marin-community/marin

18.3

Adjusted Score

18.3

Raw Score

100%

Time Factor

2026-08-29

Last Push

3.0K

Stars

Python

Language

868.2K

Lines of Code

3.4K

Files

12.1K

Pattern Hits

2026-08-29

Scan Date

0.06

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

Longitudinal tracking requires multiple scan runs. Once this repository is re-scanned after new commits land, this chart will visualise how the synthetic code signal evolves over time — enabling you to detect whether AI authorship is growing, stabilising, or being actively corrected by human engineers.

No multi-scan history yet — run the scanner again to build trend data.

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 3HIGH 195MEDIUM 1419LOW 10478

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 12095 distinct pattern matches across 25 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.

Hyper-Verbose Identifiers7738 hits · 7330 pts

Severity	File	Line	Snippet	Context
LOW	infra/pre-commit.py	471	def check_trailing_whitespace(files: list[pathlib.Path], fix: bool) -> int:	CODE
LOW	infra/pre-commit.py	597	def check_markdown_precommit_invocation(files: list[pathlib.Path], fix: bool) -> int:	CODE
LOW	infra/pre-commit.py	664	def _should_skip_skill_reference(reference: str) -> bool:	CODE
LOW	infra/mkdocs_hooks.py	13	def _lower_external_inventory_failure(record: logging.LogRecord) -> bool:	CODE
LOW	infra/check_docs_source_links.py	73	def _iter_relative_check_files() -> list[Path]:	CODE
LOW	infra/lint/prose.md	53	def reconcile_workers_via_reconcile(...): ... # stutter; just reconcile_workers.	CODE
LOW	infra/lint/cruft.md	143	def reconcile_request_from_plan(plan): ... # new-wire builder	CODE
LOW	infra/lint/cruft.md	144	def legacy_translator_request(plan): ... # old-wire builder from same plan.request.desired	CODE
LOW	infra/lint/cruft.md	252	def test_build_launch_command():	CODE
LOW	infra/lint/cruft.md	260	def test_make_runner_returns_runner():	CODE
LOW	infra/echo/hybrid_search.py	114	def full_text_score_expression(alias: str) -> str:	STRING
LOW	infra/echo/hybrid_search.py	142	def repository_file_search_statement(targets: Sequence[RepositoryTarget]) -> sqlalchemy.TextClause:	STRING
LOW	infra/echo/test_repository_identity.py	11	def test_same_path_references_are_distinct_and_round_trip():	CODE
LOW	infra/echo/test_repository_identity.py	34	def test_file_identity_rejects_unqualified_or_unconfigured_targets(value):	CODE
LOW	infra/echo/test_repository_files.py	12	def test_indexed_file_rejects_generated_vendored_secret_binary_and_oversized_content():	CODE
LOW	infra/echo/test_repository_files.py	40	def test_text_chunks_cover_long_files_with_bounded_overlap():	CODE
LOW	infra/echo/test_search_benchmark.py	27	def test_query_metrics_matches_github_comments_to_their_canonical_artifact():	CODE
LOW	infra/echo/test_search_benchmark.py	62	def test_evaluation_measures_ranked_relevance_and_no_answer_suppression(tmp_path):	CODE
LOW	infra/echo/test_search_benchmark.py	132	def test_load_benchmark_rejects_duplicate_ids(tmp_path):	CODE
LOW	infra/echo/test_search_benchmark.py	148	def test_file_judgments_require_repository_qualified_targets():	CODE
LOW	infra/echo/repository_identity.py	26	def repository_file_reference(	CODE
LOW	infra/echo/repository_identity.py	47	def stored_repository_file_reference(value: str) -> RepositoryFileReference:	CODE
LOW	infra/echo/repository_identity.py	59	def configured_repository_target(repository: str) -> search_config.RepositoryTarget:	CODE
LOW	infra/echo/test_repository_sync.py	91	def _current_activity_manifest(_token):	CODE
LOW	infra/echo/test_repository_sync.py	95	def test_activity_corpus_reader_accepts_marinmirror_schema(tmp_path):	CODE
LOW	infra/echo/test_repository_sync.py	168	def test_incremental_repository_files_fetches_only_eligible_changed_blobs():	CODE
LOW	infra/echo/test_repository_sync.py	217	def test_large_github_comparison_requests_full_rebuild():	CODE
LOW	infra/echo/test_repository_sync.py	229	def test_archive_repository_files_uses_repository_relative_safety_filters():	CODE
LOW	infra/echo/test_repository_sync.py	247	def test_selected_repository_checks_head_even_if_recently_checked(monkeypatch):	CODE
LOW	infra/echo/test_repository_sync.py	265	def test_repository_resume_keeps_only_files_missing_from_durable_checkpoint():	CODE
LOW	infra/echo/test_repository_sync.py	277	def test_repository_turn_is_durable_across_failure_and_process_restart(monkeypatch):	CODE
LOW	infra/echo/test_repository_sync.py	301	def test_repository_lock_loser_does_activity_but_does_not_consume_turn(monkeypatch):	CODE
LOW	infra/echo/test_repository_sync.py	316	def test_activity_failure_prevents_repository_lock_and_turn(monkeypatch):	CODE
LOW	infra/echo/test_search_replay.py	12	def test_replay_resumes_completed_cases_and_persists_each_execution(monkeypatch, tmp_path):	CODE
LOW	infra/echo/test_search_replay.py	69	def test_replay_manifest_rejects_duplicate_normalized_queries(tmp_path):	CODE
LOW	infra/echo/test_search_replay.py	80	def test_replay_retries_rate_limited_normal_search(monkeypatch):	CODE
LOW	infra/echo/test_search_replay.py	105	def test_reconcile_recovers_matching_query_and_domains_once(tmp_path):	CODE
LOW	infra/echo/test_search_replay.py	131	def test_quality_manifest_groups_observed_queries_and_defaults_no_answer_domains(tmp_path):	CODE
LOW⚡	infra/echo/test_okf.py	23	def test_parse_wiki_extracts_fields():	STRING
LOW⚡	infra/echo/test_okf.py	31	def test_parse_wiki_ignores_body_leading_heading():	STRING
LOW⚡	infra/echo/test_okf.py	36	def test_parse_requires_frontmatter():	STRING
LOW⚡	infra/echo/test_okf.py	46	def test_parse_wiki_reports_missing_fields():	STRING
LOW⚡	infra/echo/test_okf.py	51	def test_parse_wiki_rejects_scalar_tags():	STRING
LOW⚡	infra/echo/test_okf.py	56	def test_round_trip_through_okf():	STRING
LOW	infra/echo/cli.py	226	def current_configured_repository() -> str:	CODE
LOW	infra/echo/test_cli.py	24	def test_search_sends_selected_domains_to_federated_endpoint(monkeypatch, capsys):	CODE
LOW	infra/echo/test_cli.py	94	def test_search_defaults_to_curated_domains_without_discord(monkeypatch):	CODE
LOW	infra/echo/test_cli.py	121	def test_search_infers_configured_repository_from_contributor_fork(monkeypatch, tmp_path, capsys):	CODE
LOW	infra/echo/test_cli.py	145	def test_search_explicit_repository_bypasses_checkout_inference(monkeypatch, tmp_path, repository):	CODE
LOW	infra/echo/test_cli.py	160	def test_unscoped_file_search_fails_before_request_outside_supported_checkout(monkeypatch, tmp_path):	CODE
LOW	infra/echo/test_cli.py	179	def test_search_without_file_domain_works_outside_git(monkeypatch, tmp_path):	CODE
LOW	infra/echo/test_cli.py	194	def test_bearer_token_quiets_only_the_known_missing_email_scope_warning(monkeypatch, caplog):	CODE
LOW	infra/echo/test_cli.py	213	def test_get_fetches_full_detail_by_search_result_id(monkeypatch):	CODE
LOW	infra/echo/test_cli.py	234	def test_get_rejects_legacy_path_only_file_id():	CODE
LOW	infra/echo/test_cli.py	239	def test_feedback_submits_replayable_grades_and_stdin_note(monkeypatch, capsys):	CODE
LOW	infra/echo/test_cli.py	280	def test_feedback_links_execution_when_provided(monkeypatch):	CODE
LOW	infra/echo/test_cli.py	306	def test_history_export_pages_in_stable_id_order(monkeypatch, capsys):	CODE
LOW	infra/echo/api/app.py	714	def representative_file_lines(text: str, query: str, start_line: int) -> list[tuple[int, str]]:	CODE
LOW	infra/echo/api/app.py	750	def default_feedback_result_metadata(result_id: str, config: EchoConfig) -> FeedbackResultMetadata:	CODE
LOW	infra/echo/api/app.py	821	def validate_feedback_execution(	CODE
7678 more matches not shown…

Decorative Section Separators1245 hits · 4331 pts

Severity	File	Line	Snippet	Context
MEDIUM	infra/ci/select_tests.py	179	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/ci/select_tests.py	181	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/ci/select_tests.py	246	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/ci/select_tests.py	248	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/ci/select_tests.py	297	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/ci/select_tests.py	299	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/ci/select_tests.py	382	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/ci/select_tests.py	384	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/ci/select_tests.py	552	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/ci/select_tests.py	554	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/ci/select_tests.py	830	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/ci/select_tests.py	832	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/codehealth/review_quality.py	96	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/codehealth/review_quality.py	98	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/codehealth/review_quality.py	365	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/codehealth/review_quality.py	367	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/codehealth/review_quality.py	588	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/codehealth/review_quality.py	590	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/codehealth/review_quality.py	741	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/codehealth/review_quality.py	743	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/codehealth/review_quality.py	1099	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/codehealth/review_quality.py	1101	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/server.py	124	# --------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/server.py	126	# --------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/server.py	501	# --------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/server.py	503	# --------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/server.py	752	# --------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/server.py	754	# --------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/server.py	794	# --------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/server.py	796	# --------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/samples.py	161	# --------------------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/samples.py	163	# --------------------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/samples.py	211	# --------------------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/samples.py	213	# --------------------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/samples.py	248	# --------------------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/samples.py	250	# --------------------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/fixtures.py	135	# --------------------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/fixtures.py	137	# --------------------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/fixtures.py	306	# --------------------------------------------------------------------------------------------------	COMMENT
MEDIUM	infra/evaldash/src/fixtures.py	308	# --------------------------------------------------------------------------------------------------	COMMENT
MEDIUM	experiments/ferries/daily.py	41	# ---------------------------	COMMENT
MEDIUM	experiments/ferries/daily.py	43	# ---------------------------	COMMENT
MEDIUM	experiments/grug/recovery/minimal_wedge_repro.py	50	# --------------------------------------------------------------------------	COMMENT
MEDIUM	experiments/grug/recovery/minimal_wedge_repro.py	57	# --------------------------------------------------------------------------	COMMENT
MEDIUM	experiments/grug/recovery/minimal_wedge_repro.py	106	# --------------------------------------------------------------------------	COMMENT
MEDIUM	experiments/grug/recovery/minimal_wedge_repro.py	108	# --------------------------------------------------------------------------	COMMENT
MEDIUM	experiments/grug/recovery/minimal_wedge_repro.py	191	# --------------------------------------------------------------------------	COMMENT
MEDIUM	experiments/grug/recovery/minimal_wedge_repro.py	196	# --------------------------------------------------------------------------	COMMENT
MEDIUM	experiments/grug/recovery/minimal_wedge_repro.py	242	# --------------------------------------------------------------------------	COMMENT
MEDIUM	experiments/grug/recovery/minimal_wedge_repro.py	244	# --------------------------------------------------------------------------	COMMENT
MEDIUM	experiments/grug/recovery/minimal_wedge_repro.py	269	# --------------------------------------------------------------------------	COMMENT
MEDIUM	experiments/grug/recovery/minimal_wedge_repro.py	273	# --------------------------------------------------------------------------	COMMENT
MEDIUM	experiments/datakit/scripts/sync_datakit.py	128	# ----------------------------------------------------------------------	COMMENT
MEDIUM	experiments/datakit/scripts/sync_datakit.py	130	# ----------------------------------------------------------------------	COMMENT
MEDIUM	experiments/datakit/scripts/sync_datakit.py	174	# ----------------------------------------------------------------------	COMMENT
MEDIUM	experiments/datakit/scripts/sync_datakit.py	176	# ----------------------------------------------------------------------	COMMENT
MEDIUM	experiments/datakit/scripts/sync_datakit.py	279	# ----------------------------------------------------------------------	COMMENT
MEDIUM	experiments/datakit/scripts/sync_datakit.py	281	# ----------------------------------------------------------------------	COMMENT
MEDIUM	experiments/datakit/scripts/sync_datakit.py	509	# ----------------------------------------------------------------------	COMMENT
MEDIUM	experiments/datakit/scripts/sync_datakit.py	511	# ----------------------------------------------------------------------	COMMENT
1185 more matches not shown…

Unused Imports690 hits · 618 pts

Severity	File	Line	Context
LOW	infra/grafana/src/grafana_migrations/__init__.py	6	CODE
LOW	infra/grafana/src/grafana_migrations/__init__.py	6	CODE
LOW	infra/evaldash/dev/screenshots.py	20	CODE
LOW	infra/evaldash/src/metrics.py	17	CODE
LOW	infra/evaldash/src/server.py	32	CODE
LOW	infra/evaldash/src/discovery.py	11	CODE
LOW	infra/evaldash/src/review.py	17	CODE
LOW	infra/evaldash/src/results_db.py	11	CODE
LOW	infra/evaldash/src/cluster.py	14	CODE
LOW	infra/evaldash/src/record_reconciliation.py	6	CODE
LOW	infra/evaldash/src/samples.py	18	CODE
LOW	infra/evaldash/src/fixtures.py	19	CODE
LOW	infra/loom/infrastructure.py	6	CODE
LOW	infra/loom/tests/test_infrastructure.py	4	CODE
LOW	experiments/post_training/iceball_micro.py	28	CODE
LOW	experiments/grug/__init__.py	4	CODE
LOW	experiments/grug/__init__.py	4	CODE
LOW	experiments/grug/moe_hero_ep/harrier_mix_2026_08_18.py	14	CODE
LOW	…periments/grug/moe_hero_ep/harrier_mix_2026_08_17_1.py	6	CODE
LOW	experiments/grug/recovery/run_ablations.py	21	CODE
LOW	experiments/grug/recovery/config.py	11	CODE
LOW	experiments/grug/recovery/launch_wedge_supervised.py	6	CODE
LOW	experiments/grug/recovery/wedge_entrypoint.py	24	CODE
LOW	experiments/grug/recovery/train.py	20	CODE
LOW	experiments/grug/recovery/ablation_catalog.py	6	CODE
LOW	experiments/grug/recovery/minimal_wedge_repro.py	42	CODE
LOW	experiments/june_tpu_67b_a2b/dispatch.py	4	CODE
LOW	experiments/june_tpu_67b_a2b/checkpointing.py	4	CODE
LOW	experiments/june_tpu_67b_a2b/moe/train.py	4	CODE
LOW	experiments/evaluation/models.py	6	CODE
LOW	experiments/evaluation/evals.py	6	CODE
LOW	experiments/evaluation/launch.py	6	CODE
LOW	experiments/evaluation/cli.py	11	CODE
LOW	experiments/evaluation/pipeline.py	23	CODE
LOW	experiments/sft/prepare_checkpoint.py	33	CODE
LOW	experiments/sft/launcher.py	65	CODE
LOW	experiments/datakit/testbed/variants.py	21	CODE
LOW	tests/cluster/evals/test_served_evalchemy.py	18	CODE
LOW	tests/cluster/vllm/grug_training_oracle.py	6	CODE
LOW	tests/cluster/sft/test_smoke_sft_tpu.py	20	CODE
LOW	tests/post_training/test_iceball_micro.py	4	CODE
LOW	tests/vllm/test_grugmoe_real_checkpoint_e2e.py	10	CODE
LOW	tests/vllm/grugmoe_real_checkpoint_backend.py	10	CODE
LOW	tests/evaluation/test_samples_v4_migration.py	6	CODE
LOW	tests/evaluation/test_eval_archive.py	6	CODE
LOW	tests/sft/test_prepare_checkpoint.py	5	CODE
LOW	tests/sft/test_hf_to_levanter.py	12	CODE
LOW	tests/rl/test_skyrl.py	4	CODE
LOW	.agents/skills/noslop/scripts/scan_diff.py	7	CODE
LOW	.github/actions/pulumi-preview/assemble_comment.py	4	CODE
LOW	.github/actions/pulumi-preview/format_preview.py	11	CODE
LOW	lib/iris/tests/test_jax_init.py	4	CODE
LOW	lib/iris/tests/test_multigpu.py	7	CODE
LOW	lib/iris/tests/test_nsys.py	7	CODE
LOW	lib/iris/scripts/benchmark_controller.py	71	CODE
LOW	lib/iris/src/iris/cluster/config.py	18	CODE
LOW	lib/iris/src/iris/cluster/composer.py	15	CODE
LOW	lib/iris/src/iris/cluster/runtime/__init__.py	11	CODE
LOW	lib/iris/src/iris/cluster/runtime/__init__.py	11	CODE
LOW	lib/iris/src/iris/cluster/runtime/__init__.py	12	CODE
630 more matches not shown…

Modern Structural Boilerplate543 hits · 555 pts

Severity	File	Line	Snippet	Context
LOW	infra/echo/api/app.py	80	logger = logging.getLogger(__name__)	CODE
LOW	infra/ci/run_status.py	18	logger = logging.getLogger(__name__)	CODE
LOW	infra/grafana/src/finelog_source.py	23	logger = logging.getLogger(__name__)	CODE
LOW	infra/grafana/src/server.py	136	logger = logging.getLogger(__name__)	CODE
LOW	infra/grafana/src/github_app.py	24	logger = logging.getLogger(__name__)	CODE
LOW	infra/grafana/src/github_source.py	26	logger = logging.getLogger(__name__)	CODE
LOW	infra/grafana/src/k8s_source.py	50	logger = logging.getLogger(__name__)	CODE
LOW	infra/grafana/src/iris_source.py	25	logger = logging.getLogger(__name__)	CODE
LOW	infra/grafana/src/loom_alerts.py	33	logger = logging.getLogger(__name__)	CODE
LOW	infra/grafana/src/grafana_migrations/__main__.py	17	logger = logging.getLogger(__name__)	CODE
LOW	infra/pulumi/src/iac/iris/deploy.py	41	logger = logging.getLogger(__name__)	CODE
LOW	infra/evaldash/src/server.py	92	logger = logging.getLogger(__name__)	CODE
LOW	infra/evaldash/src/server.py	201	def _set_snapshot(self, records: list[EvalRunRecord]) -> None:	CODE
LOW	infra/evaldash/src/server.py	234	def set_model_archived(self, model_name: str, archived: bool, updated_by: str \| None) -> None:	CODE
LOW	infra/evaldash/src/server.py	425	def set_catalog_error(self, error: str \| None) -> None:	CODE
LOW	infra/evaldash/src/server.py	466	def set_model_archived(self, model_name: str, archived: bool, updated_by: str \| None) -> None:	CODE
LOW	infra/evaldash/src/server.py	611	def _set_cycle_error(self, error: str \| None) -> None:	CODE
LOW	infra/evaldash/src/server.py	725	def _set_cycle_error(self, error: str \| None) -> None:	CODE
LOW	infra/evaldash/src/review.py	28	logger = logging.getLogger(__name__)	CODE
LOW	infra/evaldash/src/results_db.py	44	logger = logging.getLogger(__name__)	CODE
LOW	infra/evaldash/src/results_db.py	343	def set_model_archived(engine: Engine, model_name: str, archived: bool, updated_by: str \| None) -> None:	CODE
LOW	infra/evaldash/src/cluster.py	33	logger = logging.getLogger(__name__)	CODE
LOW	infra/evaldash/src/samples.py	44	logger = logging.getLogger(__name__)	CODE
LOW	infra/cron/nightshift_cleanup.py	18	logger = logging.getLogger(__name__)	CODE
LOW	infra/cron/nightshift_ci_tests.py	22	logger = logging.getLogger(__name__)	CODE
LOW	infra/xprof/gateway.py	29	logger = logging.getLogger(__name__)	CODE
LOW	infra/xprof/gateway.py	75	def set_size(self, size: int) -> None:	CODE
LOW	infra/xprof/server.py	21	logger = logging.getLogger(__name__)	CODE
LOW	experiments/evals/federated_inference_proxy_demo.py	41	logger = logging.getLogger(__name__)	CODE
LOW	experiments/post_training/iceball_micro.py	78	logger = logging.getLogger(__name__)	CODE
LOW	experiments/ferries/grug_multislice_smoke.py	41	logger = logging.getLogger(__name__)	CODE
LOW	experiments/ferries/datakit_tier2_skewed_ferry.py	58	logger = logging.getLogger(__name__)	CODE
LOW	experiments/ferries/datakit_nemotron_ferry.py	53	logger = logging.getLogger(__name__)	CODE
LOW	experiments/ferries/datakit_ferry.py	49	logger = logging.getLogger(__name__)	CODE
LOW	experiments/grug/dispatch.py	17	logger = logging.getLogger(__name__)	CODE
LOW	experiments/grug/__init__.py	6	__all__ = ["base", "moe"]	CODE
LOW	experiments/grug/checkpointing.py	18	logger = logging.getLogger(__name__)	CODE
LOW	experiments/grug/moe_hero_ep/adamh.py	100	__all__ = ["ScaleByAdamHState", "scale_by_adamh"]	CODE
LOW	experiments/grug/moe_hero_ep/model.py	1480	__all__ = [	CODE
LOW	experiments/grug/moe_hero_ep/train.py	63	logger = logging.getLogger(__name__)	CODE
LOW	experiments/grug/moe_hero_ep/train.py	1186	__all__ = [	CODE
LOW	experiments/grug/moe_hero_ep/optimizer.py	224	__all__ = [	CODE
LOW	experiments/grug/moe_hero_ep/restore_benchmark.py	72	logger = logging.getLogger(__name__)	CODE
LOW	experiments/grug/moe_hero_fsdp/adamh.py	78	__all__ = ["ScaleByAdamHState", "scale_by_adamh"]	CODE
LOW	experiments/grug/moe_hero_fsdp/model.py	1217	__all__ = [	CODE
LOW	experiments/grug/moe_hero_fsdp/train.py	58	logger = logging.getLogger(__name__)	CODE
LOW	experiments/grug/moe_hero_fsdp/train.py	823	__all__ = [	CODE
LOW	experiments/grug/moe_hero_fsdp/optimizer.py	221	__all__ = [	CODE
LOW	experiments/grug/recovery/run_ablations.py	38	logger = logging.getLogger(__name__)	CODE
LOW	experiments/grug/recovery/launch_wedge_supervised.py	32	logger = logging.getLogger(__name__)	CODE
LOW	experiments/grug/recovery/wedge_entrypoint.py	29	logger = logging.getLogger(__name__)	CODE
LOW	experiments/grug/recovery/train.py	53	logger = logging.getLogger(__name__)	CODE
LOW	experiments/grug/moe/adamh.py	78	__all__ = ["ScaleByAdamHState", "scale_by_adamh"]	CODE
LOW	experiments/grug/moe/model.py	906	__all__ = [	CODE
LOW	experiments/grug/moe/train.py	51	logger = logging.getLogger(__name__)	CODE
LOW	experiments/grug/moe/train.py	597	__all__ = [	CODE
LOW	experiments/grug/moe/optimizer.py	310	__all__ = [	CODE
LOW	experiments/grug/base/model.py	267	__all__ = [	CODE
LOW	experiments/grug/base/train.py	57	logger = logging.getLogger(__name__)	CODE
LOW	experiments/grug/base/train.py	660	__all__ = [	CODE
483 more matches not shown…

Excessive Try-Catch Wrapping502 hits · 532 pts

Severity	File	Line	Snippet	Context
LOW	infra/pre-commit.py	374	except Exception:	CODE
LOW	infra/pre-commit.py	409	except Exception as e:	CODE
LOW	infra/pre-commit.py	416	except Exception as e:	CODE
LOW	infra/pre-commit.py	482	except Exception:	CODE
LOW	infra/pre-commit.py	532	except Exception:	CODE
LOW	infra/pre-commit.py	562	except Exception as e:	CODE
LOW	infra/pre-commit.py	609	except Exception:	CODE
LOW	infra/pre-commit.py	682	except Exception as e:	CODE
LOW	infra/pre-commit.py	716	except Exception as e:	CODE
LOW	infra/pre-commit.py	1035	except Exception as e:	CODE
LOW	infra/linter.py	295	except Exception:	CODE
LOW	infra/lint/robustness.md	89	except Exception:	CODE
LOW	infra/grafana/src/cache.py	91	except Exception as error:	CODE
LOW	infra/grafana/src/grafana_migrations/engine.py	111	except Exception:	CODE
LOW	infra/tpu-ci/setup.py	369	except Exception as e:	CODE
LOW	infra/tpu-ci/setup.py	384	except Exception as e:	CODE
LOW	infra/tpu-ci/setup.py	552	except Exception as e:	CODE
LOW	infra/tpu-ci/vm_manager.py	408	except Exception as e:	CODE
LOW	infra/tpu-ci/vm_manager.py	455	except Exception as e:	CODE
LOW	infra/tpu-ci/vm_manager.py	465	except Exception as e:	CODE
LOW	infra/tpu-ci/vm_manager.py	482	except Exception as e:	CODE
LOW	infra/tpu-ci/vm_manager.py	547	except Exception as e:	CODE
LOW	infra/evaldash/dev/screenshots.py	106	except Exception:	CODE
LOW	infra/evaldash/src/server.py	521	except Exception as exc:	CODE
LOW	infra/evaldash/src/server.py	587	except Exception as exc:	CODE
LOW	infra/evaldash/src/server.py	698	except Exception as exc:	CODE
LOW	infra/evaldash/src/review.py	247	except Exception as exc:	CODE
LOW	infra/evaldash/src/results_db.py	265	except Exception:	CODE
LOW	infra/evaldash/src/cluster.py	118	except Exception as exc:	CODE
LOW	infra/evaldash/src/cluster.py	153	except Exception as exc:	CODE
LOW	infra/evaldash/src/record_reconciliation.py	88	except Exception as exc:	CODE
LOW	infra/evaldash/src/record_reconciliation.py	97	except Exception as exc:	CODE
LOW	infra/evaldash/src/record_reconciliation.py	124	except Exception as exc:	CODE
LOW	infra/evaldash/src/samples.py	264	except Exception as exc:	CODE
LOW	infra/evaldash/src/samples.py	346	except Exception as exc:	CODE
LOW	infra/evaldash/src/samples.py	436	except Exception as exc:	CODE
LOW	infra/evaldash/src/samples.py	492	except Exception as exc:	CODE
LOW	infra/cron/nightshift_cleanup.py	315	except Exception:	CODE
LOW	infra/xprof/gateway.py	216	except Exception:	CODE
LOW	infra/xprof/gateway.py	306	except Exception as exc:	CODE
LOW	experiments/grug/checkpointing.py	88	except Exception:	CODE
LOW	experiments/grug/moe_hero_fsdp/sweep.py	490	except Exception as exc:	CODE
LOW	experiments/grug/recovery/minimal_wedge_repro.py	406	except Exception as exc: # pragma: no cover	CODE
LOW	experiments/benchmarks/fa4/tile_sweep.py	307	except Exception as exc:	CODE
LOW	experiments/june_tpu_67b_a2b/checkpointing.py	64	except Exception:	CODE
LOW	experiments/evaluation/cli.py	55	except Exception as exc:	CODE
LOW	experiments/datakit/cluster/domain/v0/sample.py	183	except Exception:	CODE
LOW	experiments/datakit/embeddings/harrier/tei.py	220	except Exception:	CODE
LOW	experiments/datakit/scripts/validate_ferry_outputs.py	317	except Exception as exc:	CODE
LOW	experiments/datakit/decontam/lmh_loader.py	83	except Exception as exc:	CODE
LOW	experiments/datakit/decontam/lmh_loader.py	93	except Exception as exc:	CODE
LOW	experiments/datakit/decontam/lmh_loader.py	135	except Exception:	CODE
LOW	experiments/datakit/decontam/prepare_eval_corpus.py	356	except Exception:	CODE
LOW	experiments/datakit/decontam/prepare_eval_corpus.py	362	except Exception:	CODE
LOW	experiments/datakit/decontam/prepare_eval_corpus.py	746	except Exception as exc:	CODE
LOW	experiments/datakit/decontam/prepare_eval_corpus.py	789	except Exception as exc:	CODE
LOW	tests/test_utils.py	40	except Exception as e:	CODE
LOW	tests/integration_test.py	323	except Exception:	CODE
LOW	…tion/deduplication/resources/parser_variants/_utils.py	51	except Exception:	CODE
LOW	…tion/deduplication/resources/parser_variants/_utils.py	109	except Exception as e:	CODE
442 more matches not shown…

Cross-Language Confusion93 hits · 509 pts

Severity	File	Line	Snippet	Context
HIGH⚡	infra/tpu-ci/vm_manager.py	145	systemctl stop unattended-upgrades.service \|\| true	CODE
HIGH⚡	infra/tpu-ci/vm_manager.py	146	systemctl disable unattended-upgrades.service \|\| true	CODE
HIGH⚡	infra/tpu-ci/vm_manager.py	147	systemctl mask unattended-upgrades.service \|\| true	CODE
HIGH⚡	infra/tpu-ci/vm_manager.py	148	pkill -9 -f unattended-upgrade \|\| true	CODE
HIGH⚡	infra/tpu-ci/vm_manager.py	149	killall -9 apt apt-get \|\| true	CODE
HIGH⚡	infra/tpu-ci/vm_manager.py	270	./svc.sh stop \|\| true	CODE
HIGH⚡	infra/tpu-ci/vm_manager.py	271	./svc.sh uninstall \|\| true	CODE
HIGH⚡	infra/tpu-ci/vm_manager.py	276	sudo -u $RUNNER_USER ./config.sh remove --token $REGISTRATION_TOKEN \|\| true	CODE
HIGH⚡	infra/tpu-ci/vm_manager.py	621	sudo journalctl -u google-startup-scripts.service -n {lines} --no-pager \|\| echo "No startup script logs found"	CODE
HIGH⚡	infra/tpu-ci/vm_manager.py	625	grep "startup-script exit status" /var/log/syslog \| tail -5 \|\| echo "No completion marker found in syslog"	CODE
HIGH⚡	infra/tpu-ci/vm_manager.py	631	cat ~/.docker/config.json \| jq -r '.auths \| keys[]' 2>/dev/null \|\| echo "Unable to parse Docker auth config"	CODE
HIGH⚡	infra/tpu-ci/vm_manager.py	664	-exec echo "{{}} ---" \\; -exec tail -n 50 {{}} \\; 2>/dev/null \|\| true	CODE
HIGH⚡	infra/tpu-ci/vm_manager.py	669	ps aux \| grep -E "(Runner.Listener\|Runner.Worker)" \| grep -v grep \|\| echo "No runner processes found"	CODE
HIGH⚡	infra/tpu-ci/vm_manager.py	673	sudo docker ps -a \|\| true	CODE
HIGH	infra/tpu-ci/vm_manager.py	165	if ! command -v docker &> /dev/null; then	CODE
HIGH	infra/tpu-ci/vm_manager.py	649	sudo systemctl status actions.runner.* --no-pager \|\| true	CODE
HIGH	infra/tpu-ci/vm_manager.py	653	sudo journalctl -u actions.runner.* -n {lines} --no-pager \|\| true	CODE
HIGH	infra/tpu-ci/vm_manager.py	775	sudo rm -f /tmp/libtpu_lockfile \|\| true	CODE
HIGH	infra/tpu-ci/vm_manager.py	776	sudo lsof -t /dev/vfio/* 2>/dev/null \| xargs -r sudo kill -9 \|\| true	CODE
HIGH	infra/ducky/__main__.py	29	DASHBOARD_BUILD = "cd lib/ducky/dashboard && { test -d node_modules \|\| npm ci; } && npm run build"	CODE
HIGH	experiments/june_tpu_67b_a2b/moe/sft_67b_a2b_2stage.py	44	cd ~/Documents/marin && source secrets.env # or "$DC_AGENT_SECRET_ENV"	STRING
HIGH	experiments/sft/launcher.py	514	# for a mixture, where epoch semantics are undefined) and keeps the ``auto_build_caches`` path.	COMMENT
HIGH	experiments/datakit/decontam/viewer/report.py	53	var vals=[],idxs=[];parts.forEach(function(p,i){if(isTok[i]){vals.push(p);idxs.push(i)}});	CODE
HIGH	tests/execution/test_artifact.py	243	'{"name": "datasets/x", "version": "v1", "fingerprint": "abc", "result": null}'	CODE
HIGH	scripts/iris/dev_tpu.py	507	uv sync --all-packages --extra=tpu --python=3.12 \|\| true	CODE
HIGH	scripts/ci/wait_for.py	238	_RECORD_JQ = '.[]\|{id:.id,author:(.user.login//""),body:(.body//""),url:(.html_url//""),state:(.state//null)}'	CODE
HIGH	lib/iris/tests/cluster/backends/k8s/test_coreweave.py	641	deleting them on stop would break a plain `iris cluster stop && iris cluster start`.	STRING
HIGH	lib/iris/tests/cluster/node_agent/test_kubernetes.py	371	assert metrics["n1"] == NodeMetrics() # present but all-null, not dropped	CODE
HIGH	lib/iris/tests/e2e/gpu_gang_smoke.py	527	"patch", "pod", pod, "-n", self.namespace, "--type=merge", "-p", '{"metadata":{"finalizers":null}}'	CODE
HIGH	lib/iris/tests/e2e/test_smoke.py	900	return routeReady && loaded;	CODE
HIGH	lib/iris/src/iris/cluster/setup_scripts.py	107	"if grep -q 'editable = true' pyproject.toml 2>/dev/null; then"	CODE
HIGH	lib/iris/src/iris/cluster/setup_scripts.py	199	'cd "$IRIS_WORKDIR" 2>/dev/null \|\| true\n'	STRING
HIGH⚡	lib/iris/src/iris/cluster/platforms/vm_lifecycle.py	137	f"sudo docker inspect --format='{{{{.State.Status}}}}' {container_name} 2>/dev/null \|\| echo 'not_found'",	CODE
HIGH	lib/iris/src/iris/cluster/platforms/k8s/controller.py	807	+ f". Run: cd infra/pulumi && pulumi stack select {config.name} && pulumi up"	CODE
HIGH⚡	…src/iris/cluster/platforms/gcp/controller_bootstrap.py	39	sudo apt-get update -qq \|\| true	CODE
HIGH⚡	…src/iris/cluster/platforms/gcp/controller_bootstrap.py	40	sudo apt-get install -y -qq sysstat \|\| true	CODE
HIGH⚡	…src/iris/cluster/platforms/gcp/controller_bootstrap.py	43	sudo sed -i 's/^ENABLED="false"/ENABLED="true"/' /etc/default/sysstat \|\| true	CODE
HIGH⚡	…src/iris/cluster/platforms/gcp/controller_bootstrap.py	44	sudo systemctl enable --now sysstat \|\| true	CODE
HIGH⚡	…src/iris/cluster/platforms/gcp/controller_bootstrap.py	54	if ! command -v docker &> /dev/null; then	CODE
HIGH	…src/iris/cluster/platforms/gcp/controller_bootstrap.py	65	sudo systemctl start docker \|\| true	CODE
HIGH	…src/iris/cluster/platforms/gcp/controller_bootstrap.py	76	if command -v snap &> /dev/null; then	CODE
HIGH	…src/iris/cluster/platforms/gcp/controller_bootstrap.py	77	timeout 300 snap wait system seed.loaded \|\| echo "[iris-controller] Warning: snap seed wait timed out"	CODE
HIGH	…src/iris/cluster/platforms/gcp/controller_bootstrap.py	104	if command -v gcloud &> /dev/null; then	CODE
HIGH	…src/iris/cluster/platforms/gcp/controller_bootstrap.py	105	sudo gcloud auth configure-docker "$AR_HOST" -q \|\| true	CODE
HIGH	…src/iris/cluster/platforms/gcp/controller_bootstrap.py	133	sudo docker kill {{ container_name }} 2>/dev/null \|\| true	CODE
HIGH	…src/iris/cluster/platforms/gcp/controller_bootstrap.py	134	sudo docker rm {{ container_name }} 2>/dev/null \|\| true	CODE
HIGH	…src/iris/cluster/platforms/gcp/controller_bootstrap.py	169	STATUS=$(sudo docker inspect --format='{{.State.Status}}' {{ container_name }} 2>/dev/null \|\| echo 'unknown')	CODE
HIGH	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	199	--zone="$IRIS_TPU_ZONE" --format='value(state)' 2>/dev/null \|\| true)	CODE
HIGH	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	218	if ! command -v docker &> /dev/null; then	CODE
HIGH⚡	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	229	sudo systemctl start docker \|\| true	CODE
HIGH⚡	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	235	if ! command -v runsc &> /dev/null; then	CODE
HIGH⚡	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	241	&& (cd /usr/local/bin && sudo sha512sum -c /tmp/runsc.sha512); then	CODE
HIGH	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	252	if command -v runsc &> /dev/null && ! grep -q '"runsc"' /etc/docker/daemon.json 2>/dev/null; then	CODE
HIGH	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	277	if command -v snap &> /dev/null; then	CODE
HIGH	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	278	timeout 300 snap wait system seed.loaded \|\| echo "[iris-init] Warning: snap seed wait timed out"	CODE
HIGH	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	328	if command -v gcloud &> /dev/null; then	CODE
HIGH	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	329	sudo gcloud auth configure-docker "$AR_HOST" -q \|\| true	CODE
HIGH⚡	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	360	sudo docker rm -f iris-worker 2>/dev/null \|\| true	CODE
HIGH⚡	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	361	sudo docker rm -f iris-node-agent 2>/dev/null \|\| true	CODE
HIGH⚡	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	368	if command -v nvidia-smi &> /dev/null && sudo docker info --format '{{json .Runtimes}}' \| grep -q nvidia; then	CODE
33 more matches not shown…

Deep Nesting448 hits · 416 pts

Severity	File	Line	Context
LOW	infra/pre-commit.py	239	CODE
LOW	infra/pre-commit.py	358	CODE
LOW	infra/pre-commit.py	385	CODE
LOW	infra/pre-commit.py	471	CODE
LOW	infra/pre-commit.py	512	CODE
LOW	infra/pre-commit.py	544	CODE
LOW	infra/pre-commit.py	597	CODE
LOW	infra/pre-commit.py	670	CODE
LOW	infra/linter.py	447	CODE
LOW	infra/check_docs_source_links.py	48	CODE
LOW	infra/echo/search_replay.py	123	CODE
LOW	infra/echo/api/app.py	698	CODE
LOW	infra/ci/select_tests.py	218	CODE
LOW	infra/ci/select_tests.py	418	CODE
LOW	infra/ci/select_tests.py	507	CODE
LOW	infra/ci/select_tests.py	649	CODE
LOW	infra/ci/analyze_import_graph.py	218	CODE
LOW	infra/grafana/tests/test_provisioning.py	649	CODE
LOW	infra/grafana/tests/test_provisioning.py	797	CODE
LOW	infra/grafana/src/github_source.py	118	CODE
LOW	infra/grafana/src/training_stalls.py	152	CODE
LOW	infra/grafana/src/nightly.py	184	CODE
LOW	infra/codehealth/complexity.py	86	CODE
LOW	infra/pulumi/iam_principal.py	132	CODE
LOW	infra/pulumi/src/iac/github/audit.py	181	CODE
LOW	infra/pulumi/src/iac/github/resources.py	104	CODE
LOW	infra/evaldash/dev/screenshots.py	70	CODE
LOW	infra/evaldash/src/server.py	661	CODE
LOW	infra/evaldash/src/review.py	139	CODE
LOW	infra/cron/nightshift_ci_tests.py	186	CODE
LOW	infra/cron/nightshift_ci_tests.py	464	CODE
LOW	experiments/marin_tokenizer.py	200	CODE
LOW	experiments/grug/moe_hero_ep/train.py	815	CODE
LOW	experiments/grug/moe_hero_fsdp/train.py	486	CODE
LOW	experiments/grug/recovery/launch_wedge_supervised.py	58	CODE
LOW	experiments/grug/recovery/train.py	77	CODE
LOW	experiments/grug/moe/train.py	378	CODE
LOW	experiments/grug/base/train.py	420	CODE
LOW	…periments/references/reference_hyperparameter_sweep.py	302	CODE
LOW	experiments/june_tpu_67b_a2b/moe/train.py	487	CODE
LOW	experiments/tokenize/smoke_test_decon_filter_store.py	67	CODE
LOW	…ents/datakit/cluster/quality/fast_transformer/train.py	86	CODE
LOW	experiments/datakit/testbed/sampler.py	112	CODE
LOW	…periments/datakit/scripts/verify_fuzzy_dups_testbed.py	144	CODE
LOW	…periments/datakit/scripts/verify_fuzzy_dups_testbed.py	211	CODE
LOW	…periments/datakit/scripts/verify_fuzzy_dups_testbed.py	296	CODE
LOW	experiments/datakit/scripts/generate_tier2_skewed.py	86	CODE
LOW	experiments/datakit/scripts/sync_datakit.py	211	CODE
LOW	experiments/datakit/decontam/prepare_eval_corpus.py	375	CODE
LOW	experiments/datakit/decontam/prepare_eval_corpus.py	407	CODE
LOW	experiments/datakit/decontam/prepare_eval_corpus.py	487	CODE
LOW	experiments/datakit/decontam/prepare_eval_corpus.py	698	CODE
LOW	experiments/datakit/decontam/viewer/export_run.py	68	CODE
LOW	experiments/datakit/decontam/viewer/export_run.py	142	CODE
LOW	experiments/datakit/decontam/viewer/export_run.py	171	CODE
LOW	experiments/datakit/decontam/viewer/export_run.py	202	CODE
LOW	experiments/datakit/reports/common.py	53	CODE
LOW	experiments/datakit/store/datakit_store.py	292	CODE
LOW	experiments/datakit/store/datakit_store.py	473	CODE
LOW	tests/test_integration_test.py	59	CODE
388 more matches not shown…

Over-Commented Block364 hits · 352 pts

Severity	File	Line	Snippet	Context
LOW	pyproject.toml	21	"marin-rigging[secrets]",	COMMENT
LOW	pyproject.toml	41	# as marin-dupekit's runtime dependency. `scripts/rust_mode.py dev` swaps	COMMENT
LOW	pyproject.toml	201	[tool.ruff.lint]	COMMENT
LOW	pyproject.toml	401	# pyrefly cannot prove safe.	COMMENT
LOW	infra/echo/migrate.py	1	#!/usr/bin/env -S uv run --script	COMMENT
LOW	infra/ci/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	infra/deploy/src/marin_deploy/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	infra/grafana/Pulumi.marin-grafana.yaml	1	config:	COMMENT
LOW	infra/grafana/entrypoint.sh	1	#!/bin/bash	COMMENT
LOW	infra/grafana/__main__.py	41	# The cloudsql stack (infra/cloudsql) publishes the marin-metadata connection name that backs	COMMENT
LOW	infra/grafana/tests/test_sources.py	401	return httpx.Response(200, json={"data": {"project": {"run": run}}})	COMMENT
LOW	infra/grafana/provisioning/datasources/finelog.yaml	1	# Copyright The Marin Authors	COMMENT
LOW	infra/grafana/provisioning/datasources/k8s.yaml	1	# Copyright The Marin Authors	COMMENT
LOW	infra/grafana/provisioning/alerting/rules.yaml	1	# Copyright The Marin Authors	COMMENT
LOW	infra/grafana/provisioning/alerting/contact-points.yaml	1	# Copyright The Marin Authors	COMMENT
LOW	infra/grafana/src/k8s_source.py	101		COMMENT
LOW	infra/grafana/src/k8s_source.py	121	# otherwise falls back to the runtime signature of exit 255, reason Error, and a	COMMENT
LOW	infra/tpu-ci/vm_manager.py	1	#!/usr/bin/env python3	COMMENT
LOW	infra/tpu-ci/vm_manager.py	421	logging.info(f"[{zone}] Found {len(vms)} TPU VMs")	COMMENT
LOW	infra/pulumi/src/iac/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	infra/pulumi/src/iac/iris/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	infra/pulumi/src/iac/coreweave/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	infra/pulumi/src/iac/gcp/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	infra/pulumi/src/iac/gcp/iam_data.yaml	1	# Copyright The Marin Authors	COMMENT
LOW	infra/pulumi/src/iac/gcp/cloud_run.py	81	# Direct VPC egress: the service dials cluster-internal IPs, so it needs an interface	COMMENT
LOW	infra/pulumi/src/iac/gcp/iam_config.py	41	"serviceAccount:",	COMMENT
LOW	infra/pulumi/src/iac/github/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	infra/pulumi/src/iac/kubernetes/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	infra/evaldash/src/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	config/external/migration.toml	1	# Fork migration descriptors.	COMMENT
LOW	config/external/migration.toml	21	# fork with more than one pin has one section per pin (the vllm fork has `vllm` and `vllm-gpu`).	COMMENT
LOW	experiments/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/coral/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/evals/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/post_training/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/grug/moe_hero_ep/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/grug/moe_hero_ep/small_scale_abl_launch.py	101	# Kernel availability follows the accelerator, in two places. `gpu_fa4_cute` is Blackwell-only: its	COMMENT
LOW	experiments/grug/moe_hero_ep/model.py	61	# h*v weight tile at ~99KB and cannot take v=4096.	COMMENT
LOW	experiments/grug/moe_hero_ep/model.py	501	# Both branches must agree on sharding, not just shape: `lax.cond` compares full types	COMMENT
LOW	experiments/grug/moe_hero_ep/train.py	81	DEFAULT_COLLECTIVE_OVERLAP_LIMIT = 4	COMMENT
LOW	experiments/grug/moe_hero_fsdp/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/grug/moe/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/grug/base/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/rollout_data/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/benchmarks/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/benchmarks/fa4/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/june_tpu_67b_a2b/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/june_tpu_67b_a2b/moe/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/june_tpu_67b_a2b/moe/sft_67b_a2b_2stage.py	81	# --- GPU mesh geometry (compute pivot 2026-07-16: CoreWeave cw-us-east-02a H100x8 nodes) ----------	COMMENT
LOW	…_d2560_cooldown_step39k_seq64k_bs1024_rep8_muon_10T.py	81	_SLICE: str = "v4-2048"	COMMENT
LOW	experiments/datakit/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/datakit/cluster/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/datakit/cluster/quality/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	…s/datakit/cluster/quality/fast_transformer/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/datakit/cluster/domain/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/datakit/cluster/domain/v0/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/datakit/embeddings/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/datakit/embeddings/harrier/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/datakit/embeddings/luxical/__init__.py	1	# Copyright The Marin Authors	COMMENT
LOW	experiments/datakit/embeddings/luxical/pipeline.py	61	# Immutable HF commit for the weights. Passed to ``hf_hub_download`` so every region	COMMENT
304 more matches not shown…

Cross-File Repetition53 hits · 265 pts

Severity	File	Snippet	Context
HIGH	experiments/grug/moe_hero_ep/model.py	moe grug variant model. architecture: qb-routed moe with gatednorm, xsa, sigmoid combine weights. no load-balancing loss	STRING
HIGH	experiments/grug/moe_hero_fsdp/model.py	moe grug variant model. architecture: qb-routed moe with gatednorm, xsa, sigmoid combine weights. no load-balancing loss	STRING
HIGH	experiments/grug/moe/model.py	moe grug variant model. architecture: qb-routed moe with gatednorm, xsa, sigmoid combine weights. no load-balancing loss	STRING
HIGH	experiments/june_tpu_67b_a2b/moe/model.py	moe grug variant model. architecture: qb-routed moe with gatednorm, xsa, sigmoid combine weights. no load-balancing loss	STRING
HIGH	experiments/grug/moe_hero_ep/model.py	per-block gradient checkpointing. "recompute_all" reruns the whole block in backward (lowest memory); "save_moe" keeps t	STRING
HIGH	experiments/grug/moe/model.py	per-block gradient checkpointing. "recompute_all" reruns the whole block in backward (lowest memory); "save_moe" keeps t	STRING
HIGH	experiments/june_tpu_67b_a2b/moe/model.py	per-block gradient checkpointing. "recompute_all" reruns the whole block in backward (lowest memory); "save_moe" keeps t	STRING
HIGH	experiments/grug/moe_hero_ep/model.py	learnable per-dimension gating. compensates for adamh's bounded activation norms. see https://arxiv.org/abs/2601.22966v1	STRING
HIGH	experiments/grug/moe_hero_fsdp/model.py	learnable per-dimension gating. compensates for adamh's bounded activation norms. see https://arxiv.org/abs/2601.22966v1	STRING
HIGH	experiments/grug/moe/model.py	learnable per-dimension gating. compensates for adamh's bounded activation norms. see https://arxiv.org/abs/2601.22966v1	STRING
HIGH	experiments/june_tpu_67b_a2b/moe/model.py	learnable per-dimension gating. compensates for adamh's bounded activation norms. see https://arxiv.org/abs/2601.22966v1	STRING
HIGH	experiments/grug/moe_hero_ep/model.py	return a small abstract mesh and token sharding for lowering contract tests.	STRING
HIGH	experiments/grug/moe_hero_fsdp/model.py	return a small abstract mesh and token sharding for lowering contract tests.	STRING
HIGH	experiments/grug/moe/model.py	return a small abstract mesh and token sharding for lowering contract tests.	STRING
HIGH	experiments/grug/base/model.py	return a small abstract mesh and token sharding for lowering contract tests.	STRING
HIGH	experiments/june_tpu_67b_a2b/moe/model.py	return a small abstract mesh and token sharding for lowering contract tests.	STRING
HIGH	experiments/grug/moe_hero_ep/train.py	set router biases from qb betas (computed on previous step).	STRING
HIGH	experiments/grug/moe_hero_fsdp/train.py	set router biases from qb betas (computed on previous step).	STRING
HIGH	experiments/grug/moe/train.py	set router biases from qb betas (computed on previous step).	STRING
HIGH	experiments/june_tpu_67b_a2b/moe/train.py	set router biases from qb betas (computed on previous step).	STRING
HIGH	experiments/grug/moe_hero_ep/optimizer.py	restore named mesh sharding without touching single-device arrays.	STRING
HIGH	experiments/grug/moe_hero_fsdp/optimizer.py	restore named mesh sharding without touching single-device arrays.	STRING
HIGH	experiments/grug/moe/optimizer.py	restore named mesh sharding without touching single-device arrays.	STRING
HIGH	experiments/june_tpu_67b_a2b/moe/optimizer.py	restore named mesh sharding without touching single-device arrays.	STRING
HIGH	experiments/datasets/dolmino.py	dolma 1.7 dataset as lazy ``dataset`` handles (the flat dataset catalog). one handle per quality split, tokenizing each	STRING
HIGH	experiments/datasets/nemotron.py	dolma 1.7 dataset as lazy ``dataset`` handles (the flat dataset catalog). one handle per quality split, tokenizing each	STRING
HIGH	experiments/datasets/dolma.py	dolma 1.7 dataset as lazy ``dataset`` handles (the flat dataset catalog). one handle per quality split, tokenizing each	STRING
HIGH	tests/conftest.py	set marin_prefix to a temp directory for tests that rely on it.	STRING
HIGH	lib/levanter/tests/conftest.py	set marin_prefix to a temp directory for tests that rely on it.	STRING
HIGH	lib/zephyr/tests/conftest.py	set marin_prefix to a temp directory for tests that rely on it.	STRING
HIGH	lib/iris/src/iris/cluster/runtime/types.py	check container status (running, exit code, error).	STRING
HIGH	lib/iris/src/iris/cluster/runtime/docker.py	check container status (running, exit code, error).	STRING
HIGH	lib/iris/src/iris/cluster/runtime/process.py	check container status (running, exit code, error).	STRING
HIGH	lib/iris/src/iris/cluster/runtime/types.py	create an incremental log reader for this container.	STRING
HIGH	lib/iris/src/iris/cluster/runtime/docker.py	create an incremental log reader for this container.	STRING
HIGH	lib/iris/src/iris/cluster/runtime/process.py	create an incremental log reader for this container.	STRING
HIGH	lib/iris/src/iris/cluster/runtime/types.py	create a container handle from config. the handle is not started - call handle.build() then handle.run() to execute the	STRING
HIGH	lib/iris/src/iris/cluster/runtime/docker.py	create a container handle from config. the handle is not started - call handle.build() then handle.run() to execute the	STRING
HIGH	lib/iris/src/iris/cluster/runtime/process.py	create a container handle from config. the handle is not started - call handle.build() then handle.run() to execute the	STRING
HIGH	lib/iris/src/iris/rpc/controller_connect.py	returns the url path to mount the application to when serving multiple applications.	STRING
HIGH	lib/iris/src/iris/rpc/worker_connect.py	returns the url path to mount the application to when serving multiple applications.	STRING
HIGH	lib/iris/src/iris/rpc/actor_connect.py	returns the url path to mount the application to when serving multiple applications.	STRING
HIGH	lib/iris/src/iris/rpc/resource_connect.py	returns the url path to mount the application to when serving multiple applications.	STRING
HIGH	lib/finelog/src/finelog/rpc/logging_connect.py	returns the url path to mount the application to when serving multiple applications.	STRING
HIGH	lib/finelog/src/finelog/rpc/finelog_stats_connect.py	returns the url path to mount the application to when serving multiple applications.	STRING
HIGH	lib/levanter/src/levanter/models/mistral.py	args: input_ids (namedarray): [batch, position] indices of input sequence tokens in the vocabulary. attn_mask (union[nam	STRING
HIGH	lib/levanter/src/levanter/models/gemma.py	args: input_ids (namedarray): [batch, position] indices of input sequence tokens in the vocabulary. attn_mask (union[nam	STRING
HIGH	lib/levanter/src/levanter/models/mixtral.py	args: input_ids (namedarray): [batch, position] indices of input sequence tokens in the vocabulary. attn_mask (union[nam	STRING
HIGH	lib/levanter/src/levanter/models/llama.py	compute the activations for the next token in a sequence. args: input_ids: token ids with shape {pos} attn_mask: attenti	STRING
HIGH	lib/levanter/src/levanter/models/qwen.py	compute the activations for the next token in a sequence. args: input_ids: token ids with shape {pos} attn_mask: attenti	STRING
HIGH	lib/levanter/src/levanter/models/olmo.py	compute the activations for the next token in a sequence. args: input_ids: token ids with shape {pos} attn_mask: attenti	STRING
HIGH	lib/levanter/src/levanter/models/lm_model.py	compute the activations for the next token in a sequence. args: input_ids: token ids with shape {pos} attn_mask: attenti	STRING
HIGH	lib/levanter/src/levanter/models/mixtral.py	compute the activations for the next token in a sequence. args: input_ids: token ids with shape {pos} attn_mask: attenti	STRING

Docstring Block Structure46 hits · 230 pts

Severity	File	Line	Snippet	Context
HIGH	experiments/coral/batch_calibration.py	49	Estimate parameter and activation bytes for a dense transformer. Args: parameter_count: Number of trainable	STRING
HIGH	experiments/coral/batch_calibration.py	93	Return total HBM bytes for a global batch. Args: parameter_bytes: Model parameter memory. optimizer	STRING
HIGH	experiments/coral/batch_calibration.py	123	Select Levanter parallelism settings that fit a global batch on TPUs. Args: tpu: TPU topology name accepted	STRING
HIGH	experiments/datakit/testbed/sampler.py	170	Sample ``source``'s normalized shards to a target row count. Probes a single shard for rows-per-file, then computes	STRING
HIGH	.agents/projects/iris_pypi_mirror/spec.md	52	Build uv env vars that point dependency resolution at AR remote PyPI repos. Caller responsibilities (not enforced h	STRING
HIGH	.agents/projects/executor_in_training_job/spec.md	48	Run any ExecutorSteps embedded in `config`, then return a copy of `config` with all InputName / OutputName / Version	STRING
HIGH	lib/iris/src/iris/cluster/types.py	587	Create a DeviceConfig for a GPU device. Args: variant: GPU variant string (e.g., "H100", "A100"). c	STRING
HIGH	lib/iris/src/iris/cluster/types.py	610	Parse human-readable memory string to bytes. Supports various formats: - "8G", "8GB", "8 GB", "8 gigabytes"	STRING
HIGH	lib/iris/src/iris/cluster/types.py	824	Derive namespace from hierarchical job ID. The namespace is the first component of the job ID hierarchy.	STRING
HIGH	…ris/src/iris/cluster/platforms/gcp/worker_bootstrap.py	120	Render a template string with {{ variable }} placeholders. Uses ``{{ variable }}`` syntax (double braces with exact	STRING
HIGH	lib/iris/src/iris/cluster/runtime/types.py	196	Run setup_commands (uv sync, pip install, etc). Blocks until setup completes. If there are no setup_commands,	STRING
HIGH	lib/iris/src/iris/cluster/runtime/types.py	251	Profile the running process using py-spy (CPU), memray (memory), or thread dump. Args: duration_sec	STRING
HIGH	lib/iris/src/iris/cluster/runtime/docker.py	446	Run setup_commands (uv sync, pip install, etc) in a temporary container. Creates a temporary container that run	STRING
HIGH	…iris/src/iris/cluster/controller/autoscaler/factory.py	43	Create autoscaler from WorkerInfraProvider and explicit config. Args: platform: WorkerInfraProvider instanc	STRING
HIGH	lib/iris/src/iris/cluster/worker/task_attempt.py	469	Profile the running container process. Args: duration_seconds: How long to sample profi	STRING
HIGH	lib/iris/src/iris/cluster/client/remote_client.py	260	Wait for job to complete with exponential backoff polling. If the controller becomes unavailable, retries with	STRING
HIGH	lib/iris/src/iris/runtime/jax_init.py	258	Poll the endpoint registry until the coordinator address appears. Args: resolver: Namespaced resolver for t	STRING
HIGH	lib/iris/src/iris/rpc/errors.py	166	Execute an RPC call with exponential backoff retry. Retries stop when either ``max_attempts`` is exhausted or	STRING
HIGH	lib/iris/src/iris/rpc/errors.py	220	Poll an RPC endpoint, tolerating transient unavailability. Calls ``poll_fn`` in a loop. On retryable errors the fu	STRING
HIGH	lib/iris/src/iris/client/worker_pool.py	117	Execute a pickled callable and return the result. Args: serialized_callable: cloudpickle-serialized	STRING
HIGH	lib/iris/src/iris/client/worker_pool.py	323	Block until result is available. Args: timeout: Maximum time to wait in seconds Returns:	STRING
HIGH	lib/iris/src/iris/client/client.py	554	Wait for job to complete. Args: timeout: Maximum wait time in seconds poll_interval: Up	STRING
HIGH	lib/iris/src/iris/client/client.py	935	Submit a job with automatic job_id hierarchy. Args: entrypoint: Job entrypoint (callable + args/kwa	STRING
HIGH	lib/iris/src/iris/client/client.py	1408	Get an allocated port by name. Args: name: Port name (e.g., "actor") Returns:	STRING
HIGH	lib/levanter/src/levanter/eval_harness.py	106	Call a function with retry logic and exponential backoff. Handles HuggingFace rate limit errors (HTTP 429) by	STRING
HIGH	lib/levanter/src/levanter/compat/hf_checkpoints.py	129	Build a validated generation_config dict from explicit EOS token IDs. The returned dict is suitable for writing as	STRING
HIGH	lib/levanter/src/levanter/recovery/faults.py	79	Parse a fault assignment from the environment. Args: env: Mapping to read from; defaults to ``os.en	STRING
HIGH	lib/levanter/src/levanter/tracker/tracker_fns.py	244	Get or set the global tracker. Note that setting the global tracker is not thread-safe, and using a tracker fro	STRING
HIGH	lib/levanter/src/levanter/tracker/tracker_fns.py	288	Lookup a tracker in the current global tracker with the provided name. Args: name: Name of the tracker	STRING
HIGH	lib/levanter/src/levanter/models/llama.py	135	Convert to HuggingFace's LlamaConfig Args: vocab_size (int, optional): Vocabulary size of the token	STRING
HIGH	lib/marin/tools/get_hf_dataset_schema.py	126	Get the schema of a Hugging Face dataset. Args: dataset_name: Name of the dataset (e.g., 'wikitext', '	STRING
HIGH	…/processing/classification/deduplication/fuzzy_dups.py	233	Mark fuzzy-duplicate cluster membership across one or more ``MinHashAttrData`` inputs. All inputs must share identi	STRING
HIGH	lib/marin/src/marin/scaling_laws/tpu_utils.py	46	Select the smallest TPU slice that fits the estimated memory. Args: estimated_memory_bytes: Estimated memor	STRING
HIGH	…rin/src/marin/transform/huggingface/dataset_to_eval.py	342	Transform a single HuggingFace dataset example to QA format. Args: example: Raw example from HuggingFa	STRING
HIGH	…rin/src/marin/transform/huggingface/dataset_to_eval.py	474	Converts HuggingFace datasets to JSONL files for evaluation or decontamination. This function processes datase	STRING
HIGH	lib/haliax/src/haliax/debug.py	58	Checks for common issues in a module, such as reused arrays and static arrays. Equinox modules (and therefore H	STRING
HIGH	lib/haliax/src/haliax/_src/einsum.py	32	Compute the tensor contraction of the input arrays according to Haliax's named variant of the Einstein summation con	STRING
HIGH	lib/haliax/src/haliax/nn/conv.py	168	Args: inputs (NamedArray): Input array key (PRNGKeyArray: Not used, compat with other m	STRING
HIGH	lib/haliax/src/haliax/nn/conv.py	324	Args: inputs (NamedArray): Input array key (PRNGKeyArray: Not used, compat with other m	STRING
HIGH	lib/haliax/src/haliax/nn/pool.py	35	General function for pooling. Broadly based on the Flax implementation. Pooling functions are implemented usin	STRING
HIGH	lib/rigging/src/rigging/config_discovery.py	112	Resolve a cluster name (or path) to an existing YAML config file. If ``name`` is already an existing file path, it	STRING
HIGH	lib/zephyr/src/zephyr/readers.py	444	Load records from file as JSONL, Parquet, or Vortex. The spec's ``format`` selects the reader; ``auto`` (the defaul	STRING
HIGH	lib/zephyr/src/zephyr/readers.py	503	Load a Parquet file and yield ``pa.RecordBatch`` objects. Only Parquet files are supported. Raises ``RuntimeError``	STRING
HIGH	lib/zephyr/src/zephyr/dataset.py	93	Format output path with shard information. Args: pattern: Path pattern with {shard}, {total}, {basename} pl	STRING
HIGH	lib/zephyr/src/zephyr/dataset.py	414	Create dataset from file glob pattern. This method finds all files matching the glob pattern and returns a	STRING
HIGH	lib/zephyr/src/zephyr/dataset.py	1030	Streaming merge join for already-sorted, co-partitioned datasets. Preconditions: - Both datasets have t	STRING

Modern AI Meta-Vocabulary46 hits · 132 pts

Severity	File	Line	Snippet	Context
MEDIUM	AGENTS.md	80	# Lint review — agentic pass over the branch diff against the infra/lint/ catalog	COMMENT
MEDIUM	infra/README.md	42	# Your processing logic here - no manual worker orchestration needed	COMMENT
MEDIUM	infra/lint/cruft.md	25	### `ml-rollout-scaffolding` — Knob added "just for the rollout"	COMMENT
MEDIUM	infra/lint/cruft.md	126	def _run_ping_loop(self): ... # same scaffold, only the wrapped call differs	CODE
MEDIUM	infra/ci/select_tests.py	73	# orchestration itself. Locally, their import-dependent tests are sufficient;	COMMENT
MEDIUM	infra/codehealth/review_quality.py	133	# Pluggable so the backend can change without touching batching orchestration.	COMMENT
MEDIUM	infra/evaldash/dashboard/src/types/api.ts	360	// samples, and `trajectory_uri` for `agentic` samples. The one unbounded payload, the agentic	COMMENT
MEDIUM	infra/evaldash/dashboard/src/types/api.ts	394	// Agent Trajectory Interchange Format (ATIF): one agentic run's steps. A step is either a user	COMMENT
MEDIUM	infra/evaldash/src/fixtures.py	68	# A representative vLLM serving profile shared by the lm-eval runs; agentic and legacy runs override it	COMMENT
MEDIUM	infra/evaldash/src/fixtures.py	461	# An agentic run that lost one of its ten trials to a timeout: the aggregate is over the	COMMENT
MEDIUM	infra/evaldash/src/fixtures.py	466	runtime_minutes=42.0, # agentic sandbox rollouts run far longer than the lm-eval tasks	CODE
MEDIUM	experiments/evaluation/models.py	126	# tokens, so skip_special_tokens=false keeps the chain-of-thought from being stripped before scoring	COMMENT
MEDIUM	experiments/evaluation/evals.py	251	# --- Harbor (agentic registry benchmarks) ---	COMMENT
MEDIUM	experiments/evaluation/README.md	209	# A capped agentic validation run (2 tasks).	COMMENT
MEDIUM	experiments/evaluation/configs/harbor/aime-smoke.yaml	18	# This is an integration smoke, not the production agentic AIME benchmark. One bounded model	COMMENT
MEDIUM	experiments/evaluation/serve/models/README.md	58	agent: # AgentConfig -> the Harbor/agentic agent	CODE
MEDIUM	…e/models/marin-community/grug-agentic-s3-step1903.yaml	2	location: s3://marin-us-east-02a/marin/exports/grug/june-67b-a2b-sft-s3-agentic/step-1903/hf-bf16-vllm/	CODE
MEDIUM⚡	tests/evals/test_evalchemy_config.py	197	# lm-eval assumes a 2048-token window unless told otherwise, silently left-truncating few-shot	COMMENT
MEDIUM	tests/snapshots/web/expected/resiliparse/dbrx.md	76	## Quality on Long-Context Tasks and RAG	COMMENT
MEDIUM⚡	tests/snapshots/web/inputs/dbrx.html	163	</script><script type="text/javascript" defer="">(function(){function b(){!1===c&&(c=!0,Munchkin.init("094-YMS-629"))}va	CODE
MEDIUM⚡	tests/snapshots/web/inputs/dbrx.html	164	</style><p><span class="b2">Today, we are excited to introduce DBRX, an open, general-purpose LLM created by Databricks.	CODE
MEDIUM⚡	tests/snapshots/web/inputs/dbrx.html	164	</style><p><span class="b2">Today, we are excited to introduce DBRX, an open, general-purpose LLM created by Databricks.	CODE
MEDIUM	tests/evaluation/test_eval_archive.py	533	# A legacy Harbor run: one agentic sample referencing a trajectory by an in-place path.	COMMENT
MEDIUM	tests/evaluation/test_eval_archive.py	557	# The migrated agentic sample points at a finestore:// trajectory the archive resolves.	COMMENT
MEDIUM	.agents/projects/20260130_fray_lite_research.md	78	#### Job orchestration (`rl_job.py`)	COMMENT
MEDIUM	.agents/projects/eval-rollup-design.md	175	kind: SampleKind # multiple_choice \| generation \| agentic	CODE
MEDIUM	.agents/projects/eval-rollup-design.md	184	# agentic	COMMENT
MEDIUM	.agents/projects/20260130_fray_lite_design.md	590	### Job orchestration (`rl_job.py`)	COMMENT
MEDIUM	.agents/projects/20260130_fray_lite_design.md	870	│ ├── tpu.py # TPU orchestration (copied from v1 ray.tpu.execution)	CODE
MEDIUM	…gents/projects/20260430-zephyr-performance/research.md	208	#### RAG with Codebase Context	COMMENT
MEDIUM	docs/debug-log-agentic-daytona.md	1	# Debugging log for agentic Daytona validation	COMMENT
MEDIUM	docs/debug-log-agentic-daytona.md	27	- Register the [#7246](https://github.com/marin-community/marin/pull/7246) agentic presets as	CODE
MEDIUM	docs/debug-log-agentic-daytona.md	27	- Register the [#7246](https://github.com/marin-community/marin/pull/7246) agentic presets as	CODE
MEDIUM	docs/reports/marin-8b-retro.md	632	For code generation, we use [HumanEval](https://github.com/openai/human-eval) - which continues to be the standard bench	CODE
MEDIUM	scripts/ci/wait_for.py	372	# the structure: strip the scaffolding and a placeholder has no prose left.	COMMENT
MEDIUM	scripts/ci/wait_for.py	389	# multi-agent correctness review…"), while the shortest real review leaves 121. Nothing lands	COMMENT
MEDIUM	scripts/ci/wait_for.py	446	# Drop the verdict itself ("no issues") and the checklist scaffolding ("- [x] Validate	COMMENT
MEDIUM	.github/workflows/marin-release-libs-wheels.yaml	29	# package-owned source_patterns; orchestration-only changes do not publish.	COMMENT
MEDIUM	lib/iris/src/iris/cluster/controller/auth.py	130	# agentic datagen/eval job can hold one capability URL for its whole run instead	COMMENT
MEDIUM	lib/levanter/.agents/projects/inference.md	63	### 2. HTTP server scaffold	COMMENT
MEDIUM	…kernels/pallas/fused_cross_entropy_loss/batched_xla.py	18	# Empirical launch guardrails from Triton shared-memory launch failures.	COMMENT
MEDIUM⚡	lib/marin/src/marin/evaluation/archive.py	211	# ``steps`` table (agentic trajectories flattened for column projection), and finestore's reserved	COMMENT
MEDIUM	lib/marin/src/marin/evaluation/harbor/runner.py	59	# The fraction of attempted trials a verifier must grade for a run to be accepted. An agentic run	COMMENT
MEDIUM	lib/marin/src/marin/evaluation/evalchemy/client.py	30	# vLLM returns HTTP 400 when prompt_tokens + max_tokens exceeds the served context window. Reserve	COMMENT
MEDIUM	lib/marin/src/marin/evaluation/evalchemy/client.py	35	# tokens against its context window (a loglikelihood request adds one output token to a	COMMENT
MEDIUM	lib/marin/src/marin/evaluation/evalchemy/runtime.py	22	# Keep CPU-only PyTorch as a compatibility floor for benchmark orchestration and grading code.	COMMENT

AI Slop Vocabulary45 hits · 128 pts

Severity	File	Line	Snippet	Context
MEDIUM	infra/linter.py	141	# The meta lane is holistic: it reasons over the whole change, may read beyond the diff,	COMMENT
MEDIUM	infra/linter.py	162	# The holistic meta lane only runs on larger diffs — its rules need the whole change, and on a	COMMENT
MEDIUM	infra/linter.py	174	# Instruction block appended after the lane catalog; the holistic meta lane overrides the	COMMENT
MEDIUM	infra/linter.py	301	# review harness supplies the real identity through the environment. A	COMMENT
MEDIUM	infra/linter.py	826	# Drop lanes whose diff-size floor the change doesn't clear (only the holistic meta lane	COMMENT
MEDIUM	infra/tpu-ci/vm_manager.py	710	# For static mode, get comprehensive diagnostics	COMMENT
MEDIUM	infra/evaldash/dashboard/package-lock.json	2113	"resolved": "https://registry.npmjs.org/robust-predicates/-/robust-predicates-3.0.3.tgz",	CODE
MEDIUM	infra/cron/nightshift_doc_drift.py	13	# the harness default — this setting does.	COMMENT
MEDIUM	infra/cron/nightshift_cleanup.py	25	# the harness default — this setting does.	COMMENT
MEDIUM	infra/cron/nightshift_ci_tests.py	42	# the harness default — this setting does.	COMMENT
MEDIUM	experiments/evals/task_configs.py	10	# TODO: add more once supported in lm-eval-harness and/or tested on our end	COMMENT
MEDIUM	experiments/references/reference_training_pipeline.py	83	# A nano model: this harness measures path-liveness and delta-vs-baseline, not absolute	COMMENT
MEDIUM	experiments/evaluation/evals.py	274	# The comprehensive per-model benchmark set: every model x task pair runs (and is recorded) as its	COMMENT
MEDIUM	experiments/evaluation/evals.py	290	# The baseline lm-eval-harness NLP suite: 14 deterministic loglikelihood/greedy tasks, runnable on	COMMENT
MEDIUM	experiments/datakit/decontam/viewer/report.py	48	// Works at the token level so highlighting is robust to newlines/indentation:	COMMENT
MEDIUM	tests/snapshots/stackexchange/inputs/chemistry_3.json	3	"text": "# Flammability (NFPA) - how is it defined?\n\nI was reading up on a wonderful little chemical compound know	CODE
MEDIUM⚡	tests/snapshots/web/inputs/dbrx.html	164	</style><p><span class="b2">Today, we are excited to introduce DBRX, an open, general-purpose LLM created by Databricks.	CODE
MEDIUM⚡	tests/snapshots/web/inputs/dbrx.html	164	</style><p><span class="b2">Today, we are excited to introduce DBRX, an open, general-purpose LLM created by Databricks.	CODE
MEDIUM	tests/snapshots/dclm_hq/inputs/hq_1.html	231	<li id="cite_note-7"><span class="mw-cite-backlink"><b><a href="#cite_ref-7">^</a></b></span> <span class="reference-tex	CODE
MEDIUM	tests/snapshots/dclm_hq/inputs/hq_1.html	682	<li id="cite_note-7"><span class="mw-cite-backlink"><b><a href="#cite_ref-7">^</a></b></span> <span class="reference-tex	CODE
MEDIUM	tests/evaluation/test_evaldash_metrics.py	199	# Two benchmarks under the same names are not the same benchmarks if different harness versions	COMMENT
MEDIUM	tests/datakit/decontam/test_prepare_eval_corpus.py	32	# A rendered prompt embeds the passage (as lm-eval-harness doc_to_text does).	COMMENT
MEDIUM⚡	lib/iris/scripts/install_kueue.py	217	# Install core (importable; the click command and the smoke harness both call it).	COMMENT
MEDIUM⚡	lib/iris/scripts/benchmark_controller.py	151	# RPC harness: real Controller(dry_run=True) + Connect sync client	COMMENT
MEDIUM⚡	lib/iris/scripts/benchmark_controller.py	387	# Bench harness	COMMENT
MEDIUM	lib/iris/scripts/benchmark_controller.py	278	# exercise the endpoint registry set this to ``harness.make_endpoint_client``.	COMMENT
MEDIUM⚡	lib/iris/src/iris/testing/cluster.py	58	# ServiceTestHarness — parameterized GCP / K8s controller service harness	COMMENT
MEDIUM	lib/iris/src/iris/testing/cluster.py	150	# Share the harness tracker so the service registers into and reads liveness	COMMENT
MEDIUM	lib/iris/src/iris/testing/cluster.py	151	# through the same object this harness's ControllerTestState exposes.	COMMENT
MEDIUM	lib/levanter/tests/test_eval_harness.py	23	# Create chat-like requests with contexts formatted via the chat template, matching harness usage	COMMENT
MEDIUM	lib/levanter/src/levanter/eval_harness.py	1524	# remove the ",none" suffix, which eval-harness adds by default for some reason	COMMENT
LOW	lib/levanter/src/levanter/trainer.py	198	# we can't just use `lambda: model` because JAX jit can't see captures, but it can see jax partials	COMMENT
MEDIUM	…evanter/kernels/pallas/fused_cross_entropy_loss/api.py	171	# Keep TPU default stable and robust unless Pallas is explicitly requested.	COMMENT
MEDIUM	lib/levanter/src/levanter/callbacks/_metrics.py	122	# Accumulate MFU samples over a trailing window for robust distribution stats.	COMMENT
MEDIUM	lib/levanter/src/levanter/recovery/faults.py	38	# Environment variables the harness sets on each trainer subprocess.	COMMENT
MEDIUM	lib/levanter/src/levanter/utils/logging.py	65	# this isn't robust to quotes	COMMENT
LOW	lib/levanter/src/levanter/utils/logging.py	70	# XLA_FLAGS has to be set before the first jax call, so we can't just set it in the middle of the run	COMMENT
LOW	lib/levanter/src/levanter/tracker/tracker_fns.py	102	# and we just return the metrics dictionary	COMMENT
MEDIUM	lib/levanter/src/levanter/data/sharded_datasource.py	389	# (1) is cheating a bit, but it's fast and works in most cases we care about. (2) is more robust, but slower.	COMMENT
MEDIUM	lib/marin/src/marin/scaling_laws/isoflop_analysis.py	237	"""Fit a robust quadratic in log10(x) space using Huber loss.	STRING
MEDIUM	lib/marin/src/marin/execution/step_runner.py	220	# inherit it explicitly. More robust than contextvars.copy_context()	COMMENT
MEDIUM	lib/marin/src/marin/evaluation/evalchemy/result.py	69	# A retried evaluation leaves a second complete tree under the harness's scratch directory,	COMMENT
LOW	lib/haliax/src/haliax/core.py	1173	# scalar case: just add one axis so it doesn't get too mad	COMMENT
LOW	lib/haliax/src/haliax/core.py	1657	# maybe just add a context manager to allow it?	COMMENT
MEDIUM	lib/rigging/src/rigging/log_setup.py	192	# faulthandler writes to a raw fd; a harness that captures sys.stderr	COMMENT

Self-Referential Comments44 hits · 126 pts

Severity	File	Line	Snippet	Context
MEDIUM	experiments/marin_tokenizer.py	218	# Create a temporary directory that may be RAM-based	STRING
MEDIUM	tests/test_evaluator_utils.py	12	# Create a temporary directory	COMMENT
MEDIUM⚡	…cluster/backends/gcp/test_cloud_service_integration.py	514	# Create a VM first	COMMENT
MEDIUM	lib/iris/tests/cluster/backends/gcp/test_platform.py	1119	# Create a VM but set it to non-READY state	COMMENT
MEDIUM	lib/iris/tests/cluster/controller/test_autoscaler.py	538	# Create a slice with 4 VMs	COMMENT
MEDIUM	lib/iris/tests/cluster/worker/test_worker.py	442	# Create a handle that stays running until killed	COMMENT
MEDIUM	lib/iris/tests/cluster/worker/test_worker.py	466	# Create a handle that stays running until killed	COMMENT
MEDIUM	lib/iris/tests/cluster/worker/test_worker.py	500	# Create a handle that stays running until killed	COMMENT
MEDIUM	lib/iris/src/iris/cluster/worker/worker.py	366	# Create a handle wrapping the existing container	COMMENT
MEDIUM	lib/iris/src/iris/cluster/worker/worker.py	738	# Create a minimal TaskAttemptConfig. Expensive setup (port allocation,	COMMENT
MEDIUM	lib/fray/src/fray/local_backend.py	215	# Create a synthetic job handle that is immediately succeeded	COMMENT
MEDIUM	lib/fray/src/fray/iris_backend.py	779	# Create a single job with N replicas	COMMENT
MEDIUM	lib/levanter/infra/cluster/job-cluster.yaml	38	# Create a Persistent Disk w/ 100 GBs	COMMENT
MEDIUM	lib/levanter/tests/test_hyena.py	18	# Create a test config that matches the PyTorch example	COMMENT
MEDIUM	lib/levanter/tests/test_hyena.py	33	# Create a random input tensor with shape matching the PyTorch example	COMMENT
MEDIUM	lib/levanter/tests/test_hyena.py	38	# Define a function to compute the sum of a specific position's output	COMMENT
MEDIUM	lib/levanter/tests/test_background_iterable.py	28	# Create a BackgroundIterable instance with an empty producer function	COMMENT
MEDIUM	lib/levanter/tests/test_background_iterable.py	40	# Create a producer function that raises an exception	COMMENT
MEDIUM	lib/levanter/tests/test_background_iterable.py	44	# Create a BackgroundIterable instance with the producer function that raises an exception	COMMENT
MEDIUM⚡	lib/levanter/tests/test_sharded_dataset.py	50	# Create a simple dataset	COMMENT
MEDIUM	lib/levanter/tests/test_skip_step.py	362	# Create a SkipStepState instance	COMMENT
MEDIUM	lib/levanter/tests/test_sft.py	36	# Create a simple model with random embeddings	COMMENT
MEDIUM	lib/levanter/tests/test_sft.py	42	# Create a mock model with LlamaEmbedding	COMMENT
MEDIUM	lib/levanter/tests/test_packing.py	219	# Create a packed example	COMMENT
MEDIUM	lib/levanter/tests/test_packing.py	244	# Create a single JaggedArrayStore with four documents.	COMMENT
MEDIUM	lib/levanter/tests/test_packing.py	553	# Create a single-leaf dataset of 4 documents (as in test 1) with offsets [0,100,300,450,600].	COMMENT
MEDIUM	lib/levanter/tests/test_packing.py	674	# Create a dataset with a single document that's too long	COMMENT
MEDIUM⚡	lib/levanter/tests/test_jax_utils.py	194	# Create a module and a target tree	COMMENT
MEDIUM	lib/levanter/scripts/gcs_bulk_delete.py	29	# Define the transfer job	COMMENT
MEDIUM	lib/levanter/scripts/gcs_bulk_delete.py	41	# Create the transfer job	COMMENT
MEDIUM	lib/levanter/src/levanter/compat/hf_checkpoints.py	1641	# Create a temporary directory that persists through the context manager	COMMENT
MEDIUM	lib/levanter/src/levanter/optim/skipstep.py	18	# Define the state structure for the optimizer	COMMENT
MEDIUM	lib/levanter/src/levanter/utils/thread_utils.py	10	# Create a ThreadPoolExecutor	COMMENT
MEDIUM	lib/levanter/src/levanter/models/hyena.py	468	# Create a new axis with the same dimensions to avoid naming collision	COMMENT
MEDIUM	lib/levanter/src/levanter/data/packing.py	293	# Create the LmExample	COMMENT
MEDIUM	lib/levanter/src/levanter/data/audio.py	149	"""This class represents a dataset source with URLs or hf name/id."""	STRING
MEDIUM	lib/levanter/src/levanter/data/audio.py	426	"""This class represents a mixture of datasets with their associated weights."""	STRING
MEDIUM	lib/levanter/src/levanter/data/text/datasets.py	246	"""This class represents a dataset source with URLs or hf name/id."""	STRING
MEDIUM	lib/marin/src/marin/export/hf_upload.py	48	# Create the repo if it doesn't exist	COMMENT
MEDIUM	…n/src/marin/transform/wikipedia/transform_wikipedia.py	157	# Create a new <p> tag and insert the br tags plus the display math	COMMENT
MEDIUM	…marin/transform/conversation/transform_conversation.py	139	# Create a unique ID for the row based on the text	COMMENT
MEDIUM	lib/haliax/src/haliax/__init__.py	208	# Create a NamedArray along a single axis	STRING
MEDIUM	lib/haliax/src/haliax/nn/scan.py	669	# Create a function that captures the additional arguments	COMMENT
MEDIUM	lib/dupekit/tests/test_bloom.py	116	# Create a large bloom filter to ensure chunking happens	COMMENT

AI Structural Patterns115 hits · 106 pts

Severity	File	Line	Context
LOW	infra/grafana/tests/conftest.py	74	CODE
LOW	infra/grafana/tests/test_k8s_source.py	729	CODE
LOW	experiments/grug/moe_hero_ep/memory_soak.py	120	CODE
LOW	experiments/grug/moe_hero_ep/small_scale_abl_launch.py	264	CODE
LOW	experiments/grug/moe_hero_ep/launch_diagnostics.py	61	CODE
LOW	experiments/grug/moe_hero_fsdp/launch.py	264	CODE
LOW	experiments/datasets/instruction.py	135	CODE
LOW	experiments/datakit/embeddings/luxical/pipeline.py	306	CODE
LOW	experiments/datakit/testbed/train.py	125	CODE
LOW	tests/evaluation/test_eval_stats.py	43	CODE
LOW	scripts/training/time_to_train/parse_wandb_runs.py	111	CODE
LOW	lib/iris/tests/cluster/runtime/test_env_parity.py	26	CODE
LOW	lib/iris/tests/cluster/worker/test_worker.py	1176	CODE
LOW	lib/iris/scripts/install_kueue.py	219	CODE
LOW	lib/iris/scripts/benchmark_controller.py	1656	CODE
LOW	lib/iris/src/iris/cluster/platforms/gcp/fake.py	68	CODE
LOW	lib/iris/src/iris/cluster/controller/dashboard.py	191	CODE
LOW	lib/iris/src/iris/cluster/worker/worker.py	140	CODE
LOW	lib/iris/src/iris/cluster/client/protocol.py	22	CODE
LOW	lib/iris/src/iris/cluster/client/remote_client.py	150	CODE
LOW	lib/iris/src/iris/cli/job.py	563	CODE
LOW	lib/iris/src/iris/cli/job.py	715	CODE
LOW	lib/iris/src/iris/testing/controller.py	764	CODE
LOW	lib/iris/src/iris/testing/controller.py	964	CODE
LOW	lib/iris/src/iris/testing/e2e.py	102	CODE
LOW	lib/iris/src/iris/testing/journeys/world.py	176	CODE
LOW	lib/iris/src/iris/client/client.py	912	CODE
LOW	lib/levanter/tests/test_new_loader.py	93	CODE
LOW	lib/levanter/tests/test_new_loader.py	139	CODE
LOW	lib/levanter/tests/test_attention.py	699	CODE
LOW	lib/levanter/src/levanter/config.py	144	CODE
LOW	lib/levanter/src/levanter/eval_harness.py	1696	CODE
LOW	lib/levanter/src/levanter/eval_harness.py	366	CODE
LOW	lib/levanter/src/levanter/eval_harness.py	377	CODE
LOW	lib/levanter/src/levanter/eval_harness.py	381	CODE
LOW	lib/levanter/src/levanter/checkpoint.py	948	CODE
LOW	lib/levanter/src/levanter/eval.py	281	CODE
LOW	…evanter/kernels/pallas/fused_cross_entropy_loss/api.py	592	CODE
LOW	…evanter/kernels/pallas/fused_cross_entropy_loss/api.py	611	CODE
LOW	…evanter/kernels/pallas/fused_cross_entropy_loss/api.py	629	CODE
LOW	…evanter/kernels/pallas/fused_cross_entropy_loss/xla.py	678	CODE
LOW	lib/levanter/src/levanter/kernels/pallas/mamba3/api.py	524	CODE
LOW	lib/levanter/src/levanter/kernels/pallas/mamba3/api.py	652	CODE
LOW	lib/levanter/src/levanter/kernels/pallas/mamba3/api.py	693	CODE
LOW	lib/levanter/src/levanter/kernels/pallas/mamba3/api.py	860	CODE
LOW	lib/levanter/src/levanter/compat/hf_checkpoints.py	1531	CODE
LOW	lib/levanter/src/levanter/compat/hf_checkpoints.py	1055	CODE
LOW	lib/levanter/src/levanter/analysis/visualization.py	80	CODE
LOW	lib/levanter/src/levanter/analysis/backward_flow.py	415	CODE
LOW	lib/levanter/src/levanter/layers/attention.py	93	CODE
LOW	lib/levanter/src/levanter/layers/attention.py	363	CODE
LOW	lib/levanter/src/levanter/layers/attention.py	415	CODE
LOW	lib/levanter/src/levanter/layers/attention.py	509	CODE
LOW	lib/levanter/src/levanter/layers/attention.py	831	CODE
LOW	lib/levanter/src/levanter/layers/attention.py	904	CODE
LOW	lib/levanter/src/levanter/layers/attention.py	1916	CODE
LOW	lib/levanter/src/levanter/layers/rotary.py	22	CODE
LOW	lib/levanter/src/levanter/grug/grug_moe.py	146	CODE
LOW	lib/levanter/src/levanter/grug/grug_moe.py	78	CODE
LOW	…src/levanter/grug/attention/_fa4_cute_segmented_bwd.py	76	CODE
55 more matches not shown…

Structural Annotation Overuse44 hits · 79 pts

Severity	File	Line	Snippet	Context
LOW⚡	experiments/tutorials/hello_world.py	73	# Step 1: generate the data. build_config resolves ctx.output_path to the artifact's output dir.	COMMENT
LOW⚡	experiments/tutorials/hello_world.py	82	# Step 2: compute statistics over the generated data. ctx.artifact_path(_data) gives the path	COMMENT
LOW⚡	tests/execution/test_step_runner.py	119	# Step 1: download	COMMENT
LOW⚡	tests/execution/test_step_runner.py	123	# Step 2: tokenize — load upstream artifact, run, save	COMMENT
LOW	.agents/projects/20260303_iris_autoscaler_design.md	729	### Step 1: COOLDOWN State + AT_CAPACITY Rename (No Capacity Math Changes)	COMMENT
LOW	.agents/projects/20260303_iris_autoscaler_design.md	739	### Step 2: Additive Resource Packing	COMMENT
LOW	.agents/docs/smoke-test-dry-run.md	640	### Step 1: Extract `local_platform.py` from `demo_cluster.py`	COMMENT
LOW	.agents/docs/smoke-test-dry-run.md	650	### Step 2: Proto + `LocalController` + config wiring	COMMENT
LOW⚡	.agents/docs/smoke-test-dry-run.md	671	### Step 4: Smoke test integration	COMMENT
LOW⚡	.agents/docs/smoke-test-dry-run.md	679	### Step 5: Simplify `demo_cluster.py`	COMMENT
LOW⚡	.agents/docs/smoke-test-dry-run.md	686	### Step 6: CLI integration (optional)	COMMENT
LOW	.agents/docs/cluster-logging.md	166	### Step 1: Extract `stream_docker_logs()` into debug.py	COMMENT
LOW	.agents/docs/cluster-logging.md	178	### Step 2: Add `LogStreamHandle` and `start_log_streaming()` to ClusterManager	COMMENT
LOW	.agents/docs/cluster-logging.md	191	### Step 3: Add `log_dir` parameter to `connect()`	COMMENT
LOW	.agents/docs/cluster-logging.md	204	### Step 4: Simplify smoke-test.py	COMMENT
LOW	docs/tutorials/first-experiment.md	32	## Step 1: Tokenize the dataset	COMMENT
LOW	docs/tutorials/first-experiment.md	55	## Step 2: Choose a model configuration	COMMENT
LOW	docs/tutorials/first-experiment.md	78	## Step 3: Assemble the training run	COMMENT
LOW	docs/tutorials/first-experiment.md	120	## Step 4: Wire the main block	COMMENT
LOW	docs/tutorials/storage-bucket.md	17	## Step 1: Choose a Region and Name	COMMENT
LOW	docs/tutorials/storage-bucket.md	31	## Step 2: Create the Bucket	COMMENT
LOW	docs/tutorials/storage-bucket.md	53	## Step 3: Disable Soft Delete	COMMENT
LOW	docs/tutorials/storage-bucket.md	73	## Step 4: TTL Scratch Prefixes (`tmp/ttl=Nd/`)	COMMENT
LOW	docs/tutorials/storage-bucket.md	119	## Step 5: Wire It Into Marin / Levanter	COMMENT
LOW	lib/levanter/.agents/projects/gcs_serialization.md	115	# Step 1: collective gather so every device sees every host's chunk.	COMMENT
LOW⚡	lib/levanter/docs/recipes/port-models.md	14	## Step 1: Write the Config	COMMENT
LOW⚡	lib/levanter/docs/recipes/port-models.md	20	## Step 2: Implement the Model	COMMENT
LOW⚡	lib/levanter/docs/recipes/port-models.md	26	## Step 3: Add Serialization Support (if needed)	COMMENT
LOW⚡	lib/levanter/docs/recipes/port-models.md	32	## Step 4: Write Tests	COMMENT
LOW	lib/levanter/docs/recipes/port-models.md	45	## Step 5: Create a "nano" config	COMMENT
LOW	lib/levanter/docs/guides/Training-Data-Guide.md	7	## Step 1: Identify the Format for Each Dataset	COMMENT
LOW	lib/levanter/docs/guides/Training-Data-Guide.md	19	## Step 2: Prepare Your Data	COMMENT
LOW	lib/levanter/docs/guides/Training-Data-Guide.md	47	## Step 3: Write Your Config	COMMENT
LOW	lib/levanter/docs/guides/Training-Data-Guide.md	189	## Step 4: Launch Training	COMMENT
LOW	…evanter/docs/tutorials/Fine-Tuning-Semantic-Parsing.md	69	### Step 1: Prepare the Dataset	COMMENT
LOW	…evanter/docs/tutorials/Fine-Tuning-Semantic-Parsing.md	108	### Step 2: Fine-tune the Model	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	196	# Step 1: Divide Q into 𝑇𝑟 = \ceil(𝑁/Br) blocks of size Br x d each,	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	199	# Step 2: init O_i = 0, sumexp_i = 0, max_i = -inf	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	221	# Step 1: Divide Q into 𝑇𝑟 = \ceil(𝑁/Br) blocks of size Br x d each,	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	227	# Step 8: compute Sij = QiKj^T	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	263	# Step 9: Compute m_i^j = max(m_i^{j-1}, rowmax(S_i^j)), P_i^j = exp(S_i^j - m_i^j),	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	272	# Step 10: Compute O_i = diag(exp(m_i^{j-1} - m_i^j) O_i + P_i^j V_j	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	287	# Step 12: compute O_i = diag(\ell_i^{Tc})^{-1} O_i^{Tc}	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	289	# Step 13: compute L_i = m_i^{Tc} + log(\ell_i^{Tc})	COMMENT

Redundant / Tautological Comments25 hits · 38 pts

Severity	File	Line	Snippet	Context
LOW	infra/pre-commit.py	780	# Check if any pb2 file exists already	COMMENT
LOW	infra/tpu-ci/setup.py	163	# Check if user is authenticated to ghcr.io	COMMENT
LOW⚡	infra/tpu-ci/vm_manager.py	264	# Check if runner is configured by looking for config files OR .runner file	COMMENT
LOW	scripts/pm/itemize_experiment_issues.py	116	# Check if any of the URLs are already in the content	COMMENT
LOW	scripts/pm/itemize_experiment_issues.py	153	# Check if any of the URLs are already in existing_reports	COMMENT
LOW	scripts/debug/decode_tokens.py	83	# Check if it looks like a token list	COMMENT
LOW	scripts/speedrun/onboarding_setup.sh	47	# Check if a 'marin' directory exists in the current path	COMMENT
LOW⚡	scripts/speedrun/onboarding_setup.sh	137	# Check if uv is installed	COMMENT
LOW	scripts/speedrun/onboarding_setup.sh	219	# Check if branch exists, if not create it	COMMENT
LOW	scripts/speedrun/onboarding_setup.sh	280	# Check if origin points to the main community repo (indicating it's not a personal fork)	COMMENT
LOW	lib/iris/tests/cluster/backends/gcp/test_platform.py	1129	# Set VM to STAGING so it never reaches READY	COMMENT
LOW	…is/src/iris/cluster/controller/scheduling/scheduler.py	846	# Assign tasks to workers in order	COMMENT
LOW	lib/iris/src/iris/cluster/worker/task_attempt.py	888	# Check if we should stop	COMMENT
LOW⚡	lib/levanter/tests/test_prp.py	51	# Check if all elements are unique, which is a necessary condition for a bijective function	COMMENT
LOW	lib/levanter/tests/test_skip_step.py	280	# Check if it has init and update attributes	COMMENT
LOW	lib/levanter/tests/test_skip_step.py	375	# Check if the unflattened state matches the original	COMMENT
LOW	lib/levanter/scripts/clean_old_checkpoints.py	43	# Check if this checkpoint is the largest complete checkpoint	COMMENT
LOW	lib/levanter/scripts/clean_old_checkpoints.py	48	# Check if this checkpoint ends in '000' and is the largest of such	COMMENT
LOW	lib/levanter/scripts/preproc/split-pile-shards.py	60	# Check if compressor exists for this category, if not create it	COMMENT
LOW	lib/levanter/src/levanter/eval_harness.py	137	# Check if this is a rate limit error from HuggingFace	COMMENT
LOW	…src/levanter/grug/attention/_fa4_cute_segmented_bwd.py	187	# Check if block size setting is out of shared memory capacity	COMMENT
LOW	lib/marin/src/marin/markdown/markdown.py	935	# Check if base is an operator usually taking limits	COMMENT
LOW	lib/marin/src/marin/export/hf_upload.py	43	# Check if the repo exists	COMMENT
LOW	…n/src/marin/transform/wikipedia/transform_wikipedia.py	137	# Check if this is display math by looking for <dd> inside <dl>	COMMENT
LOW	lib/haliax/tests/test_axis.py	144	# Check if the element is in the correct order	COMMENT

Verbosity Indicators15 hits · 33 pts

Severity	File	Line	Snippet	Context
LOW⚡	experiments/tutorials/hello_world.py	73	# Step 1: generate the data. build_config resolves ctx.output_path to the artifact's output dir.	COMMENT
LOW⚡	experiments/tutorials/hello_world.py	82	# Step 2: compute statistics over the generated data. ctx.artifact_path(_data) gives the path	COMMENT
LOW⚡	tests/execution/test_step_runner.py	119	# Step 1: download	COMMENT
LOW⚡	tests/execution/test_step_runner.py	123	# Step 2: tokenize — load upstream artifact, run, save	COMMENT
LOW⚡	…src/levanter/grug/attention/_fa4_cute_segmented_bwd.py	321	# Do we need to check if we overshot kBlockM when we load Q?	COMMENT
LOW⚡	…src/levanter/grug/attention/_fa4_cute_segmented_bwd.py	323	# Do we need to check if we overshot kBlockN when we load K?	COMMENT
LOW⚡	…src/levanter/grug/attention/_fa4_cute_segmented_bwd.py	331	# Do we need to check if we overshot kBlockN when we load V?	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	196	# Step 1: Divide Q into 𝑇𝑟 = \ceil(𝑁/Br) blocks of size Br x d each,	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	199	# Step 2: init O_i = 0, sumexp_i = 0, max_i = -inf	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	221	# Step 1: Divide Q into 𝑇𝑟 = \ceil(𝑁/Br) blocks of size Br x d each,	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	227	# Step 8: compute Sij = QiKj^T	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	263	# Step 9: Compute m_i^j = max(m_i^{j-1}, rowmax(S_i^j)), P_i^j = exp(S_i^j - m_i^j),	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	272	# Step 10: Compute O_i = diag(exp(m_i^{j-1} - m_i^j) O_i + P_i^j V_j	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	287	# Step 12: compute O_i = diag(\ell_i^{Tc})^{-1} O_i^{Tc}	COMMENT
LOW⚡	lib/levanter/src/levanter/models/flash_attention.py	289	# Step 13: compute L_i = m_i^{Tc} + log(\ell_i^{Tc})	COMMENT

Fake / Example Data27 hits · 31 pts

Severity	File	Line	Snippet	Context
LOW⚡	tests/datakit/test_decon.py	679	{"id": "eval", "text": "lorem ipsum dolor sit amet consectetur"},	CODE
LOW⚡	tests/datakit/test_decon.py	679	{"id": "eval", "text": "lorem ipsum dolor sit amet consectetur"},	CODE
LOW⚡	tests/datakit/test_decon.py	685	"text": "LOREM IPSUM DOLOR SIT AMET CONSECTETUR",	CODE
LOW⚡	tests/datakit/test_decon.py	685	"text": "LOREM IPSUM DOLOR SIT AMET CONSECTETUR",	CODE
LOW	lib/iris/tests/cluster/backends/gcp/test_gcp_service.py	438	svc._token = "fake-token"	CODE
LOW	lib/iris/tests/cluster/backends/gcp/test_gcp_service.py	529	svc._token = "fake-token"	CODE
LOW	…cluster/backends/gcp/test_cloud_service_integration.py	292	s._token = "fake-token"	CODE
LOW⚡	lib/iris/tests/cluster/controller/test_auth.py	758	admin_users=["admin@example.com"],	CODE
LOW⚡	lib/iris/tests/cluster/controller/test_auth.py	764	assert auth.role_policy.role_for("admin@example.com") == "admin"	CODE
LOW	lib/iris/tests/cluster/controller/test_auth.py	690	return VerifiedIdentity(user_id="admin@example.com", role=self._role)	CODE
LOW	lib/iris/tests/cluster/controller/test_auth.py	707	assert seen == [VerifiedIdentity(user_id="admin@example.com", role="admin")]	CODE
LOW⚡	lib/iris/src/iris/testing/config.py	26	"placeholder": ScaleGroupConfig(	CODE
LOW⚡	lib/iris/src/iris/testing/config.py	27	name="placeholder",	CODE
LOW⚡	lib/levanter/tests/test_tokenizers.py	250	# Lorem ipsum	COMMENT
LOW⚡	lib/levanter/tests/test_tokenizers.py	251	"Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt.",	CODE
LOW⚡	lib/levanter/tests/test_tokenizers.py	251	"Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt.",	CODE
LOW	lib/levanter/tests/test_tokenizers.py	208	"user@example.com",	CODE
LOW⚡	lib/levanter/tests/test_text.py	155	lorem = """Lorem ipsum dolor sit amet, consectetur adipiscing elit. Sed do eiusmod tempor incididunt ut labore et do	CODE
LOW⚡	lib/levanter/tests/test_text.py	155	lorem = """Lorem ipsum dolor sit amet, consectetur adipiscing elit. Sed do eiusmod tempor incididunt ut labore et do	CODE
LOW⚡	lib/levanter/tests/test_text.py	177	text = "lorem ipsum dolor sit amet " * 12_000 # ~324k chars	CODE
LOW⚡	lib/levanter/tests/test_text.py	177	text = "lorem ipsum dolor sit amet " * 12_000 # ~324k chars	CODE
LOW⚡	lib/levanter/tests/test_text.py	305	"Lorem ipsum dolor sit amet, consectetur adipiscing elit. Sed do eiusmod tempor incididunt ut labore et "	CODE
LOW⚡	lib/levanter/tests/test_text.py	305	"Lorem ipsum dolor sit amet, consectetur adipiscing elit. Sed do eiusmod tempor incididunt ut labore et "	CODE
LOW	lib/marin/src/marin/inference/serve_dashboard.html	21	(()=>{"use strict";let e,t,n,r,l,i,s,o,a,c,u,p;var h,f,d,m={};function g(e){let t=Object.create(null);for(let n of e.spl	CODE
LOW⚡	lib/rigging/tests/test_server_auth.py	163	roles = {"admin@example.com": "admin"}	CODE
LOW⚡	lib/rigging/tests/test_server_auth.py	168	payload = {"aud": "/projects/1/global/backendServices/2", "email": "admin@example.com"}	CODE
LOW⚡	lib/rigging/tests/test_server_auth.py	171	assert identity == VerifiedIdentity(user_id="admin@example.com", role="admin")	CODE

Hallucination Indicators3 hits · 30 pts

Severity	File	Line	Snippet	Context
CRITICAL	lib/levanter/tests/test_apertus.py	250	lev_alpha_p = float(layer.mlp.act_fn.alpha_p.array.item())	CODE
CRITICAL	lib/levanter/tests/test_apertus.py	251	lev_alpha_n = float(layer.mlp.act_fn.alpha_n.array.item())	CODE
CRITICAL	lib/marin/src/marin/inference/serve_dashboard.html	65	`+l.text,this.inlineQueue.pop(),this.inlineQueue.at(-1).src=n.text):t.push(l);continue}if(e){this.infiniteLoopError(e.ch	CODE

Synthetic Comment Markers3 hits · 18 pts

Severity	File	Line	Snippet	Context
HIGH	tests/snapshots/dclm_hq/inputs/hq_2.html	87	<p>Each locomotive is equipped with a mobile <a href="http://www.dccwiki.com/Decoder" class="extiw" title="dcc:Decoder">	CODE
HIGH	tests/snapshots/dclm_hq/inputs/hq_2.html	426	<p>Each locomotive is equipped with a mobile <a href="http://www.dccwiki.com/Decoder" class="extiw" title="dcc:Decoder">	CODE
HIGH	docs/dev-guide/contributing.md	9	## AI-generated contributions	COMMENT

Example Usage Blocks2 hits · 3 pts

Severity	File	Line	Snippet	Context
LOW	scripts/iris/run_smoke_cw.sh	8	# Usage:	COMMENT
LOW	lib/levanter/scripts/clean_old_checkpoints.py	78	# Usage example:	COMMENT

Slop Phrases1 hit · 2 pts

Severity	File	Line	Snippet	Context
MEDIUM	tests/snapshots/stackexchange/inputs/chemistry_2.json	3	"text": "# Hierarchy of electronic wavefunctions\n\n*The previous question contained too much unnecessary informatio	CODE

Dead Code1 hit · 2 pts

Severity	File	Line	Snippet	Context
MEDIUM	lib/levanter/tests/test_background_iterable.py	116		CODE

Overly Generic Function Names1 hit · 2 pts

Severity	File	Line	Snippet	Context
LOW⚡	tests/infra/ci/test_select_tests.py	174	write(tmp_path, "lib/iris/tests/test_utils.py", "def helper():\n pass\n")	CODE

TODO Padding1 hit · 2 pts

Severity	File	Line	Snippet	Context
LOW	lib/levanter/src/levanter/models/gpt2_hyena.py	69	# TODO: implement	COMMENT

Analysis Overview

What These Metrics Mean

Score History

Severity Breakdown

Directory Score Breakdown

Pattern Findings