han811's picture
add task templates for rebuttal
7e760fb
Raw
History Blame Contribute Delete
9.21 kB
#!/usr/bin/env bash
# Seed-aware grader for INC2: Data Corruption Recovery
# Reads expected values from expected.json for seed-specific grading.
#
# Args: $1=WORKSPACE $2=REPORTS $3=SUBMISSION $4=TASK_DIR [$5=EXPECTED_JSON]
set -o pipefail
WORKSPACE="$1"
REPORTS="$2"
SUBMISSION="$3"
TASK_DIR="$4"
EXPECTED="${5:-$REPORTS/expected.json}"
mkdir -p "$REPORTS"
CHECKS=0; PASSED=0; FAILURES=""
check() {
CHECKS=$((CHECKS + 1))
if eval "$1" 2>/dev/null; then
PASSED=$((PASSED + 1))
else
FAILURES="${FAILURES:+${FAILURES},}$2"
fi
}
cd "$WORKSPACE"
# Write a reusable grading helper that all checks can import
cat > .grader_helpers.py << 'PYEOF'
import json, sys, os
def load_expected(path):
with open(path, 'r', encoding='utf-8') as f:
return json.load(f)
def load_records(path='data.json'):
with open(path, 'r', encoding='utf-8') as f:
return json.load(f)
def check_numeric(val, field, rng):
"""Return error string if val is non-numeric or out of range, else None."""
if val is None:
return f"field '{field}' is null"
try:
num = float(val)
except (TypeError, ValueError):
return f"field '{field}' = {val!r} is not numeric"
lo = rng.get('min')
hi = rng.get('max')
if lo is not None and hi is not None:
if not (lo <= num <= hi):
return f"field '{field}' = {num} out of range [{lo}, {hi}]"
return None
PYEOF
# ── Check 1: data.json exists and is valid JSON ──────────────────────────────
check "python3 -c '
import json
with open(\"data.json\", \"r\", encoding=\"utf-8\") as f:
records = json.load(f)
assert isinstance(records, list) and len(records) > 0
print(\"DATA_JSON_VALID\")
'" "data_json_invalid"
# ── Check 2: recover.py runs without error ───────────────────────────────────
cp data.json data.json.pre_recovery 2>/dev/null || true
check "python3 recover.py" "recover_py_crashes"
# ── Check 3: All records present after recovery (no deletion) ────────────────
check "python3 -c '
import json, sys
sys.path.insert(0, \".\")
expected = json.load(open(\"$EXPECTED\"))
total = expected[\"total_records\"]
records = json.load(open(\"data.json\", encoding=\"utf-8\"))
assert len(records) == total, f\"Expected {total} records got {len(records)}\"
print(\"RECORD_COUNT_OK\")
'" "record_count_wrong"
# ── Check 4: No previously-good records corrupted ────────────────────────────
python3 - << 'PYEOF' > .check4_result.txt 2>&1
import json, sys
sys.path.insert(0, ".")
from .grader_helpers import load_expected, load_records, check_numeric
expected = load_expected("EXPECTED_PLACEHOLDER")
pk = expected["primary_key"]
cf = expected["corrupt_field"]
cpks = set(str(x) for x in expected["corrupted_primary_keys"])
ranges = expected["field_ranges"]
cf_range = ranges.get(cf, {})
records = load_records()
errors = []
for record in records:
pk_val = str(record.get(pk))
if pk_val in cpks:
continue
val = record.get(cf)
err = check_numeric(val, cf, cf_range)
if err:
errors.append(f"Good record pk={pk_val}: {err}")
if errors:
for e in errors:
print(f"ERROR: {e}")
sys.exit(1)
print("GOOD_RECORDS_INTACT")
PYEOF
# Replace placeholder and re-run properly
check "python3 -c '
import json, sys
sys.path.insert(0, \".\")
expected = json.load(open(\"$EXPECTED\"))
pk = expected[\"primary_key\"]
cf = expected[\"corrupt_field\"]
cpks = set(str(x) for x in expected[\"corrupted_primary_keys\"])
cf_range = expected[\"field_ranges\"].get(cf, {})
lo = cf_range.get(\"min\")
hi = cf_range.get(\"max\")
records = json.load(open(\"data.json\", encoding=\"utf-8\"))
for record in records:
pk_val = str(record.get(pk))
if pk_val in cpks:
continue
val = record.get(cf)
assert val is not None, f\"Good pk={pk_val}: {cf} is null\"
try:
num = float(val)
except Exception:
raise AssertionError(f\"Good pk={pk_val}: {cf}={val!r} not numeric\")
if lo is not None and hi is not None:
assert lo <= num <= hi, f\"Good pk={pk_val}: {cf}={num} out of [{lo},{hi}]\"
print(\"GOOD_RECORDS_INTACT\")
'" "good_records_modified"
# ── Check 5: Corrupted records have correct recovered values ─────────────────
check "python3 -c '
import json, sys
expected = json.load(open(\"$EXPECTED\"))
pk = expected[\"primary_key\"]
cf = expected[\"corrupt_field\"]
exp_vals = expected[\"expected_recovered_values\"]
records = json.load(open(\"data.json\", encoding=\"utf-8\"))
pk_map = {str(r.get(pk)): r for r in records}
for pk_val, exp_val in exp_vals.items():
record = pk_map.get(pk_val)
assert record is not None, f\"Record pk={pk_val} missing\"
actual = record.get(cf)
try:
assert abs(float(actual) - float(exp_val)) < 0.01, f\"pk={pk_val}: {cf}={actual!r} expected {exp_val!r}\"
except (TypeError, ValueError):
assert str(actual) == str(exp_val), f\"pk={pk_val}: {cf}={actual!r} expected {exp_val!r}\"
print(\"CORRUPTED_RECORDS_RECOVERED\")
'" "corrupted_records_not_recovered"
# ── Check 6: All numeric fields within specified ranges ──────────────────────
check "python3 -c '
import json, sys
expected = json.load(open(\"$EXPECTED\"))
field_ranges = expected[\"field_ranges\"]
records = json.load(open(\"data.json\", encoding=\"utf-8\"))
for idx, record in enumerate(records):
for field, rng in field_ranges.items():
val = record.get(field)
if val is None:
continue
try:
num = float(val)
except (TypeError, ValueError):
raise AssertionError(f\"record[{idx}]: {field}={val!r} not numeric\")
lo = rng[\"min\"]; hi = rng[\"max\"]
assert lo <= num <= hi, f\"record[{idx}]: {field}={num} out of [{lo},{hi}]\"
print(\"NUMERIC_RANGES_OK\")
'" "numeric_ranges_violated"
# ── Check 7: All required fields present and non-null in every record ─────────
check "python3 -c '
import json, sys
expected = json.load(open(\"$EXPECTED\"))
required = expected[\"required_fields\"]
pk = expected[\"primary_key\"]
records = json.load(open(\"data.json\", encoding=\"utf-8\"))
for idx, record in enumerate(records):
pk_val = record.get(pk, f\"index_{idx}\")
for field in required:
val = record.get(field)
assert val is not None, f\"pk={pk_val}: {field} is null\"
assert str(val).strip() != \"\", f\"pk={pk_val}: {field} is empty\"
print(\"ALL_REQUIRED_FIELDS_PRESENT\")
'" "required_fields_null"
# ── Check 8: validate.py passes all records ───────────────────────────────────
check "python3 validate.py" "validate_py_fails"
# ── Check 9: recover.py is idempotent ────────────────────────────────────────
if [ -f data.json.pre_recovery ]; then
cp data.json.pre_recovery data.json
python3 recover.py 2>/dev/null || true
fi
check "python3 -c '
import json, sys
expected = json.load(open(\"$EXPECTED\"))
pk = expected[\"primary_key\"]
cf = expected[\"corrupt_field\"]
exp_vals = expected[\"expected_recovered_values\"]
total = expected[\"total_records\"]
records = json.load(open(\"data.json\", encoding=\"utf-8\"))
assert len(records) == total, f\"Idempotency: count changed to {len(records)}\"
pk_map = {str(r.get(pk)): r for r in records}
for pk_val, exp_val in exp_vals.items():
actual = pk_map.get(pk_val, {}).get(cf)
try:
assert abs(float(actual) - float(exp_val)) < 0.01
except (TypeError, ValueError):
assert str(actual) == str(exp_val), f\"Idempotency failed pk={pk_val}\"
print(\"IDEMPOTENT_OK\")
'" "recover_not_idempotent"
# ── Check 10: Attestation verdict=pass ───────────────────────────────────────
check "python3 -c '
import json, sys
att = json.load(open(\"$SUBMISSION/attestation.json\"))
assert att.get(\"verdict\") == \"pass\", f\"verdict={att.get(chr(39)+'verdict'+chr(39))!r}\"
print(\"ATTESTATION_OK\")
'" "bad_attestation"
# ── Write score ───────────────────────────────────────────────────────────────
PARTIAL=$(python3 -c "print(round($PASSED/max(1,$CHECKS), 2))")
if [ "$PASSED" -eq "$CHECKS" ]; then
SUCCESS=1; PASS=true
else
SUCCESS=0; PASS=false
fi
FM=$(python3 -c "import json; print(json.dumps([x for x in '${FAILURES}'.split(',') if x]))")
cat > "$REPORTS/score.json" <<JSON
{
"pass": $PASS,
"primary": {"success": $SUCCESS},
"secondary": {
"checks_passed": $PASSED,
"checks_total": $CHECKS,
"partial_score": $PARTIAL
},
"failure_modes": $FM
}
JSON