module_id'] for e in entries}):
|
||||
name = module + '.ru-RU.v1.json'
|
||||
data = (json.dumps([e for e in entries if e['task']['module_id'] == module], ensure_ascii=False, indent=2) + '\n').encode()
|
||||
(DEST / name).write_bytes(data)
|
||||
files[name] = hashlib.sha256(data).hexdigest()
|
||||
for path in DEST.glob('*.json'):
|
||||
if path.name != 'manifest.json' and path.name not in files:
|
||||
path.unlink()
|
||||
manifest = dict(total=len(entries), editorial_review_sha256=hashlib.sha256(review).hexdigest(), files=files)
|
||||
(DEST / 'manifest.json').write_text(json.dumps(manifest, indent=2) + '\n')
|
||||
print('Wrote 12 reviewed English-explanation pilot variants across six learning areas.')
|
||||
print(f'Wrote {len(entries)} Russian native-language variants across seven modules.')
|
||||
|
||||
|
||||
def audit():
|
||||
expected, review_bytes = compile_variants()
|
||||
expected, review = compile_variants()
|
||||
manifest = json.loads((DEST / 'manifest.json').read_text())
|
||||
assert manifest['total'] == 12
|
||||
assert manifest['editorial_review_sha256'] == hashlib.sha256(review_bytes).hexdigest()
|
||||
assert manifest['total'] == len(expected) == 840
|
||||
assert manifest['editorial_review_sha256'] == hashlib.sha256(review).hexdigest()
|
||||
actual = []
|
||||
for name, digest in manifest['files'].items():
|
||||
data = (DEST / name).read_bytes()
|
||||
assert hashlib.sha256(data).hexdigest() == digest
|
||||
actual += json.loads(data)
|
||||
assert actual == expected, 'Generated variants differ from reviewed editorial sources'
|
||||
assert len({entry['id'] for entry in actual}) == 12
|
||||
assert sorted(actual, key=lambda e: e['id']) == sorted(expected, key=lambda e: e['id'])
|
||||
assert len({e['id'] for e in actual}) == 840
|
||||
base = {e['id']: e for p in BANK.glob('*.json') if p.name != 'manifest.json' for e in json.loads(p.read_text())}
|
||||
for entry in actual:
|
||||
task = entry['task']
|
||||
base = next(x for x in json.loads((BANK / f'introductions.{task["skill"]}.json').read_text()) if x['id'] == entry['base_entry_id'])
|
||||
assert task['target_language'] == task['instruction_language'] == 'en-US'
|
||||
assert task['module_id'] == 'a1_introductions_v1' and task['instruction']
|
||||
assert entry['equivalence'] in ('translation', 'adaptation')
|
||||
assert (entry['canonical_task_id'] == entry['base_entry_id']) == (entry['equivalence'] == 'translation')
|
||||
task, original = entry['task'], base[entry['base_entry_id']]['task']
|
||||
assert entry['canonical_task_id'] == entry['base_entry_id'] and entry['equivalence'] == 'translation'
|
||||
assert task['instruction_language'] == 'ru-RU' and task['target_language'] == 'en-US'
|
||||
for field in ('audio_text', 'passage', 'reference_text'):
|
||||
assert task.get(field) == base['task'].get(field), (entry['id'], field)
|
||||
assert task.get(field) == original.get(field), (entry['id'], field)
|
||||
if task['kind'] == 'choice':
|
||||
assert task['answer_key']['correct_option_id'] == base['task']['answer_key']['correct_option_id']
|
||||
assert len({o['label'].casefold() for o in task['options']}) == 3
|
||||
if entry['equivalence'] == 'translation':
|
||||
assert [o['label'] for o in task['options'] if o['role'] == 'target'] == [o['label'] for o in base['task']['options']]
|
||||
assert all(o['role'] in ('target', 'meaning') and o['language'] for o in task['options'])
|
||||
assert task['answer_key']['correct_option_id'] == original['answer_key']['correct_option_id']
|
||||
assert [o['id'] for o in task['options']] == [o['id'] for o in original['options']]
|
||||
assert len({o['label'].casefold() for o in task['options']}) == 3, entry['id']
|
||||
if task['skill'] != 'vocabulary':
|
||||
assert [o['label'] for o in task['options']] == [o['label'] for o in original['options']]
|
||||
else:
|
||||
assert task['answer_key']['reference_answer'] == original['answer_key']['reference_answer']
|
||||
if task['skill'] == 'listening':
|
||||
assert not task.get('stimulus'), 'A private recording must never become a public stimulus'
|
||||
assert sum(x['equivalence'] == 'adaptation' for x in actual) == 3
|
||||
assert {x['task']['kind'] for x in actual} == {'choice', 'repeat_sentence', 'free_speech', 'short_text', 'free_text'}
|
||||
assert all(sum(x['task']['skill'] == skill for x in actual) == 2 for skill in ('grammar', 'vocabulary', 'listening', 'reading', 'speaking', 'writing'))
|
||||
print('PASS: 12 pilot variants; 9 equivalent translations, 3 separate adaptations; preserved keys/media; six learning areas and five renderers.')
|
||||
assert not task.get('stimulus'), 'Private recording exposed as stimulus'
|
||||
if task['kind'] == 'short_text':
|
||||
assert task['stimulus']['role'] == 'source' and task['stimulus']['language'] == 'ru-RU'
|
||||
for module in {e['task']['module_id'] for e in actual}:
|
||||
for skill in ('grammar', 'vocabulary', 'reading', 'listening', 'writing', 'speaking'):
|
||||
assert sum(e['task']['module_id'] == module and e['task']['skill'] == skill for e in actual) == 20
|
||||
print('PASS: 840 Russian translations; seven modules, six skills; unchanged canonical identities, keys and English media; private listening transcripts protected.')
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
|
||||
Reference in New Issue
Block a user