Mehrsprachige Bedienung und versionierte Aufgabenvarianten einführen

Übersetzungskataloge, getrennte Sprachpräferenzen und einen versionierten Aufgabenpilot ergänzen. Historische Aufgaben und Bewertungen bleiben erhalten; äquivalente Ansichten teilen Nachweise. Lokal mit Race-Tests und PostgreSQL, Vet, Web-Build, Bank-Audits und Browser-Fixtures geprüft. Dieser Stand dient als Ausgangspunkt für die anschließend beauftragte Zusammenführung zur Muttersprache.
This commit is contained in:
Eugenius Naumenko
2026-11-07 16:26:26 +00:00
parent 87684ba0fd
commit f3d42ee2b6
68 changed files with 3895 additions and 481 deletions
+3
View File
@@ -46,3 +46,6 @@ for e in changed:
assert len(t['answer_key']['rubric'])>=4 and 'Werte aus der Beispielantwort sind nicht verbindlich.' in ' '.join(t['answer_key']['rubric']),e['id']
assert all(e['id'].startswith('a1bank_en_v1_') for e in entries if e not in changed)
print('PASS: 144 reviewed grammar keys, 144 contextual vocabulary items, 144 factual scenes; 840 outputs, 42 balanced buckets, media completeness and balanced choice positions.')
from variants import audit as audit_variants
audit_variants()
+4 -1
View File
@@ -75,4 +75,7 @@ def build():
name=module+'.'+skill+'.json';data=(json.dumps(entries,ensure_ascii=False,indent=2)+'\n').encode();(DEST/name).write_bytes(data);manifest['files'][name]=hashlib.sha256(data).hexdigest()
(DEST/'manifest.json').write_text(json.dumps(manifest,indent=2)+'\n')
print('Wrote 840 editorial exercises in 42 files.')
if __name__=='__main__':build()
if __name__=='__main__':
build()
from variants import build as build_variants
build_variants()
+2 -2
View File
@@ -18,8 +18,8 @@
"Open speaking content/language criteria are separated from Azure pronunciation scores."
],
"sources": {
"audit.py": "5c4d78b1f2afe07a3aa6f5fd3fcca98d0fa3caaa1d3086fd8355100c241ea407",
"build.py": "7896dac6fecc35f174d0fa3a70249ec3a86cc7c38e48307966a86444d9c2a0da",
"audit.py": "ec8ad1842390b03ef366726e81baa3f6d24d03165e83aff757de876918e036cf",
"build.py": "e9c047217c6b850e98d88c1ea86995e74fd398915089f51f9552a52c7eb3e077",
"daily_life.grammar.txt": "0b84ee5b65e2fdf63add826c372c1438215a48c9731c266601ba9160e0e25192",
"daily_life.stories.txt": "62bd24776fd4bbbdf7bf5e926f1452f7720e8b061394dced91ff7cc46aed2e03",
"daily_life.vocabulary.txt": "4db8b0cf43b1f154e315cefe82970e32e410e61edb047d517ad7310e5aff23aa",
+23
View File
@@ -0,0 +1,23 @@
{
"revision": "introductions-en-explanations-v1",
"reviewed": true,
"reviewer": "Codex editorial pass",
"entry_count": 12,
"method": "Offline editorial review of all twelve visible instructions, options, reference answers and rubrics. Structural checks supplement this review; no external teacher certification or live model-quality claim.",
"findings_addressed": [
"English explanation and learning language are identical: two vocabulary meaning tasks use contextual definitions, not identity translations.",
"One short-writing translation task is replaced by explicitly guided composition with the required name visible. These three adaptations have new canonical identities.",
"Nine translations retain the assessed requirements, option IDs, answer keys, target texts, and private listening media.",
"Free answers explicitly permit real or invented details. Reference names are examples, except the explicitly requested Anna in guided composition.",
"Free speech content/language and Azure pronunciation remain separate.",
"The pilot module has reviewed English display labels; module IDs and learning goals are unchanged."
],
"sources": {
"variants.py": "fed9ca7f9a3be47bd766729b386696b115bf4016a16e3c6163b2ec8608868843",
"variant_sources.json": "0a8c2445c432d4a8cf62e8d8e922a1be2cf6f4e146cbe2a52d19722c84b075f3",
"introductions.grammar.txt": "63908be080f56268d3cf0ea03f827177d332d4a23f1e01c5a3fbb513e8b2eccf",
"introductions.vocabulary.txt": "c120cdb9ccf7d70c1f61c8fcbee0bcffc8e474ceae420edcb9c59eee8b2356f1",
"introductions.stories.txt": "d187cc61ad8873c5a8c53d956f7b723689218364cce5e196c59ebbadda48405c",
"open_instructions.json": "dc6e2239a146080c8eb2f6d54f3877c5e01afaa9d4b196104184bb6623f24728"
}
}
+152
View File
@@ -0,0 +1,152 @@
{
"schema_version": 1,
"instruction_language": "en-US",
"target_language": "en-US",
"module_id": "a1_introductions_v1",
"revision": "introductions-en-explanations-v1",
"entries": [
{
"skill": "grammar",
"number": 1,
"equivalence": "translation",
"instruction": "Complete the sentence.",
"stimulus": {
"role": "target",
"language": "en-US",
"text": "I ___ Anna."
},
"explanation": "Use am after I."
},
{
"skill": "grammar",
"number": 2,
"equivalence": "translation",
"instruction": "Complete the sentence.",
"stimulus": {
"role": "target",
"language": "en-US",
"text": "You ___ my friend."
},
"explanation": "Use are after you."
},
{
"skill": "vocabulary",
"number": 1,
"equivalence": "adaptation",
"instruction": "What does hello mean in this sentence?",
"stimulus": {
"role": "target",
"language": "en-US",
"text": "Hello, I am Anna."
},
"option_meanings": {
"hallo": "a greeting",
"tschüss": "a farewell",
"bitte": "a polite request"
},
"explanation": "Hello is a greeting. We use it when we meet someone."
},
{
"skill": "vocabulary",
"number": 2,
"equivalence": "adaptation",
"instruction": "What does name mean in this sentence?",
"stimulus": {
"role": "target",
"language": "en-US",
"text": "My name is Ben."
},
"option_meanings": {
"Name": "what a person is called",
"Alter": "how old someone is",
"Wohnort": "where someone lives"
},
"explanation": "A name is what a person is called. The person's name here is Ben."
},
{
"skill": "reading",
"number": 1,
"equivalence": "translation",
"instruction": "What is the person's name?",
"explanation": "The text says: My name is Anna. The correct answer is Anna."
},
{
"skill": "reading",
"number": 2,
"equivalence": "translation",
"instruction": "Which country is Ben from?",
"explanation": "The text says: Ben is from Canada. The correct answer is Canada."
},
{
"skill": "listening",
"number": 1,
"equivalence": "translation",
"instruction": "Which country is Anna from?",
"explanation": "The recording says: I am from Spain. The correct answer is Spain."
},
{
"skill": "listening",
"number": 2,
"equivalence": "translation",
"instruction": "Where does Ben live?",
"explanation": "The recording says: He lives in London. The correct answer is London."
},
{
"skill": "speaking",
"number": 1,
"equivalence": "translation",
"instruction": "Repeat the English sentence.",
"explanation": "Repeat the complete sentence clearly. Pronunciation feedback uses only available speech measurements.",
"rubric": [
"Repeat the displayed English sentence clearly.",
"Do not claim pronunciation or fluency errors without actual Azure measurements."
]
},
{
"skill": "speaking",
"number": 11,
"equivalence": "translation",
"instruction": "Introduce yourself. Give a name and the country you are from. You may use real or invented details. Say two or three short English sentences in at most 30 seconds.",
"explanation": "The example is only one possible answer. You may choose your own name and country. Content and language are assessed separately from Azure pronunciation.",
"rubric": [
"Introduce one person with a name and country of origin.",
"Two or three short, understandable English sentences meet the required length.",
"Real or invented names and countries are allowed. Facts in the example are not mandatory.",
"Assess only the explicitly requested content and actual language errors. Do not penalise ambiguity in the task.",
"For this free-speech response, assess content and English formulations only. Azure pronunciation is displayed separately and is not part of this score."
]
},
{
"skill": "writing",
"number": 1,
"equivalence": "adaptation",
"instruction": "Write one short English sentence to introduce yourself. For this exercise, use the name Anna.",
"stimulus": {
"role": "context",
"language": "en-US",
"text": "Name: Anna"
},
"explanation": "My name is Anna. and I am Anna. are both valid answers. Use the name Anna as requested; other correct ways to introduce yourself are allowed.",
"rubric": [
"Write one understandable English sentence to introduce yourself using the explicitly requested name Anna.",
"Accept My name is Anna., I am Anna. and other correct equivalent formulations.",
"Consider actual A1 language errors proportionately. Do not invent additional requirements."
]
},
{
"skill": "writing",
"number": 11,
"equivalence": "translation",
"instruction": "Introduce yourself. Give a name and the country you are from. You may use real or invented details. Write two or three short English sentences.",
"explanation": "The example is only one possible answer. You may choose your own name and country. Only the explicitly requested information is required.",
"rubric": [
"Introduce one person with a name and country of origin.",
"Two or three short, understandable English sentences meet the required length.",
"Real or invented names and countries are allowed. Facts in the example are not mandatory.",
"Assess only the explicitly requested content and actual language errors. Do not penalise ambiguity in the task."
]
}
],
"module_title": "Greetings & introductions",
"module_description": "Greet people, introduce a name and origin; I am, my name is"
}
+103
View File
@@ -0,0 +1,103 @@
"""Offline build/audit of explicitly reviewed explanation-language variants."""
import copy
import hashlib
import json
import pathlib
BASE = pathlib.Path(__file__).resolve().parent
DEST = BASE.parents[1] / 'apps/api/internal/store/coursebank/variants'
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
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
def compile_variants():
source, review_bytes = checked_sources()
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
def build():
entries, review_bytes = 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()})
(DEST / 'manifest.json').write_text(json.dumps(manifest, indent=2) + '\n')
print('Wrote 12 reviewed English-explanation pilot variants across six learning areas.')
def audit():
expected, review_bytes = compile_variants()
manifest = json.loads((DEST / 'manifest.json').read_text())
assert manifest['total'] == 12
assert manifest['editorial_review_sha256'] == hashlib.sha256(review_bytes).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
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')
for field in ('audio_text', 'passage', 'reference_text'):
assert task.get(field) == base['task'].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'])
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.')
if __name__ == '__main__':
build()
audit()