Spaces:
Sleeping
Sleeping
File size: 25,480 Bytes
fa1fead | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 | """
tests/test_evaluation.py
WHY THIS EXISTS: Tests the evaluation infrastructure itself — not the model.
The deployment gate must block correctly when thresholds are violated.
The custom metrics must compute correctly on known inputs.
These tests run without any API calls — pure Python logic validation.
"""
from __future__ import annotations
import json
from pathlib import Path
import sys
import pytest
sys.path.insert(0, str(Path(__file__).parent.parent))
# ---------------------------------------------------------------------------
# Deployment gate tests
# ---------------------------------------------------------------------------
class TestDeploymentGate:
def _make_report(self, **overrides) -> dict:
base = {
"english": {
"faithfulness": 0.92,
"answer_relevancy": 0.87,
"context_precision": 0.83,
"context_recall": 0.78,
},
"hindi": {
"faithfulness": 0.90,
"answer_relevancy": 0.85,
"context_precision": 0.81,
"context_recall": 0.76,
},
"faithfulness_ratio": 0.978,
"hallucination_rate": 0.04,
"overall_passed": True,
"blocking_reasons": [],
"timestamp": "2026-01-01T00:00:00Z",
}
for key, value in overrides.items():
parts = key.split(".")
obj = base
for part in parts[:-1]:
obj = obj[part]
obj[parts[-1]] = value
return base
def test_all_passing_report_is_approved(self) -> None:
from evaluation.deployment_gate import check_gate
report = self._make_report()
result = check_gate(report)
assert result.approved, f"Expected approved. Failures: {result.blocking_failures}"
def test_hindi_faithfulness_below_085_blocks(self) -> None:
from evaluation.deployment_gate import check_gate
report = self._make_report(**{"hindi.faithfulness": 0.82})
result = check_gate(report)
assert not result.approved
metric_names = [f["metric"] for f in result.blocking_failures]
assert "faithfulness_hi" in metric_names
def test_hallucination_rate_above_005_blocks(self) -> None:
from evaluation.deployment_gate import check_gate
report = self._make_report(hallucination_rate=0.07)
result = check_gate(report)
assert not result.approved
metric_names = [f["metric"] for f in result.blocking_failures]
assert "hallucination_rate" in metric_names
def test_faithfulness_ratio_below_097_blocks(self) -> None:
from evaluation.deployment_gate import check_gate
# Hindi 0.72, English 0.92 → ratio = 0.783 → BLOCKED
report = self._make_report(
**{"hindi.faithfulness": 0.72, "english.faithfulness": 0.92}
)
result = check_gate(report)
assert not result.approved
metric_names = [f["metric"] for f in result.blocking_failures]
assert any("ratio" in m for m in metric_names), f"No ratio failure: {metric_names}"
def test_english_faithfulness_below_09_blocks(self) -> None:
from evaluation.deployment_gate import check_gate
report = self._make_report(**{"english.faithfulness": 0.87})
result = check_gate(report)
assert not result.approved
metric_names = [f["metric"] for f in result.blocking_failures]
assert "faithfulness_en" in metric_names
def test_missing_metric_adds_warning_not_failure(self) -> None:
from evaluation.deployment_gate import check_gate
# Remove a metric entirely — should warn, not block
report = self._make_report()
del report["english"]["context_precision"]
result = check_gate(report)
assert any("precision_en" in w or "context_precision" in w for w in result.warnings), \
f"Expected warning for missing metric. Warnings: {result.warnings}"
def test_multiple_failures_all_reported(self) -> None:
from evaluation.deployment_gate import check_gate
report = self._make_report(
**{
"hindi.faithfulness": 0.80,
"english.answer_relevancy": 0.78,
"hallucination_rate": 0.08,
}
)
result = check_gate(report)
assert not result.approved
assert len(result.blocking_failures) >= 3
def test_gate_result_to_dict_is_json_serialisable(self) -> None:
from evaluation.deployment_gate import check_gate
report = self._make_report()
result = check_gate(report)
# Must serialise to JSON without error
serialised = json.dumps(result.to_dict())
assert len(serialised) > 10
def test_file_not_found_raises(self) -> None:
from evaluation.deployment_gate import load_and_check
with pytest.raises(FileNotFoundError):
load_and_check(Path("/nonexistent/evaluation_report.json"))
def test_nested_key_resolver_missing_key(self) -> None:
from evaluation.deployment_gate import _get_nested
report = {"english": {"faithfulness": 0.91}}
assert _get_nested(report, "english.faithfulness") == pytest.approx(0.91)
assert _get_nested(report, "english.missing_key") is None
assert _get_nested(report, "missing_section.key") is None
def test_exact_boundary_values(self) -> None:
"""Values exactly at threshold must PASS (>= not just >)."""
from evaluation.deployment_gate import check_gate
report = self._make_report(
**{
"hindi.faithfulness": 0.85, # exactly at threshold
"english.faithfulness": 0.90, # exactly at threshold
"hallucination_rate": 0.049, # just under 0.05
"faithfulness_ratio": 0.9722, # above 0.97
}
)
result = check_gate(report)
assert result.approved, f"Exact boundary values should pass. Failures: {result.blocking_failures}"
# ---------------------------------------------------------------------------
# Golden test set validation tests
# ---------------------------------------------------------------------------
class TestGoldenTestSet:
@pytest.fixture(scope="class")
def test_set(self) -> list[dict]:
path = Path(__file__).parent.parent / "evaluation" / "golden_test_set.json"
with open(path, encoding="utf-8") as f:
return json.load(f)
def test_total_count_is_60(self, test_set) -> None:
assert len(test_set) == 60, f"Expected 60 samples, got {len(test_set)}"
def test_exactly_30_english_samples(self, test_set) -> None:
en = [s for s in test_set if s["language"] == "en"]
assert len(en) == 30
def test_exactly_30_hindi_samples(self, test_set) -> None:
hi = [s for s in test_set if s["language"] == "hi"]
assert len(hi) == 30
def test_all_samples_have_required_fields(self, test_set) -> None:
required = {"id", "language", "query", "ground_truth", "contexts", "doc_type", "question_type"}
for sample in test_set:
missing = required - set(sample.keys())
assert not missing, f"Sample {sample.get('id')} missing fields: {missing}"
def test_all_ids_are_unique(self, test_set) -> None:
ids = [s["id"] for s in test_set]
assert len(ids) == len(set(ids)), "Duplicate IDs found in test set"
def test_all_contexts_are_non_empty_lists(self, test_set) -> None:
for sample in test_set:
assert isinstance(sample["contexts"], list), f"Sample {sample['id']}: contexts must be list"
assert len(sample["contexts"]) >= 1, f"Sample {sample['id']}: contexts must be non-empty"
for ctx in sample["contexts"]:
assert isinstance(ctx, str) and len(ctx) > 10, \
f"Sample {sample['id']}: context string too short"
def test_hindi_queries_contain_devanagari(self, test_set) -> None:
"""Hindi samples must have actual Hindi text — not translated English."""
import unicodedata
hi_samples = [s for s in test_set if s["language"] == "hi"]
for sample in hi_samples:
query = sample["query"]
has_devanagari = any(
unicodedata.category(ch) == "Lo" and "\u0900" <= ch <= "\u097F"
for ch in query
)
assert has_devanagari, (
f"Hindi sample {sample['id']} has no Devanagari characters: {query[:60]}"
)
def test_hindi_ground_truths_contain_devanagari(self, test_set) -> None:
"""Ground truth for Hindi samples must be in Hindi, not English."""
import unicodedata
hi_samples = [s for s in test_set if s["language"] == "hi"]
for sample in hi_samples:
gt = sample["ground_truth"]
has_devanagari = any(
unicodedata.category(ch) == "Lo" and "\u0900" <= ch <= "\u097F"
for ch in gt
)
assert has_devanagari, (
f"Hindi ground truth for {sample['id']} has no Devanagari: {gt[:60]}"
)
def test_doc_types_cover_all_expected_categories(self, test_set) -> None:
doc_types = {s["doc_type"] for s in test_set}
expected = {"gst_notice", "insurance_policy", "legal_agreement",
"government_circular", "tax_notice"}
assert expected.issubset(doc_types), \
f"Missing doc types: {expected - doc_types}"
def test_question_types_are_valid(self, test_set) -> None:
valid = {"factual", "analytical", "comparative", "conversational"}
for sample in test_set:
assert sample["question_type"] in valid, \
f"Sample {sample['id']}: invalid question_type '{sample['question_type']}'"
def test_no_empty_queries_or_ground_truths(self, test_set) -> None:
for sample in test_set:
assert sample["query"].strip(), f"Sample {sample['id']}: empty query"
assert sample["ground_truth"].strip(), f"Sample {sample['id']}: empty ground_truth"
# ---------------------------------------------------------------------------
# Custom metrics tests
# ---------------------------------------------------------------------------
class TestLanguageAccuracy:
@pytest.fixture(scope="class")
def test_set(self) -> list[dict]:
path = Path(__file__).parent.parent / "evaluation" / "golden_test_set.json"
with open(path, encoding="utf-8") as f:
return json.load(f)
def test_perfect_score_when_answers_match_query_language(self, test_set) -> None:
from evaluation.custom_metrics import compute_language_accuracy
# Use ground_truth as the "generated" answer — should be perfect score
answers = [s["ground_truth"] for s in test_set]
result = compute_language_accuracy(test_set, answers)
assert result.total_samples == 60
assert result.accuracy >= 0.95, (
f"Ground-truth answers should achieve >= 0.95 language accuracy. "
f"Got {result.accuracy:.4f}. Failures: {result.failures[:3]}"
)
def test_english_answers_for_hindi_queries_fail(self, test_set) -> None:
from evaluation.custom_metrics import compute_language_accuracy
hi_samples = [s for s in test_set if s["language"] == "hi"]
en_samples = [s for s in test_set if s["language"] == "en"]
# Give English answers to Hindi queries — accuracy should drop
# For English queries, use correct English answers
mixed_samples = hi_samples + en_samples
mixed_answers = (
["This is an English answer that should not match Hindi queries."] * len(hi_samples)
+ [s["ground_truth"] for s in en_samples]
)
result = compute_language_accuracy(mixed_samples, mixed_answers)
# Should fail — English answers for Hindi queries
assert result.accuracy < 0.85, (
f"Expected accuracy to drop with wrong language answers. Got {result.accuracy:.4f}"
)
def test_empty_answers_counted_as_correct(self, test_set) -> None:
from evaluation.custom_metrics import compute_language_accuracy
samples = test_set[:5]
answers = [""] * 5 # Empty answers skip language check (fail open)
result = compute_language_accuracy(samples, answers)
assert result.correct == 5, "Empty answers should all count as correct (fail open)"
def test_no_info_responses_counted_as_correct(self, test_set) -> None:
from evaluation.custom_metrics import compute_language_accuracy
en_samples = [s for s in test_set if s["language"] == "en"][:3]
hi_samples = [s for s in test_set if s["language"] == "hi"][:3]
no_info_en = "I don't have enough information in your document to answer this."
no_info_hi = "मुझे इस प्रश्न का उत्तर आपके दस्तावेज़ में नहीं मिला।"
samples = en_samples + hi_samples
answers = [no_info_en] * 3 + [no_info_hi] * 3
result = compute_language_accuracy(samples, answers)
assert result.correct == 6, \
"No-info responses should all pass language check"
def test_mismatched_lengths_raise_value_error(self, test_set) -> None:
from evaluation.custom_metrics import compute_language_accuracy
with pytest.raises(ValueError, match="same length"):
compute_language_accuracy(test_set[:5], ["answer"] * 3)
def test_language_accuracy_result_passed_flag(self) -> None:
from evaluation.custom_metrics import LanguageAccuracyResult
passing = LanguageAccuracyResult(100, 96, 0.96, [], True)
assert passing.passed is True
failing = LanguageAccuracyResult(100, 93, 0.93, [], False)
assert failing.passed is False
class TestCrossLangRecall:
@pytest.fixture(scope="class")
def test_set(self) -> list[dict]:
path = Path(__file__).parent.parent / "evaluation" / "golden_test_set.json"
with open(path, encoding="utf-8") as f:
return json.load(f)
def test_perfect_recall_when_context_used_as_chunk(self, test_set) -> None:
from evaluation.custom_metrics import compute_cross_lang_recall
hi_samples = [s for s in test_set if s["language"] == "hi"]
# Use the reference contexts as the "retrieved chunks" — should be perfect
chunks_per_sample = [
[{"chunk_text": ctx} for ctx in s["contexts"]]
for s in hi_samples
]
result = compute_cross_lang_recall(hi_samples, chunks_per_sample)
assert result.recall >= 0.80, (
f"Using reference contexts as retrieved should score >= 0.80. "
f"Got {result.recall:.4f}"
)
assert result.passed
def test_empty_chunks_do_not_crash(self, test_set) -> None:
from evaluation.custom_metrics import compute_cross_lang_recall
hi_samples = [s for s in test_set if s["language"] == "hi"][:5]
chunks_per_sample = [[] for _ in hi_samples] # No chunks retrieved
result = compute_cross_lang_recall(hi_samples, chunks_per_sample)
assert result.total_queries == 5
assert result.relevant_retrieved == 0
assert result.recall == 0.0
def test_irrelevant_chunks_score_low(self, test_set) -> None:
from evaluation.custom_metrics import compute_cross_lang_recall
hi_samples = [s for s in test_set if s["language"] == "hi"][:10]
# Completely unrelated chunks
chunks_per_sample = [
[{"chunk_text": "The quick brown fox jumps over the lazy dog."}]
for _ in hi_samples
]
result = compute_cross_lang_recall(hi_samples, chunks_per_sample)
# English fox sentence has zero overlap with Hindi legal/GST content
assert result.recall < 0.50, (
f"Irrelevant chunks should score low. Got {result.recall:.4f}"
)
def test_mismatched_lengths_raise(self, test_set) -> None:
from evaluation.custom_metrics import compute_cross_lang_recall
hi_samples = [s for s in test_set if s["language"] == "hi"][:5]
with pytest.raises(ValueError):
compute_cross_lang_recall(hi_samples, [[]] * 3)
# ---------------------------------------------------------------------------
# Language badge builder tests
# ---------------------------------------------------------------------------
class TestLanguageBadge:
def test_hindi_document_gets_hindi_badge(self) -> None:
from api.routes.ingest import _build_language_badge
hindi_text = (
"यह एक हिंदी दस्तावेज़ है। इसमें जीएसटी नोटिस की जानकारी है। "
"करदाता को 30 दिनों के भीतर जवाब देना होगा।"
)
badge = _build_language_badge(hindi_text, "hi", 0.97)
assert badge.primary_language == "hi"
assert badge.primary_language_name == "Hindi"
assert "Hindi" in badge.badge_label
assert "🇮🇳" in badge.badge_label
assert not badge.is_bilingual
def test_english_document_gets_english_badge(self) -> None:
from api.routes.ingest import _build_language_badge
english_text = (
"This is an English document. It contains GST notice information. "
"The taxpayer must respond within 30 days."
)
badge = _build_language_badge(english_text, "en", 0.99)
assert badge.primary_language == "en"
assert badge.primary_language_name == "English"
def test_badge_confidence_stored_correctly(self) -> None:
from api.routes.ingest import _build_language_badge
badge = _build_language_badge("Sample text", "hi", 0.923)
assert badge.confidence == pytest.approx(0.923, abs=0.001)
def test_badge_label_format_hindi(self) -> None:
from api.routes.ingest import _build_language_badge
badge = _build_language_badge("हिंदी पाठ", "hi", 0.95)
# Badge label must contain language name
assert "Hindi" in badge.badge_label
def test_unknown_language_code_handled(self) -> None:
from api.routes.ingest import _build_language_badge
# Unknown code should not crash — uses uppercased code as name
badge = _build_language_badge("Some text", "xy", 0.80)
assert badge.primary_language == "xy"
assert badge.primary_language_name == "XY"
# ---------------------------------------------------------------------------
# SSE stream parser tests
# ---------------------------------------------------------------------------
class TestSSEParser:
def _make_mock_response(self, sse_text: str):
"""Creates a mock requests.Response that yields SSE lines."""
lines = sse_text.encode("utf-8").split(b"\n")
class MockResponse:
def iter_lines(self, decode_unicode=True):
for line in lines:
decoded = line.decode("utf-8") if isinstance(line, bytes) else line
yield decoded
return MockResponse()
def test_parses_sources_event(self) -> None:
from ui.components.answer_panel import _parse_sse_stream
sse = (
'event: sources\n'
'data: {"chunks": [{"chunk_id": "abc", "title": "GST Notice"}], "crag_triggered": false, "total_count": 1}\n'
'\n'
)
events = list(_parse_sse_stream(self._make_mock_response(sse)))
assert len(events) == 1
event_type, data = events[0]
assert event_type == "sources"
assert data["total_count"] == 1
assert len(data["chunks"]) == 1
def test_parses_token_events(self) -> None:
from ui.components.answer_panel import _parse_sse_stream
sse = (
'event: token\n'
'data: {"token": "The"}\n'
'\n'
'event: token\n'
'data: {"token": " GST"}\n'
'\n'
)
events = list(_parse_sse_stream(self._make_mock_response(sse)))
assert len(events) == 2
tokens = [d["token"] for _, d in events]
assert tokens == ["The", " GST"]
def test_parses_done_event(self) -> None:
from ui.components.answer_panel import _parse_sse_stream
sse = (
'event: done\n'
'data: {"language": "hi", "cost_inr": 0.83, "elapsed_ms": 1240}\n'
'\n'
)
events = list(_parse_sse_stream(self._make_mock_response(sse)))
assert len(events) == 1
event_type, data = events[0]
assert event_type == "done"
assert data["language"] == "hi"
assert data["cost_inr"] == pytest.approx(0.83)
def test_parses_error_event(self) -> None:
from ui.components.answer_panel import _parse_sse_stream
sse = (
'event: error\n'
'data: {"message": "Document not found", "code": "not_found"}\n'
'\n'
)
events = list(_parse_sse_stream(self._make_mock_response(sse)))
assert len(events) == 1
event_type, data = events[0]
assert event_type == "error"
assert data["code"] == "not_found"
def test_ignores_comment_lines(self) -> None:
from ui.components.answer_panel import _parse_sse_stream
sse = (
': this is a comment\n'
'event: token\n'
'data: {"token": "Hello"}\n'
'\n'
)
events = list(_parse_sse_stream(self._make_mock_response(sse)))
assert len(events) == 1
assert events[0][1]["token"] == "Hello"
def test_multi_line_data_joined(self) -> None:
from ui.components.answer_panel import _parse_sse_stream
sse = (
'event: token\n'
'data: {"tok\n'
'data: en": "Hi"}\n'
'\n'
)
events = list(_parse_sse_stream(self._make_mock_response(sse)))
# Multi-line data joined with \n — may fail JSON parse, stored as raw
assert len(events) == 1
def test_empty_stream_yields_no_events(self) -> None:
from ui.components.answer_panel import _parse_sse_stream
events = list(_parse_sse_stream(self._make_mock_response("")))
assert events == []
# ---------------------------------------------------------------------------
# HTML escape tests
# ---------------------------------------------------------------------------
class TestHTMLEscape:
def test_escapes_angle_brackets(self) -> None:
from ui.components.answer_panel import _escape_html
result = _escape_html("<script>alert('xss')</script>")
assert "<script>" not in result
assert "<script>" in result
def test_escapes_ampersand(self) -> None:
from ui.components.answer_panel import _escape_html
result = _escape_html("GST & Income Tax")
assert "&" in result
assert " & " not in result
def test_newlines_converted_to_br(self) -> None:
from ui.components.answer_panel import _escape_html
result = _escape_html("Line 1\nLine 2\nLine 3")
assert result.count("<br>") == 2
assert "\n" not in result
def test_hindi_text_not_modified_except_escaping(self) -> None:
from ui.components.answer_panel import _escape_html
hindi = "जीएसटी देनदारी ₹2,45,000 है।"
result = _escape_html(hindi)
# Hindi chars must be preserved
assert "जीएसटी" in result
assert "₹2,45,000" in result
def test_empty_string_returns_empty(self) -> None:
from ui.components.answer_panel import _escape_html
assert _escape_html("") == ""
# ---------------------------------------------------------------------------
# User context middleware tests
# ---------------------------------------------------------------------------
class TestUserContextMiddleware:
def test_valid_uuid_accepted(self) -> None:
import re
pattern = re.compile(
r"^[0-9a-f]{8}-[0-9a-f]{4}-4[0-9a-f]{3}-[89ab][0-9a-f]{3}-[0-9a-f]{12}$",
re.IGNORECASE,
)
valid_uuid = "550e8400-e29b-41d4-a716-446655440000"
assert pattern.match(valid_uuid)
def test_invalid_uuid_rejected(self) -> None:
import re
pattern = re.compile(
r"^[0-9a-f]{8}-[0-9a-f]{4}-4[0-9a-f]{3}-[89ab][0-9a-f]{3}-[0-9a-f]{12}$",
re.IGNORECASE,
)
invalid_ids = [
"not-a-uuid",
"12345",
"550e8400-e29b-31d4-a716-446655440000", # Version 3, not 4
"",
"injection'; DROP TABLE chunks; --",
]
for bad_id in invalid_ids:
assert not pattern.match(bad_id), f"Should reject: {bad_id}"
def test_exempt_paths_list_contains_health(self) -> None:
from api.middleware.user_context import EXEMPT_PATHS
assert "/health" in EXEMPT_PATHS
assert "/docs" in EXEMPT_PATHS
assert "/openapi.json" in EXEMPT_PATHS |