raullenchai/Rapid-MLX

32.1

Adjusted Score

32.1

Raw Score

100%

Time Factor

2026-08-29

Last Push

3.6K

Stars

Python

Language

909.2K

Lines of Code

2.1K

Files

22.1K

Pattern Hits

2026-08-29

Scan Date

0.05

HC Hit Rate

What These Metrics Mean

Adjusted Score: Primary synthetic code indicator. Raw score normalised per 1,000 lines of code and multiplied by the temporal discount factor. This is the definitive comparative metric — use it to rank repositories by AI authorship density.
Raw Score: The unmodified sum of all severity-weighted, context-multiplied pattern match scores before temporal discounting. Reflects the absolute signal strength independent of when the repository was last active.
Time Factor: The temporal discount multiplier (0–100%) applied to the raw score. Repositories last updated before ChatGPT's launch (Nov 2022) receive a 5% factor. Full signal is only assigned to repositories active in the post-adoption era (Jan 2024+).
Pattern Hits: Total count of individual pattern matches across all files and categories. A high hit count with a low score may indicate a very large codebase with isolated AI snippets; a low count with a high score indicates dense, concentrated AI signatures.
HC Hit Rate: High+Critical pattern hits per file, averaged across the repository. This orthogonal signal catches repositories where a few files are densely packed with high-severity AI tells — a strong indicator even when the normalised score appears moderate due to codebase size.
Lines of Code / Files: Total lines and files analysed. The scanner examines 94 file extensions. These denominators are used to normalise the score, enabling fair comparison between repositories of vastly different sizes.

Score History

This chart maps the temporal evolution of the adjusted synthetic code score across successive scan runs. An upward trajectory indicates ongoing incorporation of AI-generated code or expanding LLM-assisted scaffolding; a stable or declining trajectory may reflect active human refactoring, code removal, or the adoption of stricter authorship policies. The dashed secondary line (right axis) independently tracks total raw pattern hit count, which can diverge from the normalised score when codebase size changes significantly between scans.

Severity Breakdown

Classifies detected patterns by their diagnostic confidence and structural impact. CRITICAL patterns (coefficient 10) represent definitive synthetic signatures — hallucinated imports, explicit LLM attribution metadata — virtually never produced by human authors. HIGH (5) indicates strong structural tells such as cross-file repetition or cross-linguistic idioms. MEDIUM (2) covers recognisable conversational padding and AI-specific vocabulary. LOW (1) captures subtle indicators like tautological comments and generic boilerplate that require density to carry independent signal.

CRITICAL 1HIGH 109MEDIUM 3707LOW 18293

Directory Score Breakdown

This horizontal bar chart decomposes the repository's raw synthetic code score by top-level directory, allowing you to pinpoint precisely which modules or components carry the highest AI authorship density. Directories with disproportionately high scores relative to their size warrant targeted manual review: concentrated AI signatures often trace back to mass-generated configuration layers, auto-ported test suites, LLM-scaffolded boilerplate classes, or entire subsystems authored under heavy copilot assistance. Use this view to prioritise your human code-review effort.

Pattern Findings

The scanner identified 22110 distinct pattern matches across 23 syntactic categories. Each entry below represents a discrete location in the source code where the engine recorded a statistically significant AI authorship indicator. Expand any category row to inspect the individual file paths, line numbers, code snippets, and the lexical context (CODE, COMMENT, or STRING) in which each match was detected.

Reading the findings table: The Severity column indicates the diagnostic confidence level (CRITICAL / HIGH / MEDIUM / LOW). The Context column identifies whether the match occurred inside executable code, an inline comment, or a string literal — comment-context matches receive a ×1.5 weight because LLMs systematically over-annotate. The ⚡ bolt icon marks clustered matches: three or more patterns within a 10-line window, each receiving an additional ×1.5 density multiplier as dense clusters constitute far stronger evidence of synthetic authorship than isolated hits.

Decorative Section Separators3194 hits · 11404 pts

Severity	File	Line	Snippet	Context
MEDIUM	install.sh	16	# ── Helpers ──────────────────────────────────────────────────────────────────	COMMENT
MEDIUM	install.sh	45	# ── Interpreter + venv helpers (also sourced by tests/install) ────────────────	COMMENT
MEDIUM	install.sh	245	# ── Banner ───────────────────────────────────────────────────────────────────	COMMENT
MEDIUM	install.sh	254	# ── 1. Check platform ───────────────────────────────────────────────────────	COMMENT
MEDIUM	install.sh	276	# ── 2. Detect RAM → recommend model ──────────────────────────────────────────	COMMENT
MEDIUM	install.sh	290	# ── 3. Find or install Python 3.10+ ─────────────────────────────────────────	COMMENT
MEDIUM	install.sh	371	# ── 4. Migrate from old install location ─────────────────────────────────────	COMMENT
MEDIUM	install.sh	379	# ── 5. Create or update venv + install ───────────────────────────────────────	COMMENT
MEDIUM	install.sh	421	# ── 6. Create symlinks ──────────────────────────────────────────────────────	COMMENT
MEDIUM	install.sh	437	# ── 7. Ensure ~/.local/bin is in PATH ────────────────────────────────────────	COMMENT
MEDIUM	install.sh	458	# ── 8. Verify + done ────────────────────────────────────────────────────────	COMMENT
MEDIUM⚡	evals/run_eval.py	399	# =============================================================================	COMMENT
MEDIUM⚡	evals/run_eval.py	401	# =============================================================================	COMMENT
MEDIUM⚡	evals/run_eval.py	631	# =============================================================================	COMMENT
MEDIUM⚡	evals/run_eval.py	633	# =============================================================================	COMMENT
MEDIUM	evals/run_eval.py	552	# =============================================================================	COMMENT
MEDIUM	evals/run_eval.py	554	# =============================================================================	COMMENT
MEDIUM	evals/run_eval.py	1279	# =============================================================================	COMMENT
MEDIUM	evals/run_eval.py	1281	# =============================================================================	COMMENT
MEDIUM	evals/run_eval.py	1402	# =============================================================================	COMMENT
MEDIUM	evals/run_eval.py	1404	# =============================================================================	COMMENT
MEDIUM	evals/run_eval.py	1555	# =============================================================================	COMMENT
MEDIUM	evals/run_eval.py	1557	# =============================================================================	COMMENT
MEDIUM	evals/run_eval.py	1865	# =============================================================================	COMMENT
MEDIUM	evals/run_eval.py	1867	# =============================================================================	COMMENT
MEDIUM⚡	evals/results/baseline_benchmark.json	333	"content": "# Building a Distributed System on Apple Silicon: A Comprehensive Technical Guide## IntroductionThe ad	CODE
MEDIUM⚡	evals/results/baseline_benchmark.json	333	"content": "# Building a Distributed System on Apple Silicon: A Comprehensive Technical Guide## IntroductionThe ad	CODE
MEDIUM⚡	evals/results/tier1_benchmark.json	333	"content": "# Building a Distributed System on Apple Silicon: A Comprehensive Technical Guide## IntroductionApple	CODE
MEDIUM	community-benchmarks/scripts/aggregate.py	73	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	community-benchmarks/scripts/aggregate.py	75	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	community-benchmarks/scripts/aggregate.py	95	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	community-benchmarks/scripts/aggregate.py	97	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	community-benchmarks/scripts/aggregate.py	136	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	community-benchmarks/scripts/aggregate.py	138	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	community-benchmarks/scripts/aggregate.py	226	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	community-benchmarks/scripts/aggregate.py	228	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_tool_call_value_fidelity.py	699	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_tool_call_value_fidelity.py	707	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	tests/test_tool_call_value_fidelity.py	43	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	tests/test_tool_call_value_fidelity.py	46	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	tests/test_tool_call_value_fidelity.py	76	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	tests/test_tool_call_value_fidelity.py	79	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	tests/test_tool_call_value_fidelity.py	196	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	tests/test_tool_call_value_fidelity.py	203	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_reasoning_parsers.py	87	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_reasoning_parsers.py	89	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_reasoning_parsers.py	1404	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_reasoning_parsers.py	1406	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_reasoning_parsers.py	1614	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_reasoning_parsers.py	1616	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_reasoning_parsers.py	1839	# -----------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_reasoning_parsers.py	1856	# -----------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_reasoning_parsers.py	2221	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_reasoning_parsers.py	2223	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_reasoning_parsers.py	2374	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/test_reasoning_parsers.py	2376	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	tests/test_reasoning_parsers.py	22	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	tests/test_reasoning_parsers.py	24	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	tests/test_reasoning_parsers.py	54	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	tests/test_reasoning_parsers.py	56	# ---------------------------------------------------------------------------	COMMENT
3134 more matches not shown…

Hyper-Verbose Identifiers11725 hits · 10241 pts

Severity	File	Line	Snippet	Context
LOW	community-benchmarks/scripts/validate.py	352	def _check_summary_matches_rounds(bucket_name: str, bucket: dict) -> None:	CODE
LOW	community-benchmarks/scripts/validate.py	399	def _check_path_in_submissions(path: Path) -> None:	CODE
LOW	community-benchmarks/scripts/validate.py	423	def _check_no_duplicate_submission_id(	CODE
LOW	community-benchmarks/scripts/validate.py	465	def _load_submission_id_index(	CODE
LOW⚡	tests/test_tool_call_value_fidelity.py	715	def test_scan_rejects_parameter_with_no_closing_marker():	CODE
LOW	tests/test_tool_call_value_fidelity.py	313	def _recovered_value_parser_only(parser_name: str, text: str) -> str \| None:	CODE
LOW	tests/test_tool_call_value_fidelity.py	318	def test_renderer_is_valid_for_parser(parser_name, wire_format):	CODE
LOW	tests/test_tool_call_value_fidelity.py	347	def test_parser_and_scanner_agree_on_covered_pairs(parser_name, wire_format):	CODE
LOW	tests/test_tool_call_value_fidelity.py	378	def test_repeated_calls_to_the_same_tool_are_all_kept():	CODE
LOW	tests/test_tool_call_value_fidelity.py	403	def test_argument_value_survives_round_trip(parser_name, wire_format, case, value):	CODE
LOW	tests/test_tool_call_value_fidelity.py	440	def test_known_broken_are_still_broken(parser_name, wire_format, case):	CODE
LOW	tests/test_tool_call_value_fidelity.py	466	def test_scanner_path_preserves_values(case, value):	CODE
LOW	tests/test_tool_call_value_fidelity.py	513	def test_malformed_emissions_produce_no_tool_call(case, text):	CODE
LOW	tests/test_tool_call_value_fidelity.py	552	def test_parser_itself_does_not_invent_parameters(parser_name, case, value):	CODE
LOW	tests/test_tool_call_value_fidelity.py	570	def test_ambiguous_marker_ordering_does_not_invent_parameters(case, value):	CODE
LOW	tests/test_tool_call_value_fidelity.py	600	def test_undeclared_sibling_after_close_refuses_parser_call(parser_name, case, value):	CODE
LOW	tests/test_tool_call_value_fidelity.py	614	def test_undeclared_sibling_after_close_refuses_fallback_call(case, value):	CODE
LOW	tests/test_tool_call_value_fidelity.py	625	def test_undeclared_sibling_never_splices_into_previous_value():	CODE
LOW	tests/test_tool_call_value_fidelity.py	640	def test_repeated_parameter_name_is_last_value_wins(parser_name):	CODE
LOW	tests/test_tool_call_value_fidelity.py	681	def test_every_parser_is_covered_or_exempt():	CODE
LOW	tests/test_tool_call_value_fidelity.py	730	def test_scan_requires_outer_marker_when_one_is_requested():	CODE
LOW	tests/test_tool_call_value_fidelity.py	753	def test_marker_text_in_a_value_cannot_fabricate_a_second_call():	CODE
LOW	tests/test_tool_call_value_fidelity.py	779	def test_a_genuinely_repeated_call_is_still_two_calls():	CODE
LOW	tests/test_tool_call_value_fidelity.py	800	def test_no_declared_tools_keeps_the_position_only_rules():	CODE
LOW	tests/test_tool_call_value_fidelity.py	821	def test_a_standalone_undeclared_call_is_not_emitted():	CODE
LOW	tests/test_tool_call_value_fidelity.py	838	def test_the_nemotron_parser_gates_calls_too():	CODE
LOW	tests/test_tool_call_value_fidelity.py	859	def test_the_nemotron_gate_survives_streaming():	CODE
LOW	tests/test_tool_call_value_fidelity.py	890	def test_a_streamed_refusal_still_reaches_the_user_as_text():	CODE
LOW	tests/test_tool_call_value_fidelity.py	922	def test_every_refused_block_reaches_the_user_and_none_twice():	CODE
LOW	tests/test_tool_call_value_fidelity.py	949	def test_reset_clears_the_content_watermark():	CODE
LOW	tests/test_tool_call_value_fidelity.py	987	def test_refused_block_before_valid_call_in_one_delta_is_not_swallowed():	CODE
LOW	tests/test_tool_call_value_fidelity.py	1008	def test_plain_less_than_after_call_is_not_mistaken_for_partial_markup():	CODE
LOW	tests/test_tool_call_value_fidelity.py	1021	def test_disambiguated_partial_opener_releases_every_withheld_byte():	CODE
LOW	tests/test_tool_call_value_fidelity.py	1040	def test_wrapper_close_is_accounted_before_following_prose():	CODE
LOW	tests/test_tool_call_value_fidelity.py	1064	def test_gating_is_off_when_the_request_declares_no_tools():	CODE
LOW⚡	tests/test_reasoning_parsers.py	69	def extract_reasoning_streaming(self, prev, curr, delta):	CODE
LOW⚡	tests/test_reasoning_parsers.py	75	def test_finalize_streaming_default_none(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	80	def extract_reasoning_streaming(self, prev, curr, delta):	CODE
LOW⚡	tests/test_reasoning_parsers.py	104	def test_both_tags_empty_reasoning(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	110	def test_both_tags_empty_content(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	116	def test_both_tags_whitespace_reasoning(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	122	def test_only_end_tag_implicit(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	275	def test_both_tags_in_single_delta(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	283	def test_start_tag_only_in_delta(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	290	def test_no_tags_early_defaults_to_reasoning(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	354	def test_streaming_no_tag_past_threshold(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	361	def test_streaming_no_tag_under_threshold(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	368	def test_finalize_short_no_tag_correction(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	392	def test_finalize_long_no_tag_no_correction(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	399	def test_finalize_with_tags_no_correction(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	406	def test_finalize_empty_no_correction(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	1410	def test_extract_channel_analysis(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	1415	def test_extract_channel_final(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	1420	def test_extract_channel_not_found(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	1425	def test_extract_channel_empty_content(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	1430	def test_extract_channel_with_constrain(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	1435	def test_channel_regex_matches_analysis(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	1441	def test_channel_regex_matches_final(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	1447	def test_channel_regex_matches_constrain(self):	CODE
LOW⚡	tests/test_reasoning_parsers.py	1453	def test_structural_tokens_regex(self):	CODE
11665 more matches not shown…

Over-Commented Block3579 hits · 2830 pts

Severity	File	Line	Snippet	Context
LOW	pyproject.toml	41	# default_stream / new_stream / new_thread_local_stream) are materialised	COMMENT
LOW	pyproject.toml	61	# tokenizer_utils.py:505) — so under 5.13 even a bare ``import mlx_lm``	COMMENT
LOW	pyproject.toml	81	# introduces an HY3-specific regression, we tighten the floor with a	COMMENT
LOW	pyproject.toml	101	# existed, so every fresh-write test would still have passed.	COMMENT
LOW	pyproject.toml	121	# >=3.6 is required: a recent Python 3.12 micro release added a 4th	COMMENT
LOW	pyproject.toml	141	# PR-5 made grammar-constrained tool-calling DEFAULT-ON in 0.10.14, but a	COMMENT
LOW	pyproject.toml	161	# Keep the optional runtime identical to the signed Desktop sidecar.	COMMENT
LOW	pyproject.toml	181	# MTP (Multi-Token Prediction) speculative decoding — Gemma 4 external	COMMENT
LOW	pyproject.toml	261	# in the linter/typer toolchain.	COMMENT
LOW	pyproject.toml	281	# to catch format regressions. NOT used at runtime — the route	COMMENT
LOW	pyproject.toml	301	# #1208-class quantized-KV round-trip bounds) that example tests can't.	COMMENT
LOW	pyproject.toml	321	# Install mode differs per lane because `ci-linux` must stay no-MLX:	COMMENT
LOW	pyproject.toml	401	"mypy>=1.0.0",	COMMENT
LOW	pyproject.toml	421	# Audio dependencies for TTS/STT (mlx-audio)	COMMENT
LOW	pyproject.toml	441	"soundfile>=0.12.0",	COMMENT
LOW	pyproject.toml	461	# Vanilla ``phonemizer>=3.3.0`` removed it in favour of a	COMMENT
LOW	pyproject.toml	481	# F5 cloning, Kokoro's language-specific G2P stack, or voice design. Keeping	COMMENT
LOW	pyproject.toml	621	"vllm_mlx/models/deepseek_v4*.py" = ["UP", "B", "SIM", "N", "F", "I", "E"]	COMMENT
LOW	evals/run_eval.py	1141	else:	COMMENT
LOW	community-benchmarks/scripts/aggregate.py	81	# Metrics we surface per bucket. Keep this list in lockstep with what	COMMENT
LOW	community-benchmarks/scripts/validate.py	81		COMMENT
LOW	community-benchmarks/scripts/validate.py	141	"and re-run."	COMMENT
LOW	community-benchmarks/scripts/validate.py	161	# loose: it accepts date-only strings (``"2026-06-16"``) and	COMMENT
LOW	community-benchmarks/scripts/validate.py	301	)	COMMENT
LOW	community-benchmarks/scripts/validate.py	541	# and the file read. The previous version called	COMMENT
LOW	community-benchmarks/scripts/validate.py	621	# id) so two files with the same id are correctly flagged: removing	COMMENT
LOW	config/requirements-ci-linux.txt	1	# Linux CI lane (`.github/workflows/ci.yml` → `test-matrix`) test dependency set.	COMMENT
LOW	tests/test_tool_call_value_fidelity.py	201	# XPASS, so it cannot serve as the reminder it looks like (see	COMMENT
LOW	tests/test_tool_call_value_fidelity.py	401	"case,value", HOSTILE_VALUES, ids=[c for c, _ in HOSTILE_VALUES]	COMMENT
LOW	tests/test_reasoning_parsers.py	1841	#	COMMENT
LOW	tests/test_reasoning_parsers.py	1881	"Here's the scratchpad:",	COMMENT
LOW	tests/test_reasoning_parsers.py	2121	)	COMMENT
LOW	tests/test_disk_stream_cli_wiring.py	241	mock.patch.object(cli, "_ensure_model_downloaded", lambda a, *k: None),	COMMENT
LOW	tests/test_disk_stream_cli_wiring.py	421		COMMENT
LOW	tests/test_no_mllm_flag.py	181	# `--no-think` is a hidden back-compat alias of `--no-thinking` on	COMMENT
LOW	tests/test_no_mllm_flag.py	941	# Previously the SOP checks hardcoded ("cli.py", "server.py", "benchmark.py")	COMMENT
LOW	tests/test_no_mllm_flag.py	1001	# relative: from .server import load_model	COMMENT
LOW	tests/test_no_mllm_flag.py	1121	# call scan, so a new entrypoint using `from vllm_mlx import	COMMENT
LOW	tests/test_no_mllm_flag.py	1381	test_file = pathlib.Path(__file__).resolve()	COMMENT
LOW	tests/test_no_mllm_flag.py	1821	except UnicodeDecodeError:	COMMENT
LOW	tests/test_no_mllm_flag.py	1921	# Prong 2: routing-shaped string constants anywhere in the	COMMENT
LOW	tests/test_no_mllm_flag.py	2961	# only when there's something to restore — find_spec-based	COMMENT
LOW	tests/test_gemma4_grammar_558.py	161	"%llguidance {}\n"	COMMENT
LOW	tests/test_gemma4_grammar_558.py	221		COMMENT
LOW	tests/test_gemma4_grammar_558.py	1381	}	COMMENT
LOW	tests/test_tool_grammar_558.py	621	# Under ``required`` (non-reasoning) the FIRST call has NO prefix at all — it	COMMENT
LOW	tests/test_tool_grammar_558.py	681	f"required rejected two newline-separated calls ({accepted}/{total}, "	COMMENT
LOW	tests/test_tool_grammar_558.py	821	"zero-or-one)"	COMMENT
LOW	tests/test_tool_grammar_558.py	841	# * FORCED + reasoning OPTS OUT of the #558 grammar — ``build_tool_grammar``	COMMENT
LOW	tests/test_tool_grammar_558.py	1541	)	COMMENT
LOW	tests/test_harmony_stop_final_channel_only.py	421	assert stop_str == "</execute_ipython>"	COMMENT
LOW	tests/test_deep_nest_dos.py	81	# NOTE: ``ChatCompletionRequest`` MUST be referenced from a	COMMENT
LOW	tests/test_envelope_field_extraction.py	601	_walk_loc_with_root,	COMMENT
LOW	tests/test_audio_upload_size_limit.py	21	# SKIP NOTE: this test file has heavy environmental requirements that	COMMENT
LOW	tests/conftest.py	1	# SPDX-License-Identifier: Apache-2.0	COMMENT
LOW	tests/conftest.py	281		COMMENT
LOW	tests/test_no_out_of_band_routing.py	61	# Codex round-D fix (PR #409): allowlisting by NAME alone is bypass-	COMMENT
LOW	tests/test_no_out_of_band_routing.py	81	"_coerce", # model_aliases.py — builds AliasProfile.	COMMENT
LOW	tests/test_no_out_of_band_routing.py	121	{	COMMENT
LOW	tests/test_no_out_of_band_routing.py	141	# apply_top_p + apply_top_k + categorical_sampling chain when set;	COMMENT
3519 more matches not shown…

Excessive Try-Catch Wrapping1037 hits · 928 pts

Severity	File	Line	Snippet	Context
LOW	bench/bench_spec_decode_mtp.py	679	except Exception as exc: # pragma: no cover — bench	CODE
LOW⚡	evals/run_eval.py	411	except Exception:	CODE
LOW⚡	evals/run_eval.py	621	except Exception:	CODE
LOW	evals/run_eval.py	426	except Exception:	CODE
LOW	evals/run_eval.py	436	except Exception:	CODE
LOW	evals/run_eval.py	1254	except Exception as e:	CODE
LOW	evals/run_eval.py	1379	except Exception as e:	CODE
LOW	evals/run_eval.py	1533	except Exception as e:	CODE
LOW	evals/run_eval.py	1844	except Exception as e:	CODE
LOW	evals/run_eval.py	1986	except Exception:	CODE
LOW	evals/coherence_gate.py	117	except Exception:	CODE
LOW	evals/coherence_gate.py	172	except Exception as exc: # server/protocol error mid-run -> a gate failure	CODE
LOW	evals/perf_gate.py	205	except Exception:	CODE
LOW	evals/perf_gate.py	487	except Exception: # noqa: BLE001 — warm-up result is deliberately ignored	CODE
LOW	tests/test_no_mllm_flag.py	813	except Exception:	CODE
LOW	tests/test_mcp_client_mcp2_compat.py	141	except Exception as e:	STRING
LOW	tests/test_gemma4_grammar_558.py	599	except Exception: # pragma: no cover - very old hub	CODE
LOW	tests/test_gemma4_grammar_558.py	605	except Exception: # pragma: no cover - defensive	CODE
LOW	tests/test_continuous_batching.py	68	except Exception:	CODE
LOW	tests/test_tool_grammar_558.py	394	except Exception: # pragma: no cover - old hub without these names	CODE
LOW	tests/test_tool_grammar_558.py	400	except Exception: # pragma: no cover - requests not present	CODE
LOW	tests/test_tool_grammar_558.py	473	except Exception as exc: # noqa: BLE001 — re-raised unless offline cache-miss	CODE
LOW	tests/test_tool_grammar_558.py	508	except Exception as exc: # noqa: BLE001 - re-raised below if all fail	CODE
LOW	tests/test_tool_grammar_558.py	1486	except Exception as exc: # noqa: BLE001 — re-raised unless offline cache-miss	CODE
LOW	tests/test_mllm_executor_cancel.py	199	except Exception:	CODE
LOW	tests/test_mllm.py	48	except Exception:	CODE
LOW	tests/test_mllm.py	76	except Exception:	CODE
LOW	tests/test_tool_grammar_matcher_cache_558.py	140	except Exception as exc: # noqa: BLE001 - record, never swallow silently	CODE
MEDIUM	tests/test_tool_grammar_matcher_cache_558.py	134	def worker():	CODE
LOW	tests/regression_suite.py	39	except Exception:	CODE
LOW	tests/regression_suite.py	443	except Exception as e:	CODE
LOW	tests/regression_suite.py	593	except Exception as e:	CODE
LOW	tests/regression_suite.py	608	except Exception as e:	CODE
LOW	tests/regression_suite.py	680	except Exception as e:	CODE
LOW	tests/test_anthropic_stop_sequences.py	279	except Exception:	CODE
LOW⚡	tests/test_structure_info_hermes_qwen_558.py	400	except Exception: # pragma: no cover - old hub without these names	CODE
LOW⚡	tests/test_structure_info_hermes_qwen_558.py	407	except Exception: # pragma: no cover - requests not present	CODE
LOW⚡	tests/test_structure_info_hermes_qwen_558.py	415	except Exception: # pragma: no cover - httpx not present	CODE
LOW	tests/test_structure_info_hermes_qwen_558.py	577	except Exception as exc: # noqa: BLE001 - re-raised below if all fail	CODE
LOW	tests/test_telemetry_cli.py	198	except Exception:	CODE
LOW	tests/test_batching_deterministic.py	73	except Exception as e: # pragma: no cover - environment-dependent	CODE
MEDIUM	tests/test_batching_deterministic.py	68	def _load():	CODE
LOW	tests/test_event_loop.py	262	except Exception as e:	CODE
LOW	tests/test_batching.py	455	except Exception as e:	CODE
LOW	tests/test_streaming_latency.py	153	except Exception as e:	CODE
LOW	tests/test_body_receive_timeout.py	646	except Exception:	CODE
LOW⚡	tests/test_metal_error_recovery.py	49	except Exception as e:	CODE
LOW⚡	tests/test_mtp_real_weights.py	170	except Exception:	CODE
LOW	tests/test_idle_event_wakeup.py	96	except Exception:	CODE
MEDIUM	tests/test_server.py	341	def make_requests():	CODE
MEDIUM	tests/test_server.py	440	def register_files():	CODE
MEDIUM	tests/test_server.py	481	def manipulate_counter():	CODE
LOW	tests/test_server.py	346	except Exception as e:	CODE
LOW	tests/test_server.py	449	except Exception as e:	CODE
LOW	tests/test_server.py	488	except Exception as e:	CODE
LOW	tests/test_telemetry_error_wiring.py	93	except Exception:	CODE
LOW	tests/test_model_sizes.py	110	except Exception:	CODE
MEDIUM	tests/test_model_sizes.py	105	def test_every_audio_alias_has_a_manifest_entry():	CODE
LOW	tests/test_cache_routes.py	1674	except Exception:	CODE
LOW	tests/test_anthropic_stream_finalize.py	332	except Exception:	CODE
977 more matches not shown…

Unused Imports1064 hits · 890 pts

Severity	File	Line	Context
LOW	bench/bench_spec_decode_mtp.py	39	CODE
LOW	bench/bench_radix_vs_hash.py	44	CODE
LOW	evals/coherence_gate.py	34	CODE
LOW	evals/perf_gate.py	51	CODE
LOW	community-benchmarks/scripts/aggregate.py	62	CODE
LOW	community-benchmarks/scripts/validate.py	37	CODE
LOW	tests/test_tool_call_value_fidelity.py	35	CODE
LOW	tests/test_audio_boot_check.py	15	CODE
LOW	tests/test_audio_extras_lockin.py	24	CODE
LOW	tests/test_disk_stream_cli_wiring.py	49	CODE
LOW	tests/test_no_mllm_flag.py	25	CODE
LOW	tests/test_cli_embeddings_extra.py	30	CODE
LOW	tests/test_mcp_client_mcp2_compat.py	19	CODE
LOW	tests/test_sidecar_vision_smoke.py	1	CODE
LOW	tests/test_generation_config_loader.py	4	CODE
LOW	tests/test_completions_log_redaction.py	24	CODE
LOW	tests/test_seed_reproducibility.py	36	CODE
LOW	tests/test_ddtree_integration.py	4	CODE
LOW	tests/test_anthropic_stream_reasoning.py	19	CODE
LOW	tests/test_r15_routes_stability_batch.py	36	CODE
LOW	tests/test_mllm_penalty_passthrough.py	24	CODE
LOW	tests/test_version_sync.py	13	CODE
LOW	tests/test_reasoning_content_null_rescue.py	37	CODE
LOW	tests/test_video_capabilities.py	1	CODE
LOW	tests/test_alias_subfolder.py	17	CODE
LOW	tests/test_step_coalescing.py	11	CODE
LOW	tests/test_llama_tool_parser_bare_json.py	19	CODE
LOW	tests/test_harmony_stop_final_channel_only.py	36	CODE
LOW	tests/test_audio_probe_consistency.py	25	CODE
LOW	tests/test_mllm_stats.py	4	CODE
LOW	tests/test_telemetry_queue.py	12	CODE
LOW	tests/test_deep_nest_dos.py	40	CODE
LOW	tests/test_telemetry_request_wiring.py	14	CODE
LOW	tests/test_r12_reasoning_sanitizer_required.py	50	CODE
LOW	tests/test_envelope_field_extraction.py	43	CODE
LOW	tests/test_minicpm_tool_parser.py	4	CODE
LOW	tests/test_sidecar_distribution_constraints.py	1	CODE
LOW	tests/test_model_file_guard.py	4	CODE
LOW	tests/test_gdn_in_proj_fusion.py	13	CODE
LOW	tests/test_audio_upload_size_limit.py	10	CODE
LOW	tests/test_mllm_executor_cancel.py	41	CODE
LOW	tests/test_cli_response_cache_flag.py	20	CODE
LOW	tests/test_cli_response_cache_flag.py	31	CODE
LOW	tests/test_agent_query_stdin_and_trust.py	33	CODE
LOW	tests/test_prefix_cache_radix_e2e.py	21	CODE
LOW	tests/conftest.py	360	CODE
LOW	tests/test_deferred_prefix_cache_load.py	16	CODE
LOW	tests/test_forced_alignment_route_hardening.py	28	CODE
LOW	tests/test_kokoro_espeak_repair.py	21	CODE
LOW	tests/test_pflash_server_module_wiring.py	26	CODE
LOW	tests/test_chat_route_vlm_image.py	33	CODE
LOW	tests/test_stt_corrupted_file.py	16	CODE
LOW	tests/test_no_out_of_band_routing.py	29	CODE
LOW	tests/test_audio_r11_b_bundle.py	40	CODE
LOW	tests/test_request_time_alias_resolution.py	27	CODE
LOW	tests/test_response_cache_metrics.py	10	CODE
LOW	tests/test_mllm_usage_tracking.py	26	CODE
LOW	tests/test_share_cli.py	14	CODE
LOW	tests/test_orphan_tool_validation.py	34	CODE
LOW	tests/test_ui_tars_fixes.py	37	CODE
1004 more matches not shown…

AI Slop Vocabulary238 hits · 766 pts

Severity	File	Line	Snippet	Context
MEDIUM	install.sh	229	# When sourced by the test harness we only want the definitions above, not the	COMMENT
MEDIUM	evals/results/baseline_benchmark.json	156	"content": "# Performance Analysis & BenchmarksHere's a comprehensive analysis with benchmarks:\n\n```pythonim	CODE
MEDIUM⚡	evals/results/baseline_benchmark.json	333	"content": "# Building a Distributed System on Apple Silicon: A Comprehensive Technical Guide## IntroductionThe ad	CODE
MEDIUM⚡	evals/results/baseline_benchmark.json	333	"content": "# Building a Distributed System on Apple Silicon: A Comprehensive Technical Guide## IntroductionThe ad	CODE
MEDIUM⚡	evals/results/baseline_benchmark.json	333	"content": "# Building a Distributed System on Apple Silicon: A Comprehensive Technical Guide## IntroductionThe ad	CODE
MEDIUM⚡	evals/results/tier1b_benchmark.json	333	"content": " about building a distributed system on Apple Silicon. They want it to cover architecture design, netw	CODE
MEDIUM⚡	evals/results/tier1b_benchmark.json	333	"content": " about building a distributed system on Apple Silicon. They want it to cover architecture design, netw	CODE
MEDIUM⚡	evals/results/tier1_benchmark.json	333	"content": "# Building a Distributed System on Apple Silicon: A Comprehensive Technical Guide## IntroductionApple	CODE
MEDIUM⚡	evals/results/tier1_benchmark.json	333	"content": "# Building a Distributed System on Apple Silicon: A Comprehensive Technical Guide## IntroductionApple	CODE
MEDIUM⚡	evals/results/tier1_benchmark.json	333	"content": "# Building a Distributed System on Apple Silicon: A Comprehensive Technical Guide## IntroductionApple	CODE
MEDIUM⚡	community-benchmarks/scripts/aggregate.py	90	# adding harness pass-rate columns). Consumers should skip aggregates	COMMENT
MEDIUM⚡	tests/test_disk_stream_cli_wiring.py	187	# 3 & 4. Subprocess harness — real bench_command load path, no model	COMMENT
MEDIUM⚡	tests/test_disk_stream_cli_wiring.py	369	# dead on the benchmark harness. Named analogously to both the bench	STRING
MEDIUM⚡	tests/test_no_mllm_flag.py	1775	# serving entrypoints (test helpers, internal probes, doctor harness	COMMENT
MEDIUM⚡	tests/test_no_mllm_flag.py	1781	# Eval harness — bench / scoring tool, not a serving entrypoint.	COMMENT
MEDIUM	tests/test_harmony_stop_final_channel_only.py	6	Discovered by PR #1048's real Docker OpenHands E2E harness: the	STRING
MEDIUM	tests/test_no_out_of_band_routing.py	221	# Port for doctor harness probe checks, not engine routing.	COMMENT
MEDIUM	tests/test_no_out_of_band_routing.py	346	# harness name (or comma-separated subset) when running a scoped	COMMENT
MEDIUM	tests/test_no_out_of_band_routing.py	348	# profiles_filter`` consumes it to whittle the harness loop down	COMMENT
MEDIUM⚡	tests/test_audio_r11_b_bundle.py	72	# with the same harness pattern.	COMMENT
MEDIUM	tests/test_mtp_cli_wiring.py	667	"""Codex round-D BLOCKING #1 regression guard (bench-harness path).	STRING
MEDIUM	tests/test_mtp_cli_wiring.py	937	"""Codex round-G NIT #4 + round-D cross-check: bench-harness path	STRING
MEDIUM	tests/regression_suite.py	19	# Port can be overridden by the doctor harness (which picks a free port).	COMMENT
MEDIUM⚡	tests/test_stream_include_usage_honored.py	112	# Test harness	COMMENT
MEDIUM⚡	tests/test_pflash_needle.py	39	# in ``pytest.ini`` is reserved for the deferred engine-level harness	COMMENT
MEDIUM	tests/test_pflash_needle.py	91	[4_096, 8_192, 16_384], # token-level only — engine harness exercises longer	CODE
MEDIUM⚡	tests/test_pflash_needle.py	198	# Engine-level harness — deferred.	COMMENT
MEDIUM⚡	tests/test_pflash_needle.py	215	# When someone re-introduces an engine-level harness here, the right	COMMENT
MEDIUM	tests/test_pflash_needle.py	228	# Env contract for that future harness: ``PFLASH_NEEDLE_MODEL`` selects	COMMENT
MEDIUM	tests/test_pflash_needle.py	230	# Re-instate as module-level reads when the harness lands; reading them	COMMENT
MEDIUM⚡	tests/test_audio_r8_a_bundle.py	66	# Helpers (copied from r7-C so this file stands alone — the harness pattern	COMMENT
MEDIUM	tests/test_audio_r8_a_bundle.py	747	# The test harness may run on a workstation whose HF cache	COMMENT
MEDIUM	tests/test_audio_r8_a_bundle.py	813	# snapshot in the test harness) so the cold-start fallback	COMMENT
MEDIUM	tests/test_bench_tier_submit_combo.py	362	# tier=smoke / harness / all). The speed numbers carry through	COMMENT
MEDIUM⚡	tests/test_prefix_cache_persistence.py	2138	# and the next save attempt (e.g. an embedded harness that does multiple	COMMENT
MEDIUM⚡	tests/test_routes_models_effective_parsers.py	44	# Test harness	COMMENT
MEDIUM⚡	tests/test_completions_spec_parity.py	1021	# representation is robust against future ``Annotated[int, ...]``	COMMENT
MEDIUM⚡	tests/test_ui_tars_parser.py	224	# pyautogui, the Anthropic computer-tool harness) — expects	COMMENT
MEDIUM⚡	tests/test_dflash_integration.py	3274	# response. Mirrors the PoC bench harness but goes through our server.	COMMENT
MEDIUM	tests/test_hermes_harness_contract.py	41	# on a real server. The point of this test is the harness invocation	COMMENT
MEDIUM	tests/test_bench_tier_all.py	157	# Speed failed → overall fail, but harness must still have run.	COMMENT
MEDIUM⚡	tests/test_responses_bundle.py	26	# Lightweight engine harness — copy of routes/test_responses_route.py's	COMMENT
MEDIUM	tests/test_community_bench_upload.py	282	Codex round 2 on #1403: the smoke/harness presence rules were gated on	STRING
MEDIUM	tests/test_responses_route.py	157	# first request, so isolate that unrelated boundary in this harness.	COMMENT
MEDIUM⚡	tests/test_agent_query_context_skip.py	116	# the harness wanted vs what the model offered.	COMMENT
MEDIUM	tests/test_agent_query_context_skip.py	193	# genuine harness misconfiguration.	COMMENT
MEDIUM	tests/test_cache_routes.py	2905	# the shared device default-stream context. This is test-harness plumbing	COMMENT
MEDIUM	tests/test_probe_fastpath.py	555	"""Codex r3 BLOCKING #1: an ASGI server / harness that ships	STRING
MEDIUM	tests/test_release_check_m3_port_thread.py	46	# integration harness with a different env-var convention.	COMMENT
MEDIUM	tests/test_pflash_metrics.py	152	assert dropped_expected > 0 # sanity — the harness must be compressing	CODE
MEDIUM⚡	tests/test_tool_choice_enforcement.py	74	# Test harness — recording mock engine + client builder	COMMENT
MEDIUM⚡	tests/test_kv_quant_gate.py	307	# Baseline KV-cache dtype detection (harness helper, hermetic — no mlx load)	COMMENT
MEDIUM⚡	tests/test_kv_quant_gate.py	357	# Decode strips special tokens (harness helper) — codex round 4 blocking #1	COMMENT
MEDIUM⚡	tests/test_kv_quant_gate.py	737	# Gated end-to-end smoke — real harness, cached small model only.	COMMENT
MEDIUM	tests/test_bench_tier_harness.py	28	# tier=harness tests that incidentally hit the smoke probe	COMMENT
MEDIUM	tests/test_bench_tier_harness.py	101	# Each harness name should appear in the per-tier detail block.	COMMENT
MEDIUM	tests/test_bench_tier_harness.py	197	# Crash → tier-level FAIL, but every harness was visited.	COMMENT
MEDIUM	tests/test_bench_tier_harness.py	314	_ = _time.time() - t0 # touched for clarity; harness sweep is mocked	CODE
MEDIUM	tests/test_bench_tier_harness.py	50	the post-#682 harness session would conclude the (mock) server is	STRING
MEDIUM	tests/parsers/fake_tokenizer.py	118	# harness aligned with the existing OutputRouter test fixtures.	COMMENT
178 more matches not shown…

Modern AI Meta-Vocabulary182 hits · 568 pts

Severity	File	Line	Snippet	Context
MEDIUM	pyproject.toml	436	# F5-TTS (pure-MLX, no torch): EN+ZH multilingual + zero-shot voice cloning.	COMMENT
MEDIUM⚡	evals/results/tier1b_benchmark.json	333	"content": " about building a distributed system on Apple Silicon. They want it to cover architecture design, netw	CODE
MEDIUM⚡	tests/test_reasoning_parsers.py	1844	# supposed to emit its chain-of-thought followed by ``</think>`` and	COMMENT
MEDIUM⚡	tests/test_reasoning_parsers.py	1850	# response — which is pure chain-of-thought — into ``content`` and	COMMENT
MEDIUM	tests/test_reasoning_parsers.py	2160	# reclassified as the model's own chain-of-thought.	COMMENT
MEDIUM⚡	tests/test_reasoning_content_null_rescue.py	827	# in the same orchestration the chat route runs, against every	COMMENT
MEDIUM	tests/test_reasoning_content_null_rescue.py	1071	# parser-specific reasoning trace's last RESCUE_TAIL_LENGTH	COMMENT
MEDIUM	tests/test_no_out_of_band_routing.py	391	# chars of the reasoning trace on length-cut mid-think. Power	COMMENT
MEDIUM	tests/test_no_out_of_band_routing.py	457	# F-K-WHISPER-961 anti-hallucination VAD pre-trim opt-out	COMMENT
MEDIUM	tests/test_no_out_of_band_routing.py	461	# before Whisper for silence-hallucination suppression on	COMMENT
MEDIUM	tests/test_model_auto_config.py	1799	# V3Beta hallucination — V3 pattern needs the same boundary.	COMMENT
MEDIUM	tests/test_embeddings_timeout_admission.py	175	# H-09 (route guard) requires the embedding model to be configured.	COMMENT
MEDIUM⚡	tests/test_chatterbox_expressiveness.py	377	# D) Route — Chatterbox zero-shot cloning IS reachable via /v1/audio/speech	COMMENT
MEDIUM⚡	tests/test_pflash_needle.py	206	# async-runtime scaffolding into this test module or (b) drift from the	COMMENT
MEDIUM⚡	tests/test_r12_h06_repair_context_guard.py	45	# tokenizer + a tunable max-context window so we can hit the H-06	COMMENT
MEDIUM	tests/test_embeddings.py	187	# H-09: the route now requires the embedding model to be	COMMENT
MEDIUM⚡	tests/test_finalize_harmony_raw_text.py	506	# the byte-identical reasoning trace into BOTH ``content`` AND	COMMENT
MEDIUM	tests/test_finalize_harmony_raw_text.py	861	# started with the REMAINDER of the reasoning trace ("answer succinctly,	COMMENT
MEDIUM	tests/test_finalize_harmony_raw_text.py	933	# Long no-tag reasoning trace (chat-template pre-injected <think>,	COMMENT
MEDIUM	tests/test_batching_deterministic.py	158	"Bisected to PR #280 (event-driven idle wakeup). The test adds 4 "	CODE
MEDIUM⚡	tests/test_prefix_cache_persistence.py	2132	# from a fs-event-driven antivirus touching cache_dir mid-rename, observed	COMMENT
MEDIUM	…ts/test_r12_m3_responses_stream_leading_items_order.py	376	# The reasoning summary must carry the model's chain-of-thought.	COMMENT
MEDIUM⚡	tests/test_chat_route_tool_tag_leak.py	241	# ``reasoning_content`` must contain the chain-of-thought.	COMMENT
MEDIUM	tests/test_anthropic_think_leak_r12_m1b.py	497	f"(PR #802 design — agentic auto-retry pattern-matches the prefix); "	CODE
MEDIUM	tests/test_routes.py	1206	# context window — both ids share the loaded engine.	COMMENT
MEDIUM	tests/test_routes.py	1528	# H-09 route guard requires an embedding model to be	COMMENT
MEDIUM⚡	tests/test_ui_tars_parser.py	409	# parser owns the chain-of-thought channel, the tool parser owns	COMMENT
MEDIUM⚡	tests/test_ui_tars_parser.py	412	# mirror so the chain-of-thought doesn't double-render when the	COMMENT
MEDIUM	tests/test_stt_vad_pretrim.py	217	# anti-hallucination invariant.	COMMENT
MEDIUM⚡	tests/test_pr_validate_codex.py	1776	"### context window exceeded\nNo blocking issues found.",	CODE
MEDIUM	tests/test_mxfp4_moe_guardrail.py	416	# passed to e.g. monkeypatch.setattr in test scaffolding, which	COMMENT
MEDIUM	tests/test_harmony_finalize.py	588	# The reasoning trace must still flow on the reasoning channel	COMMENT
MEDIUM	tests/test_reasoning_budget_generation.py	939	# tied-embedding model (qwen3): logits come from embed_tokens.as_linear, so	COMMENT
MEDIUM	tests/test_routes_models.py	102	# Both cards must be present — the chat model AND the embedding model.	COMMENT
MEDIUM	tests/test_streaming_newlines.py	303	# The reasoning trace from inside ``<think>...</think>`` MUST	COMMENT
MEDIUM	tests/test_agent_query_context_skip.py	81	# ~100 cols so the phrase "context window" ends up split as	COMMENT
MEDIUM⚡	tests/test_agent_query_context_skip.py	131	initial #659 fix used a literal ``"context window" in output``	STRING
MEDIUM	tests/test_silent_drop_rescue_569.py	540	# The rescued content is the accumulated reasoning trace.	COMMENT
MEDIUM	tests/test_silent_drop_rescue_569.py	943	# them should carry the reasoning trace. Per-delta reasoning	COMMENT
MEDIUM	tests/test_doctor_runner.py	220	# The doctor CLI no longer owns server-boot orchestration; that moved to	COMMENT
MEDIUM	tests/test_release_check_random.py	48	# Excluded: the 4 B tier cannot hold an agentic loop. It does not	COMMENT
MEDIUM	tests/test_chat_template_tool_call_arguments.py	296	env = jinja2.Environment( # noqa: S701 - test scaffolding	CODE
MEDIUM	tests/test_chat_logprobs_channel_routing.py	189	# #569: rescue fires — content carries the reasoning trace so the	COMMENT
MEDIUM	tests/test_hermes_malformed_tool_feedback.py	48	# The orchestration layer must reject this before dispatch and return its	COMMENT
MEDIUM⚡	tests/test_mtp_gemma4_assistant_inject.py	337	# SIGNATURE contract, and unlike the PR #989 scaffold path, the	COMMENT
MEDIUM⚡	tests/test_mtp_gemma4_assistant_inject.py	758	# make_mtp_cache returns a list from the inner scaffold — assert	COMMENT
MEDIUM	tests/test_tool_choice_enforcement.py	184	# No-fake-API clause names the most common hallucination shapes.	COMMENT
MEDIUM⚡	tests/test_api_validation_bundle.py	472	# - C3: chat rejects image/video on text-only models (no silent hallucination)	COMMENT
MEDIUM	tests/test_api_validation_bundle.py	270	# H-09 route guard rejects requests when no embedding model is	COMMENT
MEDIUM	tests/integrations/test_agents_matrix.py	10	Existing wire cells (from #1030 scaffold):	STRING
MEDIUM	tests/integrations/agent_smoke.sh	38	# reasoning: cold (agentic tool/long-context kernels still compiling) it runs	COMMENT
MEDIUM	tests/integrations/agent_smoke.sh	208	# model — with thinking on it emits a chain-of-thought whose length is highly	COMMENT
MEDIUM	tests/integrations/agent_smoke.sh	214	# verify the end-to-end agentic tool-calling path (serve → tool schema → agent	COMMENT
MEDIUM	tests/integrations/agent_smoke.sh	538	# agentic run above (SERVE_PID still listening), so there is NO second model load.	COMMENT
MEDIUM	tests/integrations/agent_smoke.sh	540	# This closes the L2 gap where the release gate proved the agentic path but never	COMMENT
MEDIUM⚡	tests/integrations/test_hermes.py	160	# context window is below what Hermes's full tool-rich prompt	COMMENT
MEDIUM⚡	tests/integrations/test_hermes.py	167	# literal ``"context window"`` substring would miss when	COMMENT
MEDIUM⚡	tests/integrations/test_hermes.py	575	# Deep agentic tests (requires hermes binary, tests real workflows)	COMMENT
MEDIUM	docs/getting-started/quickstart.md	12	rapid-mlx chat --think # surface chain-of-thought reasoning	CODE
MEDIUM	docs/guides/embeddings.md	17	### Start the server with an embedding model	COMMENT
122 more matches not shown…

Deep Nesting561 hits · 404 pts

Severity	File	Line	Context
LOW	bench/bench_spec_decode_mtp.py	612	CODE
LOW	evals/run_eval.py	331	CODE
LOW	evals/run_eval.py	474	CODE
LOW	evals/run_eval.py	636	CODE
LOW	evals/run_eval.py	683	CODE
LOW	evals/run_eval.py	722	CODE
LOW	evals/run_eval.py	793	CODE
LOW	evals/run_eval.py	1307	CODE
LOW	evals/run_eval.py	1707	CODE
LOW	evals/run_eval.py	1872	CODE
LOW	evals/generate_scorecard.py	84	CODE
LOW	evals/perf_gate.py	213	CODE
LOW	community-benchmarks/scripts/aggregate.py	193	CODE
LOW	tests/test_no_mllm_flag.py	969	CODE
LOW	tests/test_no_mllm_flag.py	1087	CODE
LOW	tests/test_no_mllm_flag.py	1358	CODE
LOW	tests/test_no_mllm_flag.py	1588	CODE
LOW	tests/test_no_mllm_flag.py	1882	CODE
LOW	tests/test_no_mllm_flag.py	2126	CODE
LOW	tests/test_docs_cli_defaults.py	37	CODE
LOW	tests/test_docs_cli_defaults.py	51	CODE
LOW	tests/test_reasoning_content_null_rescue.py	1435	CODE
LOW	tests/test_reasoning_content_null_rescue.py	1918	CODE
LOW	tests/test_pflash_server_module_wiring.py	81	CODE
LOW	tests/test_stt_corrupted_file.py	184	CODE
LOW	tests/test_no_out_of_band_routing.py	719	CODE
LOW	tests/test_no_out_of_band_routing.py	875	CODE
LOW	tests/test_no_out_of_band_routing.py	1002	CODE
LOW	tests/test_no_out_of_band_routing.py	1157	CODE
LOW	tests/test_ui_tars_fixes.py	718	CODE
LOW	tests/test_serve_listen_fd.py	145	CODE
LOW	tests/test_responses_engine_failure_envelope.py	986	CODE
LOW	tests/test_responses_engine_failure_envelope.py	1407	CODE
LOW	tests/test_ci_lane_promotion.py	61	CODE
LOW	tests/regression_suite.py	44	CODE
LOW	tests/test_anthropic_stop_sequences.py	245	CODE
LOW	tests/test_anthropic_stop_sequences.py	296	CODE
LOW	tests/test_ubc_evict.py	83	CODE
LOW	tests/test_responses_sse_event_order.py	170	CODE
LOW	tests/test_fake_sidecar_image_catalog.py	273	CODE
LOW	tests/test_anthropic_streaming_reasoning.py	77	CODE
LOW	tests/test_anthropic_streaming_reasoning.py	111	CODE
LOW	tests/test_truncation_no_synthetic_text.py	235	CODE
LOW	tests/test_truncation_no_synthetic_text.py	352	CODE
LOW	tests/test_truncation_no_synthetic_text.py	456	CODE
LOW	tests/test_response_format_json_schema_strict.py	608	CODE
LOW	tests/test_community_bench.py	1311	CODE
LOW	tests/test_community_bench.py	1335	CODE
LOW	tests/test_r12_h06_repair_context_guard.py	101	CODE
LOW	tests/test_cohere_command_reasoning_parser.py	407	CODE
LOW	tests/test_gdn_prefill_kernel.py	168	CODE
LOW	tests/test_fp8_repack.py	125	CODE
LOW	tests/test_prefix_cache_persistence.py	1249	CODE
LOW	tests/test_prefix_cache_persistence.py	1270	CODE
LOW	tests/test_responses_591_followups.py	253	CODE
LOW	tests/test_pr_validate_stress_timeout.py	108	CODE
LOW	tests/test_responses_budget_exhaust_streaming.py	627	CODE
LOW	tests/test_responses_budget_exhaust_streaming.py	926	CODE
LOW	tests/test_streaming_latency.py	28	CODE
LOW	…ts/test_r12_m3_responses_stream_leading_items_order.py	232	CODE
501 more matches not shown…

Cross-Language Confusion71 hits · 351 pts

Severity	File	Line	Snippet	Context
HIGH	evals/coherence_gate.py	108	f"assistant content must be a string or null, got {type(content).__name__}"	CODE
HIGH⚡	tests/test_gemma4_grammar_558.py	1009	"code", ["a < b && c > d", "vector<int> v", "obj = {x:1}", "print('ok')"]	CODE
HIGH	tests/test_gemma4_grammar_558.py	796	"a < b && c > d",	CODE
HIGH	tests/test_reasoning_content_null_rescue.py	1023	strict-null, but the sentinel-absence assertion is what	STRING
HIGH	tests/test_envelope_field_extraction.py	11	"param":null,"code":null}}	STRING
HIGH	tests/test_no_pydantic_error_leak.py	15	"v/missing","type":"invalid_request_error","code":null,	STRING
HIGH	tests/test_no_pydantic_error_leak.py	16	"param":null}}	STRING
HIGH	tests/test_r10_scrub_validation_bundle.py	32	``reasoning_effort`` accepted any value (int, list, null,	STRING
HIGH	tests/regression_suite.py	486	# every leaf in the schema (qty must be int\|null, no extra keys,	COMMENT
HIGH	tests/test_stream_include_usage_honored.py	205	f"chunk (not just emit null); got {len(chunks_with_usage_key)} "	CODE
HIGH	tests/test_stream_include_usage_honored.py	233	f"chunk (not just emit null); got {len(chunks_with_usage_key)} "	CODE
HIGH	tests/test_stream_include_usage_honored.py	304	f"every chunk (not just emit null); got "	CODE
HIGH	tests/test_stream_include_usage_honored.py	332	f"KEY from every chunk (not just emit null); got "	CODE
HIGH⚡	tests/test_stream_include_usage_honored.py	182	be absent from every chunk (not just present-but-null). Codex	STRING
HIGH	tests/test_residency_load_field_names.py	9	dtype and TurboQuant are mutually exclusive","param":null}}	STRING
HIGH	tests/test_check_desktop_publish.py	250	echo ' {{"id": 11, "event": "workflow_dispatch", "head_sha": "{ACCEPTED}", "head_branch": "rapid-mac-v0.13.0-	CODE
HIGH	tests/test_finalize_harmony_raw_text.py	1009	spec — the tool call IS the visible response, ``content`` is null).	STRING
HIGH	tests/test_streaming_tool_filter.py	305	'<｜DSML｜parameter name="cmd" string="true">pwd && ls',	CODE
HIGH	tests/test_routes_models_effective_parsers.py	470	f"entry's live ``tool_call_parser=None`` must surface as null; "	CODE
HIGH	tests/test_routes_models_effective_parsers.py	474	f"entry's live ``reasoning_parser=None`` must surface as null; "	CODE
HIGH⚡	tests/test_gui_walk_completeness.py	16	assert "elementWalkComplete && windowListComplete" in source	CODE
HIGH	tests/test_api_models.py	1516	# D-MISSING-CONTENT-KEY (r12-7): empty string, not null, so the	COMMENT
HIGH	tests/test_api_models.py	169	carrying ``{"role":"user","content":null}`` (or "system" /	STRING
HIGH⚡	tests/test_openai_tool_choice_validation.py	234	"""``tool_choice=null`` (JSON null) — same semantics as	STRING
HIGH	tests/test_chat_template_registry.py	468	"arguments": '{"command":"ls /tmp","limit":null}',	CODE
HIGH⚡	tests/test_responses_bundle.py	397	'<｜DSML｜parameter name="cmd" string="true">pwd && ls',	CODE
HIGH	tests/test_responses_bundle.py	452	assert json.loads(argument_deltas) == {"cmd": "pwd && ls"}	CODE
HIGH	tests/test_responses_bundle.py	463	assert json.loads(completed_calls[0]["arguments"]) == {"cmd": "pwd && ls"}	CODE
HIGH	tests/test_kv_estimation.py	429	"sliding_attention", # 1 — window UNKNOWN (null)	CODE
HIGH⚡	tests/test_version_check.py	972	naive ``curl A \|\| curl B \| bash`` that returns bash's status and would	STRING
HIGH⚡	tests/test_lfm_tool_parser.py	549	'"cursor": null, "n": 3})]',	CODE
HIGH⚡	tests/test_lfm_tool_parser.py	550	'[browse({"refresh": true, "cached": false, "cursor": null, "n": 3})]',	CODE
HIGH	tests/test_model_sizes.py	125	'{"sizes": null}', # sizes present but not a dict	CODE
HIGH⚡	tests/test_qwen3coder_xml_grammar_558.py	671	@pytest.mark.parametrize("code", ["a < b && c > d", "vector<int> v", "print('ok')"])	CODE
HIGH⚡	tests/test_mcp_security.py	70	validator.validate_command("npx && rm -rf /", "test-server")	CODE
HIGH⚡	tests/test_mcp_security.py	66	"""Test that command injection via && is blocked."""	STRING
HIGH	tests/test_ram_tier_recommendations_agree.py	508	"!quickstart.done && QuickstartCoordinator.retiredStarters.contains" in caller	CODE
HIGH	tests/test_audio_r11_b_pure.py	161	"""``{"format": null}`` is the JSON shape an SDK might emit when	STRING
HIGH⚡	tests/test_casual_chat_auto_disable_thinking.py	438	"""Codex r1 MEDIUM #3: ``reasoning={"effort": null}`` is	STRING
HIGH⚡	tests/test_casual_chat_auto_disable_thinking.py	443	round-trips ``reasoning={"effort": null}`` (e.g. an SDK that	STRING
HIGH	tests/test_guided.py	815	for t in inner.encode('a","tag":null}'):	CODE
HIGH	tests/test_guided.py	826	produced = json.loads('{"inner":{"x":1},"kind":"a","tag":null}')	CODE
HIGH⚡	tests/test_tool_param_enforcement.py	203	_validate_tool_call_params([_call("nick", '{"name": null}')], tools)	CODE
HIGH	tests/test_bench_vs_ollama.py	142	'data: {"choices":[null,{"delta":null},{"delta":[]}]}',	CODE
HIGH	tests/test_bench_vs_ollama.py	175	'{"message":null}',	CODE
HIGH	tests/test_vision_extra_install.py	322	f"`pip install rapid-mlx && rapid-mlx serve gemma-4-12b-4bit` "	CODE
HIGH	tests/test_vision_extra_install.py	489	"`pip install rapid-mlx && rapid-mlx serve gemma-4-12b-4bit` "	CODE
HIGH	tests/test_vision_extra_install.py	13	``pip install rapid-mlx==0.10.0 && rapid-mlx serve gemma-4-12b-4bit``	STRING
HIGH	scripts/agent_promo_gen.py	154	pip install rapid-mlx && rapid-mlx serve <model>	CODE
HIGH	vllm_mlx/_download_gate.py	1262	# A non-string value (list, number, null) is enough to make	COMMENT
HIGH⚡	vllm_mlx/kv_quant_gate.py	166	mean_kl: Mean forward KL ``D(P_baseline \|\| P_candidate)`` in nats over	STRING
HIGH	vllm_mlx/kv_quant_gate.py	323	null). Used to require the candidate to preserve the baseline's structure.	STRING
HIGH	vllm_mlx/tool_parsers/muse_tool_parser.py	325	* ``null`` -> None only when the schema explicitly permits null;	STRING
HIGH	vllm_mlx/tool_parsers/gemma4_tool_parser.py	441	# Try to parse as JSON literal (int, float, bool, null). Catch	COMMENT
HIGH	vllm_mlx/middleware/body_size.py	579	"code": "request_too_large", "param": null}}	STRING
HIGH	vllm_mlx/mcp/types.py	25	``{"mcpServers": null}``) raises rather than silently falling back.	STRING
HIGH	vllm_mlx/models/gemma4_text.py	591	# * ``None`` reaching here means the key was ABSENT (not explicit null,	COMMENT
HIGH	vllm_mlx/models/gemma4_text.py	611	# ``TypeError`` on a ``None`` that reached it via an explicit config null).	COMMENT
HIGH	vllm_mlx/models/gemma4_vendored/__init__.py	8	lean; but a fresh `pip install rapid-mlx==0.10.0 && rapid-mlx serve	STRING
HIGH	vllm_mlx/spec_decode/mtp/quantized_argmax.py	179	if (simd_gid == 0 && simd_lid == 0) {	CODE
11 more matches not shown…

Modern Structural Boilerplate176 hits · 176 pts

Severity	File	Line	Snippet	Context
LOW	tests/integrations/conftest.py	809	__all__ = [	CODE
LOW	videox_fun_mlx/pipeline/scheduler.py	83	def set_timesteps(self, num_inference_steps: int) -> None:	CODE
LOW	examples/video_benchmark.py	31	logger = logging.getLogger(__name__)	CODE
LOW	examples/test_video.py	24	logger = logging.getLogger(__name__)	CODE
LOW	scripts/extract_hy3_mtp.py	47	logger = logging.getLogger(__name__)	CODE
LOW	scripts/extract_mtp_weights.py	32	logger = logging.getLogger(__name__)	CODE
LOW	scripts/pr_validate/steps/codex_review.py	254	1. [BLOCKING] vllm_mlx/routes/chat.py:918 — `assert isinstance(_msg, dict)` is stripped under `python -O`, leaving the g	CODE
LOW	vllm_mlx/_tempfile_safe.py	365	__all__ = ["managed_tempfile_path"]	CODE
LOW	vllm_mlx/mllm_cache.py	31	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/_mlx_compat.py	47	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/memory_cache.py	39	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/server.py	167	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/server.py	2563	def _set_resident_primary(entry: ModelEntry \| None) -> None:	CODE
LOW	vllm_mlx/vision_embedding_cache.py	26	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/output_router.py	37	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/gdn_prefill.py	43	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/embedding.py	15	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/coherence.py	38	__all__ = [	CODE
LOW	vllm_mlx/singleton_cache_fastpath.py	76	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/request.py	253	def set_finished(self, status: RequestStatus, reason: str \| None = None) -> None:	CODE
LOW	vllm_mlx/turboquant.py	52	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/multimodal_processor.py	25	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/__init__.py	99	__all__ = [	CODE
LOW⚡	vllm_mlx/kv_cache_dtype.py	46	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/optimizations.py	30	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/_parent_watchdog.py	61	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/output_router_harmony.py	58	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/engine_core.py	34	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/fp8_repack.py	56	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/pflash.py	33	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/mllm_scheduler.py	58	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/mllm_scheduler.py	679	def set_generation_paused(self, paused: bool, *, add_allowance: int = 0) -> None:	CODE
LOW	vllm_mlx/paged_cache.py	34	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/model_auto_config.py	34	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/_signal_observability.py	56	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/prefix_cache.py	29	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/scheduler.py	136	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/scheduler.py	6114	def set_generation_paused(self, paused: bool, *, add_allowance: int = 0) -> None:	CODE
LOW	vllm_mlx/tool_call_scan.py	39	__all__ = [	CODE
LOW	vllm_mlx/gdn_in_proj_fusion.py	51	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/_mxfp4_moe_guardrail.py	76	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/model_registry.py	21	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/mllm_batch_generator.py	49	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/moe_fusion.py	37	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/tool_parsers/lfm_tool_parser.py	33	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/tool_parsers/deepseek_v3_tool_parser.py	74	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/tool_parsers/qwen3coder_tool_parser.py	42	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/tool_parsers/__init__.py	82	__all__ = [	CODE
LOW	vllm_mlx/tool_parsers/seed_oss_tool_parser.py	32	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/tool_parsers/deepseekv31_tool_parser.py	52	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/tool_parsers/nemotron_tool_parser.py	32	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/tool_parsers/ui_tars_tool_parser.py	63	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/middleware/auth.py	20	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/middleware/probe_fastpath.py	67	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/middleware/body_size.py	51	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/middleware/body_depth.py	56	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/middleware/request_logging.py	30	logger = logging.getLogger(__name__)	CODE
LOW⚡	vllm_mlx/kernels/turboquant_fused.py	28	logger = logging.getLogger(__name__)	CODE
LOW	vllm_mlx/bench/_server.py	242	__all__ = ["serve", "find_free_port", "ServerStartFailed"]	CODE
LOW	vllm_mlx/video/__init__.py	6	__all__ = ["VideoGenerationEngine", "WanVideoEngine"]	CODE
116 more matches not shown…

Cross-File Repetition21 hits · 105 pts

Severity	File	Snippet	Context
HIGH	tests/test_forced_alignment_route_hardening.py	mount the audio router on a bare fastapi app, bypassing auth.	STRING
HIGH	tests/test_audio_routes_bundle.py	mount the audio router on a bare fastapi app, bypassing auth.	STRING
HIGH	tests/test_stt_response_format.py	mount the audio router on a bare fastapi app, bypassing auth.	STRING
HIGH	tests/test_audio_music_route.py	mount the audio router on a bare fastapi app, bypassing auth.	STRING
HIGH	tests/test_chatterbox_expressiveness.py	minimal fake so the tts-lane probe passes without the real extra.	STRING
HIGH	tests/test_speech_inline_clone_route.py	minimal fake so the tts-lane probe passes without the real extra.	STRING
HIGH	tests/test_qwen3_tts_audio.py	minimal fake so the tts-lane probe passes without the real extra.	STRING
HIGH	tests/test_qwen3_tts_voicedesign.py	minimal fake so the tts-lane probe passes without the real extra.	STRING
HIGH	tests/test_stop_and_timeout_request_schema.py	patch the global config singleton and restore on teardown.	STRING
HIGH	tests/test_sampling_param_finite_range.py	patch the global config singleton and restore on teardown.	STRING
HIGH	tests/test_sampling_validation.py	patch the global config singleton and restore on teardown.	STRING
HIGH	tests/integrations/test_hermes.py	streaming mode: tool calls arrive as structured deltas.	STRING
HIGH	vllm_mlx/_integration_tests/test_hermes.py	streaming mode: tool calls arrive as structured deltas.	STRING
HIGH	vllm_mlx/agents/testing.py	streaming mode: tool calls arrive as structured deltas.	STRING
HIGH	vllm_mlx/_mxfp4_moe_guardrail.py	test-only hook: zero the counters between cases. production code must not call this — prometheus counters are contractua	STRING
HIGH	vllm_mlx/runtime/ubc_evict.py	test-only hook: zero the counters between cases. production code must not call this — prometheus counters are contractua	STRING
HIGH	vllm_mlx/api/response_format_metrics.py	test-only hook: zero the counters between cases. production code must not call this — prometheus counters are contractua	STRING
HIGH	…s/RapidTests/DeferredTelemetryConsentWiringTests.swift	data: {"choices":[{"delta":{"content":"ok"},"finish_reason":"stop"}]}\n data: [done]\n	STRING
HIGH	…-mac/Tests/RapidTests/ChatStreamRequestBodyTests.swift	data: {"choices":[{"delta":{"content":"ok"},"finish_reason":"stop"}]}\n data: [done]\n	STRING
HIGH	…pid-mac/Tests/RapidTests/CustomInstructionsTests.swift	data: {"choices":[{"delta":{"content":"ok"},"finish_reason":"stop"}]}\n data: [done]\n	STRING
HIGH	…mac/Tests/RapidTests/CurrentDateTimeContextTests.swift	data: {"choices":[{"delta":{"content":"ok"},"finish_reason":"stop"}]}\n data: [done]\n	STRING

Docstring Block Structure17 hits · 85 pts

Severity	File	Line	Snippet	Context
HIGH	vllm_mlx/embedding.py	287	Generate embeddings for one or more texts. Args: texts: A single string or list of strings	STRING
HIGH	vllm_mlx/embedding.py	352	Embed pre-tokenized inputs (OpenAI spec input formats 3 and 4). Skips the tokenizer entirely — the caller has a	STRING
HIGH	vllm_mlx/paged_cache.py	276	Pop n non-pinned blocks from the front. Pinned blocks are skipped to prevent eviction of important	STRING
HIGH	vllm_mlx/paged_cache.py	653	Allocate multiple blocks at once (vLLM style). Args: num_blocks: Number of blocks to alloc	STRING
HIGH	vllm_mlx/model_registry.py	58	Attempt to acquire ownership of a model. Args: model: The MLX model engine: Th	STRING
HIGH	vllm_mlx/tool_parsers/abstract_tool_parser.py	512	Retrieve a registered ToolParser class by name. Args: name: Parser name (e.g., 'mistral',	STRING
HIGH	vllm_mlx/runtime/model_registry.py	109	Get the engine for a model name. Falls back to default. Args: model_name: Model name from request.	STRING
HIGH	vllm_mlx/mcp/config.py	41	Load MCP configuration from file. Search order: 1. Explicit path argument 2. RAPID_MLX_MCP_CONFIG envi	STRING
HIGH	vllm_mlx/mcp/config.py	127	Validate and parse configuration dictionary. Args: data: Raw configuration dictionary tolerant	STRING
HIGH	vllm_mlx/mcp/tools.py	50	Parse OpenAI tool call back to MCP format. Args: tool_call: OpenAI tool call from model response	STRING
HIGH	vllm_mlx/models/mllm.py	502	Decode base64 image to bytes. Args: base64_string: Base64 encoded image (optionally with data URL pref	STRING
HIGH	vllm_mlx/models/mllm.py	858	Download image from URL and return local path. Args: url: Image URL timeout: Download timeout	STRING
HIGH	vllm_mlx/models/mllm.py	950	Download video from URL and return local path. Args: url: Video URL (http/https) timeout: Down	STRING
HIGH	vllm_mlx/models/mllm.py	1053	Decode base64 video to temp file and return path. Supports format: data:video/mp4;base64,AAAA... Args:	STRING
HIGH	vllm_mlx/reasoning/__init__.py	51	Get a reasoning parser class by name. Args: name: Name of the parser (e.g., "qwen3", "deepseek_r1").	STRING
HIGH	vllm_mlx/spec_decode/mtp/qwen3_5_inject.py	403	Inject MTP support into a loaded Qwen3.5 / Qwen3.6 model. Args: model: A model loaded via ``mlx_lm.load()``	STRING
HIGH	vllm_mlx/audio/music.py	142	Generate ``seconds`` of audio for ``prompt`` → ``out_path`` (wav). Args: prompt: Natural-language d	STRING

Self-Referential Comments28 hits · 80 pts

Severity	File	Line	Snippet	Context
MEDIUM	tests/test_tool_logits.py	126	# Create a token ID that corresponds to the last char of '<invoke'	COMMENT
MEDIUM	tests/test_tool_parsers.py	8	# This file is the Linux parser-matrix entry point. Import the focused	COMMENT
MEDIUM	tests/test_server.py	373	# Create a real temp file	COMMENT
MEDIUM	tests/test_fix_reasoning_mtp_doctor.py	97	# Create a small QuantizedSwitchLinear	COMMENT
MEDIUM	tests/test_memory_stability.py	41	# Create a mock BatchGenerator with close()	COMMENT
MEDIUM	tests/test_memory_stability.py	172	# Create a mock request with extracted cache (dict-state format)	COMMENT
MEDIUM	tests/test_memory_stability.py	219	# Create a mock response with prompt_cache	COMMENT
MEDIUM	tests/test_turboquant.py	544	# Create a mix: KVCache + non-KVCache	COMMENT
MEDIUM	tests/test_mtp_gemma4_assistant_inject.py	1238	# Create the well-known name AND an "extra" shard.	COMMENT
MEDIUM	tests/test_mllm_continuous_batching.py	818	# Create a request with stop sequences	COMMENT
MEDIUM	tests/test_mllm_cache.py	77	# Create a temp file with some content	COMMENT
MEDIUM	tests/integrations/test_hermes.py	670	# Create a test file first	COMMENT
MEDIUM	examples/test_openai_compatibility.py	63	# Create a simple 100x100 red square image	COMMENT
MEDIUM	examples/test_openai_compatibility.py	490	# Create a simple video with 3 colored frames (red, green, blue)	COMMENT
MEDIUM	examples/test_video.py	68	# Create a simple video with colored frames	COMMENT
MEDIUM	examples/demo_openai_image.py	80	# Create a simple red square image for testing	COMMENT
MEDIUM	examples/demo_openai_image.py	86	# Create a simple 100x100 red image	COMMENT
MEDIUM	examples/benchmark_audio.py	42	# Create a simple sine wave tone	COMMENT
MEDIUM	examples/demo_curl_image.sh	77	# Create a simple red PNG using Python (smallest valid PNG)	COMMENT
MEDIUM	scripts/create_release.sh	3	# Create a GitHub Release whose tag we atomically claim at $RELEASE_SHA.	COMMENT
MEDIUM	scripts/pr_validate/steps/targeted_tests.py	253	# Create a worktree pointing at base_ref (sha-pinned when given).	COMMENT
MEDIUM	vllm_mlx/gdn_prefill.py	202	# This function is the single validation authority for the kernel's	STRING
MEDIUM	vllm_mlx/gradio_app.py	298	# Create a public share link	COMMENT
MEDIUM	vllm_mlx/cli.py	5449	# Create the single mlx-step worker BEFORE loading weights and reuse it	COMMENT
MEDIUM	vllm_mlx/mllm_batch_generator.py	1389	# Create a fresh KVCache for this request's language model prefill	COMMENT
MEDIUM	vllm_mlx/_integration_tests/test_hermes.py	670	# Create a test file first	COMMENT
MEDIUM	vllm_mlx/api/guided.py	202	# Create the model dynamically	COMMENT
MEDIUM	vllm_mlx/speculative/dflash/server.py	1525	# Create the generator on the same worker that will drive it,	COMMENT

Redundant / Tautological Comments53 hits · 78 pts

Severity	File	Line	Snippet	Context
LOW	evals/run_eval.py	950	# Check if model called the recovery tool	COMMENT
LOW	tests/test_paged_cache_benefits.py	313	# Check if root is shared	COMMENT
LOW	tests/test_paged_cache.py	176	# Increment ref	COMMENT
LOW	examples/video_benchmark.py	577	# Print results	STRING
LOW	examples/mcp_tool_use.py	97	# Check if model wants to use tools	COMMENT
LOW	examples/mic_realtime.py	111	# Check if audio has content (not silence)	COMMENT
LOW	examples/test_openai_compatibility.py	805	# Check if server is reachable	STRING
LOW	examples/mllm_benchmark.py	451	# Print results	STRING
LOW	examples/mic_live.py	130	# Check if silence long enough to trigger transcription	COMMENT
LOW	examples/demo_openai_video.py	110	# Check if there's a sample video in the examples directory	COMMENT
LOW	scripts/add_mtp_weights.py	296	# Check if MTP weights already exist	COMMENT
LOW	scripts/benchmark_engines.py	500	# Check if tool call was returned in structured format	COMMENT
LOW	scripts/benchmark_engines.py	502	# Check if content contains text-format tool calls (degraded)	COMMENT
LOW	scripts/agent_discovery.py	347	# Check if it's actually an agent	COMMENT
LOW	scripts/agent_discovery.py	385	# Check if title mentions AI agents	COMMENT
LOW	vllm_mlx/memory_cache.py	1994	# Check if cached_key is a prefix of tokens_key	COMMENT
LOW	vllm_mlx/memory_cache.py	2016	# Check if tokens_key is a prefix of cached_key	COMMENT
LOW	vllm_mlx/memory_cache.py	2266	# Check if single entry exceeds limit	COMMENT
LOW	vllm_mlx/memory_cache.py	3736	# Check if it fits against the running (live+staged in	COMMENT
LOW	vllm_mlx/gradio_app.py	160	# Check if this message had media cached	COMMENT
LOW	vllm_mlx/cli.py	4382	# Set SO_REUSEADDR to match uvicorn's bind behavior — without it, this	COMMENT
LOW	vllm_mlx/mllm_scheduler.py	1131	# Check if finished	COMMENT
LOW	vllm_mlx/prefix_cache.py	154	# Check if we have a shorter prefix with cache	COMMENT
LOW	vllm_mlx/prefix_cache.py	223	# Check if cache supports trimming	COMMENT
LOW	vllm_mlx/prefix_cache.py	615	# Check if cache_data contains extracted tensor states	COMMENT
LOW	vllm_mlx/prefix_cache.py	649	# Check if this block already exists (deduplication)	COMMENT
LOW	vllm_mlx/scheduler.py	4239	# Check if it's a list of cache layers	COMMENT
LOW	vllm_mlx/scheduler.py	4247	# Check if layer has expected structure	COMMENT
LOW	vllm_mlx/scheduler.py	7286	# Check if finished	COMMENT
LOW	vllm_mlx/mllm_batch_generator.py	632	# Check if this is actually a VLM with separate language model	COMMENT
LOW	vllm_mlx/tool_parsers/qwen3coder_tool_parser.py	935	# Check if we need to advance to next tool. The tool boundary is	COMMENT
LOW	vllm_mlx/tool_parsers/granite_tool_parser.py	62	# Check if it starts with JSON array	COMMENT
LOW	vllm_mlx/tool_parsers/seed_oss_tool_parser.py	294	# Check if we need to advance to next tool	COMMENT
LOW	vllm_mlx/tool_parsers/seed_oss_tool_parser.py	309	# Check if thinking ended (or never started)	COMMENT
LOW	vllm_mlx/tool_parsers/gemma4_tool_parser.py	892	# Check if we're inside a tool call. Either the pristine wire	COMMENT
LOW	vllm_mlx/tool_parsers/gemma4_tool_parser.py	943	# Check if we have a complete text tool call	COMMENT
LOW	vllm_mlx/utils/tokenizer.py	1534	# Check if model needs fallback (e.g., Nemotron)	COMMENT
LOW	vllm_mlx/mcp/security.py	154	# Check if command is in whitelist	COMMENT
LOW	vllm_mlx/mcp/security.py	156	# Check if it's an absolute path to an allowed command	COMMENT
LOW	vllm_mlx/mcp/executor.py	459	# Check if tool exists	COMMENT
LOW⚡	vllm_mlx/models/mllm.py	1125	# Check if it's a local file (confined to regular media files; see	COMMENT
LOW⚡	vllm_mlx/models/mllm.py	1131	# Check if it's a URL	COMMENT
LOW⚡	vllm_mlx/models/mllm.py	1135	# Check if it's base64	COMMENT
LOW	vllm_mlx/models/mllm.py	1214	# Check if it's base64 FIRST (before Path.exists() which fails on long strings)	COMMENT
LOW	vllm_mlx/models/mllm.py	1218	# Check if it's a URL	COMMENT
LOW	vllm_mlx/models/deepseek_v4_cache.py	801	# Check if all caches are empty	COMMENT
LOW	vllm_mlx/reasoning/deepseek_r1_parser.py	120	# Check if any tags are in the current text	COMMENT
LOW	vllm_mlx/reasoning/minimax_parser.py	173	# Check if output starts with reasoning patterns	COMMENT
LOW⚡	vllm_mlx/reasoning/gemma4_parser.py	321	# Check if we just entered thought channel	COMMENT
LOW⚡	vllm_mlx/reasoning/gemma4_parser.py	326	# Check if we just entered content channel	COMMENT
LOW⚡	vllm_mlx/reasoning/gemma4_parser.py	331	# Check if thought ended (first <channel\|> after thought start)	COMMENT
LOW	vllm_mlx/api/tool_logits.py	353	# Check if we should start tracking a pattern	COMMENT
LOW	vllm_mlx/routes/chat.py	6312	# Check if we should include usage in the final chunk	COMMENT

Structural Annotation Overuse34 hits · 63 pts

Severity	File	Line	Snippet	Context
LOW⚡	tests/test_llama_tool_parser_bare_json.py	351	# Step 1: prose preface — passes through.	COMMENT
LOW⚡	tests/test_llama_tool_parser_bare_json.py	358	# Step 2: opening brace + name/params arrive in one go.	COMMENT
LOW⚡	tests/test_llama_tool_parser_bare_json.py	375	# Step 1: closed prose JSON — flushed as content on close.	COMMENT
LOW⚡	tests/test_llama_tool_parser_bare_json.py	384	# Step 2: trailing prose — must pass through.	COMMENT
LOW	tests/test_ui_tars_fixes.py	740	# Step 1: walk imports to learn what local names refer to	COMMENT
LOW	tests/test_ui_tars_fixes.py	758	# Step 2: walk the tree looking for Call nodes whose callable	COMMENT
LOW⚡	tests/test_metrics_route.py	531	# Step 1: hits=10 → exposed 10	COMMENT
LOW⚡	tests/test_metrics_route.py	535	# Step 2: hits=15 → exposed 15	COMMENT
LOW⚡	tests/test_metrics_route.py	540	# Step 3: simulate cache.clear() — raw drops to 0; exposed must stay ≥ 15	COMMENT
LOW⚡	tests/test_metrics_route.py	549	# Step 4: hits=3 — fresh activity after clear; exposed should be 15+3=18	COMMENT
LOW⚡	tests/test_reasoning_budget_generation.py	191	# Step 2: one generated think token → 1/1 spent → force on THIS call.	COMMENT
LOW⚡	tests/test_grammar_processor_558.py	2395	# Step 1: baseline the prompt (1 token), then feed the first generated	COMMENT
LOW⚡	tests/test_grammar_processor_558.py	2405	# Step 2: feed the second generated token — the matcher rejects it, so	COMMENT
LOW	tests/test_grammar_processor_558.py	2418	# Step 3: any subsequent call also stays dropped (logits unchanged).	COMMENT
LOW	docs/guides/server.md	285	# Step 1: Send request with tools	COMMENT
LOW	docs/guides/server.md	301	# Step 2: Check if model wants to use tools	COMMENT
LOW	docs/guides/server.md	307	# Step 3: Send tool result back	COMMENT
LOW	vllm_mlx/tool_parsers/gemma4_tool_parser.py	428	# Step 2: bare KV parse	COMMENT
LOW	vllm_mlx/utils/chat_template.py	1546	# Step 1: retry without enable_thinking (many templates don't support it).	COMMENT
LOW	vllm_mlx/utils/chat_template.py	1590	# Step 2: template also rejects tools — fall back to prompt injection.	COMMENT
LOW	vllm_mlx/service/postprocessor.py	2654	# Step 1: Separate content from reasoning	COMMENT
LOW	vllm_mlx/share/cli.py	341	# Step 1: a deliberately-wrong key must NOT return 200. If the	COMMENT
LOW	vllm_mlx/share/cli.py	352	# Step 2: the real key must return 200.	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	377	// Step 1: cached + safe (only over the known-params subset).	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	382	// Step 2: smallest known-params alias that fits.	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	390	// Step 3: smallest known-params alias overall (.tooBig but	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	395	// Step 4: catalog has zero parseable aliases. Last resort —	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	509	// Step 1: bucketed default is on disk AND runnable. No	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	515	// Step 2: any cached + fits alternative — prefer it over a	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	528	// Step 3: nothing cached fits, but the canonical pick still	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	534	// Step 4: bucketed is .tooBig OR missing — hand off to the	COMMENT
LOW⚡	…rapid-mac/Sources/Rapid/Server/ToolUseCapability.swift	493	// Step 1: denylist match (broken takes priority).	COMMENT
LOW⚡	…rapid-mac/Sources/Rapid/Server/ToolUseCapability.swift	498	// Step 2: deliberate-unknown overrides shadow the family match	COMMENT
LOW	…rapid-mac/Sources/Rapid/Server/ToolUseCapability.swift	512	// Step 3: family + size guard. An alias is .known when its	COMMENT

Verbosity Indicators32 hits · 61 pts

Severity	File	Line	Snippet	Context
LOW⚡	tests/test_llama_tool_parser_bare_json.py	351	# Step 1: prose preface — passes through.	COMMENT
LOW⚡	tests/test_llama_tool_parser_bare_json.py	358	# Step 2: opening brace + name/params arrive in one go.	COMMENT
LOW⚡	tests/test_llama_tool_parser_bare_json.py	375	# Step 1: closed prose JSON — flushed as content on close.	COMMENT
LOW⚡	tests/test_llama_tool_parser_bare_json.py	384	# Step 2: trailing prose — must pass through.	COMMENT
LOW	tests/test_ui_tars_fixes.py	740	# Step 1: walk imports to learn what local names refer to	COMMENT
LOW	tests/test_ui_tars_fixes.py	758	# Step 2: walk the tree looking for Call nodes whose callable	COMMENT
LOW⚡	tests/test_metrics_route.py	531	# Step 1: hits=10 → exposed 10	COMMENT
LOW⚡	tests/test_metrics_route.py	535	# Step 2: hits=15 → exposed 15	COMMENT
LOW⚡	tests/test_metrics_route.py	540	# Step 3: simulate cache.clear() — raw drops to 0; exposed must stay ≥ 15	COMMENT
LOW⚡	tests/test_metrics_route.py	549	# Step 4: hits=3 — fresh activity after clear; exposed should be 15+3=18	COMMENT
LOW⚡	tests/test_reasoning_budget_generation.py	189	# Step 1: [1] is baselined as prompt; 0/1 spent → "free", logits untouched.	COMMENT
LOW⚡	tests/test_reasoning_budget_generation.py	191	# Step 2: one generated think token → 1/1 spent → force on THIS call.	COMMENT
LOW⚡	tests/test_grammar_processor_558.py	2395	# Step 1: baseline the prompt (1 token), then feed the first generated	COMMENT
LOW⚡	tests/test_grammar_processor_558.py	2405	# Step 2: feed the second generated token — the matcher rejects it, so	COMMENT
LOW	tests/test_grammar_processor_558.py	2418	# Step 3: any subsequent call also stays dropped (logits unchanged).	COMMENT
LOW	vllm_mlx/tool_parsers/gemma4_tool_parser.py	428	# Step 2: bare KV parse	COMMENT
LOW	vllm_mlx/utils/chat_template.py	1546	# Step 1: retry without enable_thinking (many templates don't support it).	COMMENT
LOW	vllm_mlx/utils/chat_template.py	1590	# Step 2: template also rejects tools — fall back to prompt injection.	COMMENT
LOW	vllm_mlx/service/postprocessor.py	2654	# Step 1: Separate content from reasoning	COMMENT
LOW	vllm_mlx/share/cli.py	341	# Step 1: a deliberately-wrong key must NOT return 200. If the	COMMENT
LOW	vllm_mlx/share/cli.py	352	# Step 2: the real key must return 200.	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	377	// Step 1: cached + safe (only over the known-params subset).	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	382	// Step 2: smallest known-params alias that fits.	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	390	// Step 3: smallest known-params alias overall (.tooBig but	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	395	// Step 4: catalog has zero parseable aliases. Last resort —	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	509	// Step 1: bucketed default is on disk AND runnable. No	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	515	// Step 2: any cached + fits alternative — prefer it over a	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	528	// Step 3: nothing cached fits, but the canonical pick still	COMMENT
LOW⚡	…apid-mac/Sources/Rapid/Server/RAMBucketedDefault.swift	534	// Step 4: bucketed is .tooBig OR missing — hand off to the	COMMENT
LOW⚡	…rapid-mac/Sources/Rapid/Server/ToolUseCapability.swift	493	// Step 1: denylist match (broken takes priority).	COMMENT
LOW⚡	…rapid-mac/Sources/Rapid/Server/ToolUseCapability.swift	498	// Step 2: deliberate-unknown overrides shadow the family match	COMMENT
LOW	…rapid-mac/Sources/Rapid/Server/ToolUseCapability.swift	512	// Step 3: family + size guard. An alias is .known when its	COMMENT

AI Structural Patterns42 hits · 34 pts

Severity	File	Line	Context
LOW	tests/test_community_bench_aggregate.py	62	CODE
LOW	tests/test_cache_routes.py	84	CODE
LOW	…eox_fun_mlx/pipeline/pipeline_cogvideox_fun_inpaint.py	229	CODE
LOW	videox_fun_mlx/pipeline/scheduler.py	25	CODE
LOW	videox_fun_mlx/models/cogvideox_vae.py	307	CODE
LOW	videox_fun_mlx/models/cogvideox_vae.py	537	CODE
LOW	videox_fun_mlx/models/cogvideox_vae.py	717	CODE
LOW	videox_fun_mlx/models/cogvideox_vae.py	819	CODE
LOW	videox_fun_mlx/models/cogvideox_vae.py	950	CODE
LOW	videox_fun_mlx/models/cogvideox_vae.py	1075	CODE
LOW	videox_fun_mlx/models/cogvideox_transformer3d.py	235	CODE
LOW	videox_fun_mlx/models/cogvideox_transformer3d.py	382	CODE
LOW	videox_fun_mlx/models/cogvideox_transformer3d.py	496	CODE
LOW	examples/mcp_chat.py	47	CODE
LOW	scripts/agent_soak_test.py	272	CODE
LOW	scripts/mirror_to_r2.py	476	CODE
LOW	scripts/bench_decode_tps.py	127	CODE
LOW	vllm_mlx/server.py	1778	CODE
LOW	vllm_mlx/engine_core.py	1080	CODE
LOW	vllm_mlx/mllm_scheduler.py	529	CODE
LOW	vllm_mlx/mllm_scheduler.py	1909	CODE
LOW	vllm_mlx/gdn_in_proj_fusion.py	230	CODE
LOW	vllm_mlx/mllm_batch_generator.py	584	CODE
LOW	vllm_mlx/tool_parsers/muse_tool_parser.py	381	CODE
LOW	vllm_mlx/tool_parsers/harmony_tool_parser.py	430	CODE
LOW	vllm_mlx/video/engine.py	123	CODE
LOW	vllm_mlx/runtime/diffusion_lane.py	955	CODE
LOW	vllm_mlx/runtime/resident_models.py	582	CODE
LOW	vllm_mlx/models/mllm.py	1891	CODE
LOW	vllm_mlx/models/nemotron_labs_diffusion.py	209	CODE
LOW	vllm_mlx/models/gemma4_vendored/language.py	601	CODE
LOW	vllm_mlx/spec_decode/mtp/cache_patch.py	385	CODE
LOW	vllm_mlx/spec_decode/mtp/generator.py	199	CODE
LOW	vllm_mlx/audio/tts.py	649	CODE
LOW	vllm_mlx/speculative/dflash/server.py	610	CODE
LOW	vllm_mlx/speculative/dflash/server.py	2057	CODE
LOW	vllm_mlx/engine/batched.py	793	CODE
LOW	vllm_mlx/engine/batched.py	2324	CODE
LOW	vllm_mlx/routes/responses.py	2637	CODE
LOW	vllm_mlx/routes/images.py	442	CODE
LOW	vllm_mlx/routes/audio.py	1888	CODE
LOW	vllm_mlx/routes/video.py	594	CODE

Slop Phrases13 hits · 26 pts

Severity	File	Line	Snippet	Context
MEDIUM⚡	evals/results/tier1b_benchmark.json	333	"content": " about building a distributed system on Apple Silicon. They want it to cover architecture design, netw	CODE
MEDIUM	…0260617-apple-m3-ultra-qwen3-8b-4bit-e912922f4600.json	139	"response_excerpt": "\n\nHello! 2 + 2 equals 4. Let me know if you need help with anything else! 😊"	CODE
MEDIUM⚡	tests/test_pr_validate_codex.py	865	monkeypatch, tmp_path, "I'm unable to review this content."	CODE
MEDIUM	tests/integrations/conftest.py	455	# "I cannot provide the current weather in Tokyo as I cannot access	COMMENT
MEDIUM	scripts/agent_promo_gen.py	185	Let me know if you need help setting it up!	CODE
MEDIUM	…ests/RapidTests/GroundingConfabulationRetryTests.swift	159	"I cannot provide current data, my knowledge is only up to 2024.",	CODE
MEDIUM	…ests/RapidTests/GroundingConfabulationRetryTests.swift	239	"According to my knowledge cutoff, I cannot provide current data.",	CODE
MEDIUM⚡	apps/rapid-mac/Sources/Rapid/Chat/ChatViewModel.swift	2566	"i can't provide real-time", "i cannot provide real-time",	STRING
MEDIUM⚡	apps/rapid-mac/Sources/Rapid/Chat/ChatViewModel.swift	2567	"i can't provide current", "i cannot provide current",	STRING
MEDIUM⚡	apps/rapid-mac/Sources/Rapid/Chat/ChatViewModel.swift	2574	"i'm unable to access real-time", "i am unable to access real-time",	STRING
MEDIUM⚡	apps/rapid-mac/Sources/Rapid/Chat/ChatViewModel.swift	2575	"i'm unable to access current", "i am unable to access current",	STRING
MEDIUM⚡	apps/rapid-mac/Sources/Rapid/Chat/ChatViewModel.swift	2576	"i'm unable to provide real-time", "i am unable to provide real-time",	STRING
MEDIUM	apps/rapid-mac/Sources/Rapid/Chat/ChatViewModel.swift	2605	/// ("According to my knowledge cutoff, I cannot provide current data") and	STRING

Example Usage Blocks15 hits · 24 pts

Severity	File	Line	Snippet	Context
LOW	tests/integrations/test_aider.sh	18	# Usage:	COMMENT
LOW⚡	tests/integrations/test_openhands.sh	34	# Usage:	COMMENT
LOW	examples/demo_curl_text.sh	6	# Usage:	COMMENT
LOW	examples/demo_curl_video.sh	6	# Usage:	COMMENT
LOW	examples/demo_curl_image.sh	6	# Usage:	COMMENT
LOW	scripts/run_dogfood_mvp.sh	8	# Usage:	COMMENT
LOW	scripts/gemma4_mtp_mvp_smoke.sh	30	# Usage:	COMMENT
LOW⚡	scripts/coherence_sweep.sh	26	# Usage:	COMMENT
LOW	vllm_mlx/_integration_tests/test_aider.sh	18	# Usage:	COMMENT
LOW⚡	apps/rapid-mac/scripts/dogfood-isolate.sh	27	# Usage:	COMMENT
LOW	apps/rapid-mac/scripts/build-bootstrapper-dmg.sh	84	# Usage:	COMMENT
LOW	apps/rapid-mac/scripts/validate-dmg.sh	24	# Usage:	COMMENT
LOW	apps/rapid-mac/scripts/stage-licenses.sh	15	# Usage:	COMMENT
LOW	apps/rapid-mac/scripts/build-sidecar.sh	24	# Usage:	COMMENT
LOW	apps/rapid-mac/scripts/walkthrough.sh	14	# Usage:	COMMENT

Fake / Example Data21 hits · 18 pts

Severity	File	Line	Snippet	Context
LOW	evals/run_eval.py	1116	fake_key = f"{prefix}fake_result"	CODE
LOW	evals/run_eval.py	1118	fake_key,	CODE
LOW⚡	tests/test_gemma4_tool_parser.py	262	"endpointId": "jane@example.org",	CODE
LOW	tests/test_gemma4_tool_parser.py	76	"endpointId": "jane@example.org",	CODE
LOW	tests/test_gemma4_tool_parser.py	145	"endpointId": "jane@example.org",	CODE
LOW	tests/test_gemma4_tool_parser.py	163	"endpointId": "jane@example.org",	CODE
LOW	tests/test_tokenizer_gemma4_hybrid.py	272	"Lorem ipsum dolor sit amet",	CODE
LOW	tests/test_tokenizer_gemma4_hybrid.py	272	"Lorem ipsum dolor sit amet",	CODE
LOW⚡	tests/test_release_artifact_matrix.py	29	wheel.write_bytes(b"placeholder")	CODE
LOW⚡	tests/test_tool_param_enforcement_f141a.py	135	("email", "notanemail", "user@example.com"),	CODE
LOW	tests/test_prefix_boundary_path_parity.py	390	"""Third-party template drift keeps the conservative dummy-user LCP."""	STRING
LOW	tests/test_qwen3coder_tool_parser_streaming.py	30	"Lorem ipsum dolor sit amet, consectetur adipiscing elit. Sed do "	CODE
LOW	tests/test_qwen3coder_tool_parser_streaming.py	30	"Lorem ipsum dolor sit amet, consectetur adipiscing elit. Sed do "	CODE
LOW	tests/test_download_gate.py	2437	(snap / "vae.safetensors" / "placeholder").write_bytes(b"x" * 16)	CODE
LOW	tests/test_download_gate.py	2449	(snap / "vae.safetensors" / "placeholder").unlink()	CODE
LOW	tests/test_upstream_regression.py	390	'[TOOL_CALLS] [{"arguments":{"name": "John Doe"}, "name": "get_age"}]',	CODE
LOW	tests/test_upstream_regression.py	392	{"name": "John Doe"},	CODE
LOW	apps/rapid-mac/CHANGELOG.md	3291	"placeholder"; pill HStacks collapsed so VoiceOver doesn't	CODE
LOW	…-mac/Sources/Rapid/UI/Markdown/SyntaxHighlighter.swift	1235	"width", "height", "alt", "title", "rel", "target", "placeholder",	CODE
LOW	…wiftMath/Sources/SwiftMath/MathRender/MTMathList.swift	358	string += "{\(self.numerator?.description ?? "placeholder")}{\(self.denominator?.description ?? "placeholder")}"	CODE
LOW	…wiftMath/Sources/SwiftMath/MathRender/MTMathList.swift	407	string += "{\(self.radicand?.description ?? "placeholder")}"	CODE

Dead Code5 hits · 10 pts

Severity	File	Line	Context
MEDIUM	tests/test_r15_routes_stability_batch.py	501	CODE
MEDIUM	tests/test_r15_routes_stability_batch.py	533	CODE
MEDIUM	tests/test_engine_lifecycle.py	1007	CODE
MEDIUM	tests/test_engine_lifecycle.py	1471	CODE
MEDIUM	tests/test_engine_lifecycle.py	1487	CODE

Hallucination Indicators1 hit · 10 pts

Severity	File	Line	Snippet	Context
CRITICAL	tests/test_prompt_cache_snapshot.py	336	scheduler.memory_aware_cache.store.call_args.kwargs.get("evict_prefixes")	CODE

Overly Generic Function Names1 hit · 1 pts

Severity	File	Line	Snippet	Context
LOW	scripts/benchmark_realworld.py	94	def run_task(name, messages, max_tokens=1000, tools=None, temperature=0.0):	CODE

Analysis Overview

What These Metrics Mean

Score History

Severity Breakdown

Directory Score Breakdown

Pattern Findings