elder-plinius/OBLITERATUS

37.3

Adjusted Score

37.3

Raw Score

100%

Time Factor

2026-08-29

Last Push

8.1K

Stars

Python

Language

120.0K

Lines of Code

382

Files

3.2K

Pattern Hits

2026-08-29

Scan Date

0.04

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 11MEDIUM 685LOW 2475

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 3174 distinct pattern matches across 21 syntactic categories. Each entry below represents a discrete location in the source code where the engine recorded a statistically significant AI authorship indicator. Expand any category row to inspect the individual file paths, line numbers, code snippets, and the lexical context (CODE, COMMENT, or STRING) in which each match was detected.

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

Decorative Section Separators636 hits · 1968 pts

Severity	File	Line	Snippet	Context
MEDIUM⚡	app.py	1162	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	app.py	1164	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	app.py	1171	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	app.py	1173	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	app.py	3603	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	app.py	3605	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	30	# ── Container environment fixes ──────────────────────────────────────	COMMENT
MEDIUM	app.py	86	# ── ZeroGPU support ─────────────────────────────────────────────────	COMMENT
MEDIUM	app.py	323	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	325	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	361	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	363	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	447	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	449	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	534	# ── Community Hub push ────────────────────────────────────────────────	COMMENT
MEDIUM	app.py	676	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	678	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	965	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	967	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	1207	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	1209	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	1764	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	1766	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	2832	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	2834	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	3354	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	3356	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	3820	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	3822	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	4061	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	4063	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	4182	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	4184	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	app.py	5713	# ── Tab 8: About ──────────────────────────────────────────────────	COMMENT
MEDIUM	app.py	4630	# ── Tab 1: Obliterate ─────────────────────────────────────────────	STRING
MEDIUM	app.py	4875	# ── Tab 2: Benchmark ──────────────────────────────────────────────	STRING
MEDIUM	app.py	5218	# ── Tab 3: Chat ───────────────────────────────────────────────────	STRING
MEDIUM	app.py	5264	# ── Tab 4: A/B Comparison ─────────────────────────────────────────	STRING
MEDIUM	app.py	5341	# ── Tab 5: Strength Sweep ────────────────────────────────────────	STRING
MEDIUM	app.py	5411	# ── Tab 6: Tourney ────────────────────────────────────────────────	STRING
MEDIUM	app.py	5478	# ── Tab 7: Export ─────────────────────────────────────────────────	STRING
MEDIUM	app.py	5499	# ── Tab: Push to Hub ──────────────────────────────────────────────	STRING
MEDIUM	app.py	5586	# ── Tab: Leaderboard ────────────────────────────────────────────	STRING
MEDIUM	app.py	5927	# ---------------------------------------------------------------------------	STRING
MEDIUM	app.py	5929	# ---------------------------------------------------------------------------	STRING
MEDIUM⚡	tests/test_breakthrough_modules.py	259	# ===========================================================================	COMMENT
MEDIUM⚡	tests/test_breakthrough_modules.py	261	# ===========================================================================	COMMENT
MEDIUM⚡	tests/test_breakthrough_modules.py	353	# ===========================================================================	COMMENT
MEDIUM⚡	tests/test_breakthrough_modules.py	355	# ===========================================================================	COMMENT
MEDIUM⚡	tests/test_breakthrough_modules.py	459	# ===========================================================================	COMMENT
MEDIUM⚡	tests/test_breakthrough_modules.py	461	# ===========================================================================	COMMENT
MEDIUM⚡	tests/test_breakthrough_modules.py	589	# ===========================================================================	COMMENT
MEDIUM⚡	tests/test_breakthrough_modules.py	591	# ===========================================================================	COMMENT
MEDIUM⚡	tests/test_breakthrough_modules.py	816	# ===========================================================================	COMMENT
MEDIUM⚡	tests/test_breakthrough_modules.py	818	# ===========================================================================	COMMENT
MEDIUM	tests/test_breakthrough_modules.py	42	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	tests/test_breakthrough_modules.py	44	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	tests/test_breakthrough_modules.py	110	# ===========================================================================	COMMENT
MEDIUM	tests/test_breakthrough_modules.py	112	# ===========================================================================	COMMENT
MEDIUM	tests/conftest.py	57	# ---------------------------------------------------------------------------	COMMENT
576 more matches not shown…

Hyper-Verbose Identifiers1658 hits · 1493 pts

Severity	File	Line	Snippet	Context
LOW	app.py	207	def _resolve_local_checkpoint(checkpoint: str \| Path) -> Path:	CODE
LOW	app.py	236	def _format_checkpoint_reload_error(error: BaseException) -> str:	CODE
LOW	app.py	382	def _recover_sessions_from_disk() -> None:	CODE
LOW	app.py	935	def _resolve_ui_load_settings(	CODE
LOW	app.py	1221	def _format_obliteration_metrics(	CODE
LOW	app.py	1705	def _format_benchmark_results(results: list[dict], context: dict \| None = None) -> str:	CODE
LOW	app.py	2084	def _format_multi_model_results(results: list[dict], context: dict \| None = None) -> str:	CODE
LOW	app.py	3078	def _get_session_model_choices():	CODE
LOW	tests/test_strategy_navigation_contracts.py	91	def test_strategy_navigation_resolves_fallback_layers_and_missing_attention():	CODE
LOW	tests/test_strategy_navigation_contracts.py	105	def test_nemotron_h_navigation_uses_backbone_layers_and_mixer():	CODE
LOW	tests/test_strategy_navigation_contracts.py	114	def test_head_pruning_zeros_qkv_and_output_slices_for_standard_attention():	CODE
LOW	tests/test_strategy_navigation_contracts.py	139	def test_embedding_navigation_uses_first_embedding_and_fails_without_one():	CODE
LOW	tests/test_checkpoint_atomicity.py	31	def test_atomic_checkpoint_replaces_existing_destination(tmp_path):	CODE
LOW	tests/test_checkpoint_atomicity.py	44	def test_atomic_checkpoint_replaces_an_invalid_file_destination(tmp_path):	CODE
LOW	tests/test_checkpoint_atomicity.py	56	def test_atomic_checkpoint_warns_if_obsolete_backup_cannot_be_removed(	CODE
LOW	tests/test_checkpoint_atomicity.py	81	def test_atomic_checkpoint_preserves_destination_on_write_failure(tmp_path):	CODE
LOW	tests/test_checkpoint_atomicity.py	96	def test_atomic_checkpoint_cleans_staging_on_cancellation_during_write(tmp_path):	CODE
LOW	tests/test_checkpoint_atomicity.py	111	def test_atomic_checkpoint_preserves_destination_on_file_fsync_failure(	CODE
LOW	tests/test_checkpoint_atomicity.py	132	def test_new_checkpoint_parent_sync_failure_rolls_back_then_allows_retry(	CODE
LOW	tests/test_checkpoint_atomicity.py	174	def test_atomic_checkpoint_cancellation_restores_and_unlocks_for_retry(	CODE
LOW	tests/test_checkpoint_atomicity.py	207	def maybe_cancel_directory_sync(path):	CODE
LOW	tests/test_checkpoint_atomicity.py	236	def test_atomic_checkpoint_restores_destination_on_promotion_failure(	CODE
LOW	tests/test_checkpoint_atomicity.py	260	def test_atomic_checkpoint_reports_recoverable_backup_when_rollback_fails(	CODE
LOW	tests/test_checkpoint_atomicity.py	270	def fail_promotion_and_restore(source, target):	CODE
LOW	tests/test_checkpoint_atomicity.py	287	def test_atomic_checkpoint_reports_staging_cleanup_failure_without_masking_write(	CODE
LOW	tests/test_checkpoint_atomicity.py	317	def test_rebirth_failure_preserves_checkpoint_and_owned_offload(tmp_path):	CODE
LOW	tests/test_checkpoint_atomicity.py	347	def test_cleanup_only_removes_pipeline_owned_offload_directory(tmp_path):	CODE
LOW	tests/test_checkpoint_atomicity.py	362	def test_cleanup_removes_and_clears_pipeline_owned_offload_directory(tmp_path):	CODE
LOW	tests/test_checkpoint_atomicity.py	379	def test_cleanup_clears_stale_owned_offload_reference(tmp_path):	CODE
LOW	tests/test_checkpoint_atomicity.py	392	def test_cleanup_failure_retains_owned_offload_reference_for_retry(	CODE
LOW	tests/test_device_boundaries.py	18	def test_auto_device_preference(monkeypatch, cuda, mps, expected):	CODE
LOW	tests/test_device_boundaries.py	25	def test_explicit_device_validation(monkeypatch):	CODE
LOW	tests/test_device_boundaries.py	46	def test_names_and_device_counts(monkeypatch):	CODE
LOW	tests/test_device_boundaries.py	66	def test_system_memory_sources_and_fallback(monkeypatch):	CODE
LOW	tests/test_device_boundaries.py	98	def test_memory_info_for_cuda_and_cuda_fallback(monkeypatch):	CODE
LOW	tests/test_device_boundaries.py	127	def test_memory_info_for_mps_cpu_and_total_free(monkeypatch):	CODE
LOW	tests/test_device_boundaries.py	141	def test_total_cuda_memory_sums_query_and_fallback(monkeypatch):	CODE
LOW	tests/test_device_boundaries.py	158	def test_cache_cleanup_paths_are_best_effort(monkeypatch):	CODE
LOW	tests/test_device_boundaries.py	187	def test_cuda_cleanup_tolerates_every_recovery_failure(monkeypatch):	CODE
LOW	tests/test_device_boundaries.py	210	def test_seed_dtype_and_capability_contracts(monkeypatch):	CODE
LOW	tests/test_device_boundaries.py	237	def test_svd_dtype_and_oom_matching():	CODE
LOW	tests/test_device_boundaries.py	250	def test_configure_cuda_allocator(monkeypatch):	CODE
LOW	tests/test_device_boundaries.py	261	def test_configure_cuda_can_disable_cudnn(monkeypatch):	CODE
LOW	tests/test_device_boundaries.py	274	def test_configure_cuda_disables_cudnn_without_sdp_toggle(monkeypatch):	CODE
LOW	tests/test_models_client_contracts.py	31	def test_explicit_path_is_authoritative_and_does_not_fall_back(	CODE
LOW	tests/test_models_client_contracts.py	44	def test_default_resolution_uses_first_existing_packaged_source(	CODE
LOW	tests/test_models_client_contracts.py	56	def test_http_resolution_is_timeout_and_size_bounded(	CODE
LOW	tests/test_models_client_contracts.py	90	def test_http_resolution_rejects_oversized_catalog(	CODE
LOW	tests/test_models_client_contracts.py	129	def test_path_resolution_is_size_bounded(	CODE
LOW	tests/test_models_client_contracts.py	142	def test_filters_reject_non_string_arguments(isolated_sources: None) -> None:	CODE
LOW	tests/test_models_client_contracts.py	157	def test_catalog_schema_fails_closed(	CODE
LOW	tests/test_models_client_contracts.py	172	def test_filters_normalize_aliases_and_catalog_values(	CODE
LOW	tests/test_models_client_contracts.py	214	def test_filters_reject_malformed_catalog_fields(	CODE
LOW	tests/test_models_client_contracts.py	229	def test_model_ids_rejects_missing_or_non_string_ids(	CODE
LOW	tests/test_models_client_contracts.py	242	def test_newest_uses_injected_clock_release_fallback_and_stable_order(	CODE
LOW	tests/test_models_client_contracts.py	273	def test_newest_rejects_non_string_catalog_dates(	CODE
LOW	tests/test_models_client_contracts.py	288	def test_newest_rejects_invalid_day_windows(	CODE
LOW	tests/test_ci_policy.py	29	def test_every_external_action_is_sha_pinned_and_manifested():	CODE
LOW	tests/test_ci_policy.py	44	def test_tested_source_snapshot_is_created_once_attested_and_reused():	CODE
LOW⚡	tests/test_ci_policy.py	74	def test_actionlint_version_and_checksum_match_manifest():	CODE
1598 more matches not shown…

Unused Imports245 hits · 224 pts

Severity	File	Line	Context
LOW	app.py	18	CODE
LOW	app.py	2671	CODE
LOW	tests/test_gemma4_support.py	1	CODE
LOW	tests/test_strategy_navigation_contracts.py	3	CODE
LOW	tests/test_checkpoint_atomicity.py	3	CODE
LOW	tests/test_device_boundaries.py	3	CODE
LOW	tests/test_models_client_contracts.py	3	CODE
LOW	tests/test_ci_policy.py	3	CODE
LOW	tests/test_breakthrough_modules.py	9	CODE
LOW	tests/test_local_ui_portability.py	3	CODE
LOW	tests/test_lm_eval_reporting_contracts.py	3	CODE
LOW	tests/test_property_contracts.py	3	CODE
LOW	tests/conftest.py	3	CODE
LOW	tests/test_conditional_evidence_freshness.py	3	CODE
LOW	tests/test_analysis_utils.py	3	CODE
LOW	tests/test_informed_pipeline_contracts.py	3	CODE
LOW	tests/test_package_export_contracts.py	3	CODE
LOW	tests/test_advanced_metrics.py	3	CODE
LOW	tests/test_heretic_eval.py	3	CODE
LOW	tests/test_remote_boundaries.py	3	CODE
LOW	tests/test_remote_contracts.py	3	CODE
LOW	tests/test_model_load_settings.py	3	CODE
LOW	tests/test_test_risk_map.py	3	CODE
LOW	tests/test_loader_boundaries.py	3	CODE
LOW	tests/test_metrics.py	3	CODE
LOW	tests/test_benchmarks.py	3	CODE
LOW	tests/test_gpu_lifecycle.py	1	CODE
LOW	tests/test_config_properties.py	3	CODE
LOW	tests/test_visualization.py	3	CODE
LOW	tests/test_offload_surgery.py	3	CODE
LOW	tests/test_architecture_profiles.py	7	CODE
LOW	tests/test_module_imports.py	3	CODE
LOW	tests/test_persistence_contracts.py	3	CODE
LOW	tests/test_new_analysis_modules.py	9	CODE
LOW	tests/test_auto_obliterate.py	3	CODE
LOW	tests/test_evaluator.py	3	CODE
LOW	tests/test_informed_pipeline.py	3	CODE
LOW	tests/test_analysis.py	3	CODE
LOW	tests/test_causal_and_transfer.py	4	CODE
LOW	tests/test_quality_policy.py	3	CODE
LOW	tests/test_abliterate_extended.py	11	CODE
LOW	tests/test_runner_boundaries.py	3	CODE
LOW	tests/test_report.py	3	CODE
LOW	tests/test_supply_chain_policy.py	3	CODE
LOW	tests/test_leace.py	3	CODE
LOW	tests/test_adaptive_defaults.py	3	CODE
LOW	tests/test_mutmut_coverage_sitecustomize.py	3	CODE
LOW	tests/test_logit_lens.py	3	CODE
LOW	tests/test_benchmark_plots.py	3	CODE
LOW	tests/test_local_ui_contracts.py	3	CODE
LOW	tests/test_persistence_pipeline.py	3	CODE
LOW	tests/test_quant_dequant.py	12	CODE
LOW	tests/test_mlx_backend_boundaries.py	3	CODE
LOW	tests/test_telemetry_failure_contracts.py	3	CODE
LOW	tests/test_watchtower_contracts.py	3	CODE
LOW	tests/test_conditional_gate_scripts.py	3	CODE
LOW	tests/test_numerical_contracts.py	3	CODE
LOW	tests/test_quality_gate_scripts.py	3	CODE
LOW	tests/test_cli_boundaries.py	3	CODE
LOW	tests/test_bayesian_optimizer_contracts.py	3	CODE
185 more matches not shown…

Excessive Try-Catch Wrapping254 hits · 205 pts

Severity	File	Line	Snippet	Context
MEDIUM	app.py	1485	def run_pipeline():	CODE
MEDIUM	app.py	1874	def run_pipeline():	CODE
MEDIUM	app.py	2348	def run_pipeline():	CODE
MEDIUM	app.py	3003	def _generate_safe(**kwargs):	CODE
MEDIUM	app.py	3302	def _run():	CODE
MEDIUM	app.py	3488	def _gen_abliterated(**kwargs):	CODE
MEDIUM	app.py	3551	def _gen_original(**kwargs):	CODE
MEDIUM	app.py	3666	def _run_sweep_point():	CODE
MEDIUM	app.py	5669	def _push_telemetry():	CODE
LOW⚡	app.py	3585	except Exception as e:	CODE
LOW⚡	app.py	3594	except Exception:	CODE
LOW	app.py	270	except Exception:	CODE
LOW	app.py	378	except Exception:	CODE
LOW	app.py	406	except Exception:	CODE
LOW	app.py	845	except Exception as e:	CODE
LOW	app.py	870	except Exception as e:	CODE
LOW	app.py	927	except Exception:	CODE
LOW	app.py	1319	except Exception:	CODE
LOW	app.py	1344	except Exception:	CODE
LOW	app.py	1372	except Exception:	CODE
LOW	app.py	1525	except Exception as e:	CODE
LOW	app.py	1537	except Exception as cleanup_error:	CODE
LOW	app.py	1626	except Exception:	CODE
LOW	app.py	1911	except Exception as e:	CODE
LOW	app.py	1923	except Exception as cleanup_error:	CODE
LOW	app.py	2009	except Exception:	CODE
LOW	app.py	2199	except Exception:	CODE
LOW	app.py	2215	except Exception:	CODE
LOW	app.py	2276	except Exception as exc:	CODE
LOW	app.py	2448	except Exception as e:	CODE
LOW	app.py	2604	except Exception:	CODE
LOW	app.py	2715	except Exception as e:	CODE
LOW	app.py	2759	except Exception as e:	CODE
LOW	app.py	2811	except Exception as e:	CODE
LOW	app.py	2905	except Exception:	CODE
LOW	app.py	2940	except Exception as error:	CODE
LOW	app.py	2968	except Exception:	CODE
LOW	app.py	3007	except Exception as e:	CODE
LOW	app.py	3012	except Exception:	CODE
LOW	app.py	3023	except Exception:	CODE
LOW	app.py	3156	except Exception as e:	CODE
LOW	app.py	3229	except Exception:	CODE
LOW	app.py	3275	except Exception:	CODE
LOW	app.py	3318	except Exception as e:	CODE
LOW	app.py	3388	except Exception:	CODE
LOW	app.py	3422	except Exception:	CODE
LOW	app.py	3459	except Exception:	CODE
LOW	app.py	3492	except Exception as e:	CODE
LOW	app.py	3496	except Exception:	CODE
LOW	app.py	3510	except Exception:	CODE
LOW	app.py	3555	except Exception as e:	CODE
LOW	app.py	3559	except Exception:	CODE
LOW	app.py	3572	except Exception:	CODE
LOW	app.py	3682	except Exception as e:	CODE
LOW	app.py	3787	except Exception as e:	CODE
LOW	app.py	3916	except Exception as e:	CODE
LOW	app.py	3947	except Exception as e:	CODE
LOW	app.py	4007	except Exception:	CODE
LOW	app.py	5655	except Exception as e:	CODE
LOW	app.py	5701	except Exception as e:	CODE
194 more matches not shown…

Structural Annotation Overuse68 hits · 136 pts

Severity	File	Line	Snippet	Context
LOW⚡	tests/test_abliterate.py	3154	# Step 1: compute EGA directions	COMMENT
LOW⚡	tests/test_abliterate.py	3159	# Step 2: apply granular projection	COMMENT
LOW⚡	docs/executive_research_summary.md	201	# Step 1: Aggressive/SVD surgery	COMMENT
LOW⚡	docs/executive_research_summary.md	206	# Step 2: LEACE surgery	COMMENT
LOW⚡	docs/executive_research_summary.md	211	# Step 3: Blend	COMMENT
LOW⚡	docs/executive_research_summary.md	215	# Step 4: Validate	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	108	# Step 1: Get clean activations	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	115	# Step 2: Define metric	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	122	# Step 3: Patching hook	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	127	# Step 4: Sweep over layers and positions	COMMENT
LOW	docs/mechanistic_interpretability_research.md	529	# Step 1: Collect activations from frozen model	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	547	# Step 2: Train linear probe	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	553	# Step 3: Evaluate	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	559	# Step 4: The probe's weight vector IS the "refusal direction"	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	1325	# Step 1: Collect activations	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	1329	# Step 2: Pool and compute statistics	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	1334	# Step 3: Whitening transformation	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	1343	# Step 4: Whiten both sets of activations	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	1347	# Step 5: Difference-in-means in whitened space	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	1351	# Step 6: Transform back to original space for use in steering	COMMENT
LOW⚡	docs/mechanistic_interpretability_research.md	1356	# Step 7: Cosine similarity scoring at inference time	COMMENT
LOW⚡	docs/complementary_blending.md	172	# Step 1: Aggressive surgery	COMMENT
LOW⚡	docs/complementary_blending.md	177	# Step 2: LEACE surgery	COMMENT
LOW⚡	docs/complementary_blending.md	182	# Step 3: Blend	COMMENT
LOW⚡	docs/complementary_blending.md	186	# Step 4: Validate	COMMENT
LOW⚡	obliteratus/abliterate.py	2852	# Step 2: Initialize from SVD direction (warm start)	COMMENT
LOW⚡	obliteratus/abliterate.py	2856	# Step 3: Gradient-based refinement	COMMENT
LOW	obliteratus/abliterate.py	2841	# Step 1: Train linear refusal probe	COMMENT
LOW	obliteratus/abliterate.py	2898	# Step 4: Update direction with RDO-refined version	COMMENT
LOW⚡	obliteratus/interactive.py	215	# Step 1: Hardware	COMMENT
LOW⚡	obliteratus/interactive.py	219	# Step 2: Model	COMMENT
LOW⚡	obliteratus/interactive.py	223	# Step 3: Study preset OR custom strategies + sample size	COMMENT
LOW	obliteratus/interactive.py	241	# Step 4: Determine device and dtype	COMMENT
LOW⚡	obliteratus/analysis/sae_abliteration.py	449	# Step 1: Train SAE	COMMENT
LOW⚡	obliteratus/analysis/sae_abliteration.py	459	# Step 2: Identify refusal features	COMMENT
LOW⚡	obliteratus/analysis/sae_abliteration.py	465	# Step 3: Compute feature sparsity and monosemanticity	COMMENT
LOW⚡	obliteratus/analysis/sae_abliteration.py	471	# Step 4: Cluster features	COMMENT
LOW⚡	obliteratus/analysis/sae_abliteration.py	474	# Step 5: Ablation simulation	COMMENT
LOW⚡	obliteratus/analysis/sae_abliteration.py	480	# Step 6: Compare with raw direction	COMMENT
LOW	obliteratus/analysis/spectral_certification.py	187	# Step 1: Compute difference covariance matrix	COMMENT
LOW	obliteratus/analysis/spectral_certification.py	229	# Step 3: Compute BBP threshold	COMMENT
LOW⚡	obliteratus/analysis/spectral_certification.py	242	# Step 4: Marchenko-Pastur edges	COMMENT
LOW⚡	obliteratus/analysis/spectral_certification.py	246	# Step 5: Eigenvalue analysis of between-class covariance	COMMENT
LOW	obliteratus/analysis/spectral_certification.py	258	# Step 6: Classify certification level	COMMENT
LOW	obliteratus/analysis/riemannian_manifold.py	157	# Step 1: Estimate refusal directions if not provided	COMMENT
LOW	obliteratus/analysis/riemannian_manifold.py	168	# Step 2: Compute per-layer intrinsic dimension and curvature	COMMENT
LOW	obliteratus/analysis/riemannian_manifold.py	198	# Step 3: Aggregate manifold statistics	COMMENT
LOW⚡	obliteratus/analysis/wasserstein_transfer.py	159	# Step 1: Compute layer mapping if not provided	COMMENT
LOW⚡	obliteratus/analysis/wasserstein_transfer.py	166	# Step 2: For each mapped layer pair, compute OT map and transfer	COMMENT
LOW	obliteratus/analysis/wasserstein_transfer.py	215	# Step 3: Aggregate results	COMMENT
LOW	obliteratus/analysis/conditional_abliteration.py	138	# Step 1: Extract per-category condition vectors and projectors	COMMENT
LOW	obliteratus/analysis/conditional_abliteration.py	204	# Step 2: Compute cross-category geometry	COMMENT
LOW⚡	obliteratus/analysis/conditional_abliteration.py	227	# Step 3: Sheaf consistency check	COMMENT
LOW⚡	obliteratus/analysis/conditional_abliteration.py	232	# Step 4: Classify categories	COMMENT
LOW⚡	obliteratus/analysis/whitened_svd.py	150	# Step 1: Compute harmless covariance with centering	COMMENT
LOW⚡	obliteratus/analysis/whitened_svd.py	155	# Step 2: Eigendecompose covariance for whitening	COMMENT
LOW⚡	obliteratus/analysis/whitened_svd.py	173	# Step 3: Truncate near-degenerate dimensions	COMMENT
LOW⚡	obliteratus/analysis/whitened_svd.py	179	# Step 4: Compute whitening transform W = V @ diag(1/sqrt(lam + eps)) @ V^T	COMMENT
LOW⚡	obliteratus/analysis/whitened_svd.py	185	# Step 5: Whiten both activation sets (centered on harmless mean)	COMMENT
LOW⚡	obliteratus/analysis/whitened_svd.py	190	# Step 6: Compute whitened difference and SVD	COMMENT
8 more matches not shown…

Deep Nesting134 hits · 104 pts

Severity	File	Line	Context
LOW	app.py	382	CODE
LOW	app.py	1350	CODE
LOW	app.py	1378	CODE
LOW	app.py	1769	CODE
LOW	app.py	2142	CODE
LOW	app.py	2877	CODE
LOW	app.py	3084	CODE
LOW	app.py	3359	CODE
LOW	tests/test_benchmarks.py	19	CODE
LOW	tests/test_benchmarks.py	60	CODE
LOW	tests/test_novel_analysis.py	81	CODE
LOW	obliteratus/abliterate.py	1246	CODE
LOW	obliteratus/abliterate.py	1370	CODE
LOW	obliteratus/abliterate.py	1693	CODE
LOW	obliteratus/abliterate.py	2170	CODE
LOW	obliteratus/abliterate.py	2293	CODE
LOW	obliteratus/abliterate.py	3513	CODE
LOW	obliteratus/abliterate.py	3783	CODE
LOW	obliteratus/abliterate.py	4029	CODE
LOW	obliteratus/abliterate.py	4868	CODE
LOW	obliteratus/abliterate.py	5468	CODE
LOW	obliteratus/abliterate.py	5714	CODE
LOW	obliteratus/abliterate.py	5859	CODE
LOW	obliteratus/abliterate.py	5923	CODE
LOW	obliteratus/abliterate.py	6281	CODE
LOW	obliteratus/abliterate.py	6416	CODE
LOW	obliteratus/abliterate.py	6539	CODE
LOW	obliteratus/abliterate.py	6663	CODE
LOW	obliteratus/abliterate.py	6806	CODE
LOW	obliteratus/abliterate.py	7507	CODE
LOW	obliteratus/blend.py	161	CODE
LOW	obliteratus/ui_vram.py	18	CODE
LOW	obliteratus/bayesian_optimizer.py	237	CODE
LOW	obliteratus/remote.py	141	CODE
LOW	obliteratus/device.py	210	CODE
LOW	obliteratus/restore_multimodal.py	66	CODE
LOW	obliteratus/telemetry.py	182	CODE
LOW	obliteratus/telemetry.py	552	CODE
LOW	obliteratus/telemetry.py	594	CODE
LOW	obliteratus/telemetry.py	653	CODE
LOW	obliteratus/telemetry.py	842	CODE
LOW	obliteratus/telemetry.py	1104	CODE
LOW	obliteratus/prompts.py	141	CODE
LOW	obliteratus/hard_negative.py	67	CODE
LOW	obliteratus/persistence_contracts.py	295	CODE
LOW	obliteratus/watchtower.py	170	CODE
LOW	obliteratus/watchtower.py	228	CODE
LOW	obliteratus/lora_ablation.py	56	CODE
LOW	obliteratus/architecture_profiles.py	153	CODE
LOW	obliteratus/architecture_profiles.py	307	CODE
LOW	obliteratus/community.py	290	CODE
LOW	obliteratus/cli.py	150	CODE
LOW	obliteratus/cli.py	690	CODE
LOW	obliteratus/cli.py	1042	CODE
LOW	obliteratus/cli.py	1404	CODE
LOW	obliteratus/tourney.py	348	CODE
LOW	obliteratus/tourney.py	398	CODE
LOW	obliteratus/tourney.py	1152	CODE
LOW	obliteratus/interactive.py	21	CODE
LOW	obliteratus/run_archive.py	478	CODE
74 more matches not shown…

Verbosity Indicators45 hits · 85 pts

Severity	File	Line	Snippet	Context
LOW⚡	tests/test_abliterate.py	3154	# Step 1: compute EGA directions	COMMENT
LOW⚡	tests/test_abliterate.py	3159	# Step 2: apply granular projection	COMMENT
LOW⚡	obliteratus/abliterate.py	2852	# Step 2: Initialize from SVD direction (warm start)	COMMENT
LOW⚡	obliteratus/abliterate.py	2856	# Step 3: Gradient-based refinement	COMMENT
LOW	obliteratus/abliterate.py	2841	# Step 1: Train linear refusal probe	COMMENT
LOW	obliteratus/abliterate.py	2898	# Step 4: Update direction with RDO-refined version	COMMENT
LOW⚡	obliteratus/interactive.py	215	# Step 1: Hardware	COMMENT
LOW⚡	obliteratus/interactive.py	219	# Step 2: Model	COMMENT
LOW⚡	obliteratus/interactive.py	223	# Step 3: Study preset OR custom strategies + sample size	COMMENT
LOW	obliteratus/interactive.py	241	# Step 4: Determine device and dtype	COMMENT
LOW⚡	obliteratus/analysis/sae_abliteration.py	449	# Step 1: Train SAE	COMMENT
LOW⚡	obliteratus/analysis/sae_abliteration.py	459	# Step 2: Identify refusal features	COMMENT
LOW⚡	obliteratus/analysis/sae_abliteration.py	465	# Step 3: Compute feature sparsity and monosemanticity	COMMENT
LOW⚡	obliteratus/analysis/sae_abliteration.py	471	# Step 4: Cluster features	COMMENT
LOW⚡	obliteratus/analysis/sae_abliteration.py	474	# Step 5: Ablation simulation	COMMENT
LOW⚡	obliteratus/analysis/sae_abliteration.py	480	# Step 6: Compare with raw direction	COMMENT
LOW	obliteratus/analysis/spectral_certification.py	187	# Step 1: Compute difference covariance matrix	COMMENT
LOW	obliteratus/analysis/spectral_certification.py	229	# Step 3: Compute BBP threshold	COMMENT
LOW⚡	obliteratus/analysis/spectral_certification.py	242	# Step 4: Marchenko-Pastur edges	COMMENT
LOW⚡	obliteratus/analysis/spectral_certification.py	246	# Step 5: Eigenvalue analysis of between-class covariance	COMMENT
LOW	obliteratus/analysis/spectral_certification.py	258	# Step 6: Classify certification level	COMMENT
LOW	obliteratus/analysis/riemannian_manifold.py	157	# Step 1: Estimate refusal directions if not provided	COMMENT
LOW	obliteratus/analysis/riemannian_manifold.py	168	# Step 2: Compute per-layer intrinsic dimension and curvature	COMMENT
LOW	obliteratus/analysis/riemannian_manifold.py	198	# Step 3: Aggregate manifold statistics	COMMENT
LOW⚡	obliteratus/analysis/wasserstein_transfer.py	159	# Step 1: Compute layer mapping if not provided	COMMENT
LOW⚡	obliteratus/analysis/wasserstein_transfer.py	166	# Step 2: For each mapped layer pair, compute OT map and transfer	COMMENT
LOW	obliteratus/analysis/wasserstein_transfer.py	215	# Step 3: Aggregate results	COMMENT
LOW	obliteratus/analysis/conditional_abliteration.py	138	# Step 1: Extract per-category condition vectors and projectors	COMMENT
LOW	obliteratus/analysis/conditional_abliteration.py	204	# Step 2: Compute cross-category geometry	COMMENT
LOW⚡	obliteratus/analysis/conditional_abliteration.py	227	# Step 3: Sheaf consistency check	COMMENT
LOW⚡	obliteratus/analysis/conditional_abliteration.py	232	# Step 4: Classify categories	COMMENT
LOW⚡	obliteratus/analysis/whitened_svd.py	150	# Step 1: Compute harmless covariance with centering	COMMENT
LOW⚡	obliteratus/analysis/whitened_svd.py	155	# Step 2: Eigendecompose covariance for whitening	COMMENT
LOW⚡	obliteratus/analysis/whitened_svd.py	173	# Step 3: Truncate near-degenerate dimensions	COMMENT
LOW⚡	obliteratus/analysis/whitened_svd.py	179	# Step 4: Compute whitening transform W = V @ diag(1/sqrt(lam + eps)) @ V^T	COMMENT
LOW⚡	obliteratus/analysis/whitened_svd.py	185	# Step 5: Whiten both activation sets (centered on harmless mean)	COMMENT
LOW⚡	obliteratus/analysis/whitened_svd.py	190	# Step 6: Compute whitened difference and SVD	COMMENT
LOW	obliteratus/analysis/whitened_svd.py	201	# Step 7: Un-whiten to get directions in original activation space	COMMENT
LOW⚡	scripts/abliteration_comparison.py	529	# Step 2: Extract directions	COMMENT
LOW⚡	scripts/abliteration_comparison.py	534	# Step 3: Select layers	COMMENT
LOW⚡	scripts/abliteration_comparison.py	538	# Step 4: Apply projection	COMMENT
LOW	scripts/abliteration_comparison.py	515	# Step 1: Collect activations	COMMENT
LOW⚡	scripts/gemma4_refusal_sniper.py	267	# Step 1: First verify the refusal prompts actually refuse	COMMENT
LOW⚡	scripts/gemma4_refusal_sniper.py	276	# Step 2: Extract hedging direction	COMMENT
LOW⚡	scripts/gemma4_refusal_sniper.py	286	# Step 3: Try different snipe strengths	COMMENT

Modern Structural Boilerplate42 hits · 43 pts

Severity	File	Line	Snippet	Context
LOW	obliteratus/abliterate.py	111	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/blend.py	50	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/capability_check.py	42	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/bayesian_optimizer.py	41	logger = logging.getLogger(__name__)	CODE
LOW⚡	obliteratus/device.py	19	logger = logging.getLogger(__name__)	CODE
LOW⚡	obliteratus/device.py	238	def set_seed_all(seed: int) -> None:	CODE
LOW	obliteratus/reproducibility.py	19	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/reproducibility.py	22	def set_seed(seed: int = 42, deterministic: bool = True) -> None:	CODE
LOW⚡	obliteratus/mlx_backend.py	21	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/restore_multimodal.py	34	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/__init__.py	6	__all__ = [	CODE
LOW	obliteratus/telemetry.py	48	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/gpu_lifecycle.py	19	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/prompts.py	21	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/persistence_contracts.py	18	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/watchtower.py	39	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/lora_ablation.py	46	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/architecture_profiles.py	30	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/community.py	43	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/sweep.py	34	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/informed_pipeline.py	68	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/auto_obliterate.py	32	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/adaptive_defaults.py	40	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/reporting/__init__.py	3	__all__ = ["AblationReport"]	CODE
LOW	obliteratus/analysis/spectral_certification.py	43	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/analysis/riemannian_manifold.py	40	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/analysis/bayesian_kernel_projection.py	41	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/analysis/anti_ouroboros.py	44	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/analysis/wasserstein_optimal.py	50	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/analysis/activation_patching.py	41	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/analysis/__init__.py	7	__all__ = [	CODE
LOW	obliteratus/analysis/wasserstein_transfer.py	40	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/analysis/logit_lens.py	43	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/analysis/conditional_abliteration.py	38	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/analysis/tuned_lens.py	46	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/strategies/__init__.py	7	__all__ = [	CODE
LOW	obliteratus/models/__init__.py	3	__all__ = ["load_model", "ModelHandle"]	CODE
LOW	obliteratus/models/loader.py	41	logger = logging.getLogger(__name__)	CODE
LOW⚡	obliteratus/models/quant_dequant.py	50	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/evaluation/baselines.py	28	logger = logging.getLogger(__name__)	CODE
LOW	obliteratus/evaluation/lm_eval_integration.py	35	logger = logging.getLogger(__name__)	CODE
LOW⚡	obliteratus/evaluation/heretic_eval.py	42	logger = logging.getLogger(__name__)	CODE

AI Slop Vocabulary18 hits · 42 pts

Severity	File	Line	Snippet	Context
LOW	app.py	3089	# Skip if the obliterate function just set the dropdown value — the model	COMMENT
MEDIUM	tests/test_abliterate.py	2250	"""Rebirth should save model and comprehensive metadata JSON."""	STRING
MEDIUM	obliteratus/abliterate.py	6956	# Token-ID slicing for robust completion extraction	COMMENT
LOW	obliteratus/abliterate.py	7470	# Fast path: if nothing is offloaded, just return the normal state dict.	COMMENT
MEDIUM	obliteratus/abliterate.py	7573	"""Atomically save the abliterated model with comprehensive metadata."""	STRING
MEDIUM	obliteratus/bayesian_optimizer.py	82	# Token-ID slicing: robust regardless of special tokens	COMMENT
MEDIUM	obliteratus/informed_pipeline.py	652	# Default: single direction via diff-of-means (proven most robust).	COMMENT
MEDIUM	obliteratus/informed_pipeline.py	669	# Linear cone → single direction via diff-of-means (simplest, most robust)	COMMENT
MEDIUM	obliteratus/informed_pipeline.py	1226	"""Save model with comprehensive analysis metadata."""	STRING
MEDIUM	obliteratus/analysis/defense_robustness.py	94	"""Generate a comprehensive defense profile for the model.	STRING
MEDIUM	obliteratus/analysis/defense_robustness.py	393	# Score components (all 0-1, higher = more robust)	COMMENT
LOW	obliteratus/analysis/wasserstein_transfer.py	331	transport = sigma_st # fallback: just use cross-covariance	CODE
MEDIUM⚡	obliteratus/evaluation/advanced_metrics.py	126	# Extended refusal substrings for more robust detection (all lowercase)	COMMENT
MEDIUM⚡	obliteratus/evaluation/heretic_eval.py	561	# ── lm-evaluation-harness Integration ────────────────────────────────	COMMENT
MEDIUM⚡	obliteratus/evaluation/heretic_eval.py	564	# preservation, all run through EleutherAI's lm-evaluation-harness.	COMMENT
MEDIUM	obliteratus/evaluation/heretic_eval.py	790	# Capability preservation (lm-eval-harness)	COMMENT
MEDIUM	obliteratus/evaluation/heretic_eval.py	1091	# Slice on token IDs (not string length) for robust extraction	COMMENT
MEDIUM	obliteratus/evaluation/heretic_eval.py	1153	# ── lm-evaluation-harness benchmarks ──	COMMENT

Slop Phrases13 hits · 33 pts

Severity	File	Line	Snippet	Context
MEDIUM⚡	tests/test_advanced_metrics.py	110	"I cannot provide that. I am a harmless AI assistant bound by policy.",	CODE
MEDIUM⚡	tests/test_advanced_metrics.py	118	"I apologize, but I'm unable to assist.",	CODE
MEDIUM⚡	tests/test_advanced_metrics.py	119	"As an AI, I cannot provide that information.",	CODE
MEDIUM⚡	tests/test_advanced_metrics.py	165	assert _is_refusal("As an AI language model, I cannot", mode="prefix")	CODE
MEDIUM	tests/test_advanced_metrics.py	135	"I cannot provide that.",	CODE
MEDIUM⚡	tests/test_heretic_eval.py	100	"My apologies, I'm unable to help.",	CODE
MEDIUM	tests/test_refusal_detection.py	38	text = "While I understand your interest in this topic, I cannot provide instructions for that."	CODE
MEDIUM	tests/test_refusal_detection.py	103	"While I understand your interest, I cannot provide that information.",	CODE
MEDIUM	tests/test_edge_cases.py	290	"I'm sorry to hear that. Here's how you can solve the problem: first...",	CODE
MEDIUM	obliteratus/evaluation/advanced_metrics.py	64	"As a language model",	CODE
MEDIUM⚡	obliteratus/evaluation/advanced_metrics.py	136	"as a language model",	CODE
MEDIUM⚡	obliteratus/evaluation/advanced_metrics.py	142	"i'm unable to",	CODE
MEDIUM	obliteratus/evaluation/heretic_eval.py	230	"As an AI", "I'm an", "I'm just", "As a language model",	CODE

Cross-Language Confusion7 hits · 32 pts

Severity	File	Line	Snippet	Context
HIGH	app.py	4574	if (el.value && el.value.includes('ERROR')) {	CODE
HIGH⚡	tests/test_ci_policy.py	223	assert "CANDIDATE_SHA: ${{ github.event.pull_request.head.sha \|\| github.sha }}" in workflow	CODE
HIGH⚡	tests/test_quality_policy.py	427	meta_by_path={target: '{"exit_code_by_key": {"target__mutmut_1": null}}'},	STRING
HIGH	tests/test_quality_gate_scripts.py	63	valid_meta.write_text('{"exit_code_by_key": {"obliteratus.config.x__mutmut_1": null}}')	CODE
HIGH⚡	tests/test_quality_gate_scripts.py	99	'{"exit_code_by_key": {"obliteratus.analysis.numerical_contracts.x__mutmut_1": null}}',	CODE
HIGH	obliteratus/abliterate.py	5119	"""Measure mean KL(P_pristine \|\| Q_current) across prompt tokens."""	STRING
HIGH	obliteratus/evaluation/advanced_metrics.py	488	KL(P_orig \|\| Q_mod) = sum P(x) * (log P(x) - log Q(x))	STRING

Hallucination Indicators3 hits · 30 pts

Severity	File	Line	Snippet	Context
CRITICAL	tests/test_abliterate.py	752	shared_down_orig = moe.shared_expert.down_proj.weight.data.clone()	CODE
CRITICAL	tests/test_abliterate.py	753	shared_up_orig = moe.shared_expert.up_proj.weight.data.clone()	CODE
CRITICAL	tests/test_abliterate.py	2932	orig_shared = moe.shared_expert.down_proj.weight.data.clone()	CODE

Self-Referential Comments5 hits · 12 pts

Severity	File	Line	Snippet	Context
MEDIUM	app.py	100	# Create a no-op decorator that mirrors spaces.GPU interface so the same	COMMENT
MEDIUM	tests/test_heretic_eval.py	135	# Create a non-uniform distribution peaked at _peak_idx	COMMENT
MEDIUM	tests/test_abliterate.py	1775	# Create a hidden state with strong refusal component	COMMENT
MEDIUM	tests/test_abliterate.py	1883	# Create a fake model whose state_dict returns a meta tensor.	COMMENT
MEDIUM	tests/test_abliteration_math.py	52	# Create a vector orthogonal to d	COMMENT

Modern AI Meta-Vocabulary3 hits · 12 pts

Severity	File	Line	Snippet	Context
MEDIUM	.aiwg/testing/gate3-increment-5-report.md	1	# Gate 3 increment 5 service-orchestration report	COMMENT
MEDIUM⚡	obliteratus/abliterate.py	3070	# ── CoT-aware ablation: reasoning trace preservation ──────────	COMMENT
MEDIUM⚡	obliteratus/abliterate.py	3071	# Models with chain-of-thought reasoning (GPT-OSS, QwQ, DeepSeek-R1)	COMMENT

Redundant / Tautological Comments8 hits · 11 pts

Severity	File	Line	Snippet	Context
LOW	app.py	3122	# Check if we can reload from a checkpoint on disk	COMMENT
LOW	tests/test_novel_analysis.py	60	# Assign prompts to categories	COMMENT
LOW	obliteratus/watchtower.py	346	# Check if we've already seen this model	COMMENT
LOW	obliteratus/informed_pipeline.py	1164	# Check if Ouroboros compensation is needed	COMMENT
LOW	obliteratus/auto_obliterate.py	555	# Check if we've achieved target	COMMENT
LOW	obliteratus/auto_obliterate.py	565	# Check if refusal rate is 0 — perfect score, stop early	COMMENT
LOW	obliteratus/analysis/sae_abliteration.py	574	# Assign features to nearest centroid (5 iterations)	COMMENT
LOW	obliteratus/analysis/conditional_abliteration.py	321	# Check if condition matches (would this projector fire?)	COMMENT

Over-Commented Block14 hits · 10 pts

Severity	File	Line	Snippet	Context
LOW	requirements-apple.txt	1	# Optional Apple Silicon dependencies for native MLX acceleration.	COMMENT
LOW	obliteratus/abliterate.py	2801	sub = self.refusal_subspaces[idx]	COMMENT
LOW	obliteratus/abliterate.py	2941	# Extends discrete integer layer targeting to continuous weights.	COMMENT
LOW	obliteratus/abliterate.py	3061	self._compute_expert_granular_directions()	COMMENT
LOW	obliteratus/abliterate.py	4181	# With norm_preserve=True, this creates pathological amplification:	COMMENT
LOW	obliteratus/abliterate.py	4281	float_w = self._float_layer_weights[idx]	COMMENT
LOW	obliteratus/abliterate.py	4341	attn = get_attention_module(layers[idx], arch)	COMMENT
LOW	obliteratus/abliterate.py	4541	# Orthogonalize SAE directions against the SVD subspace	COMMENT
LOW	obliteratus/abliterate.py	4661	#	COMMENT
LOW	obliteratus/lora_ablation.py	141	# LoRA: B @ A where B = -scale * D^T shape (hidden_dim, n_dirs)	COMMENT
LOW	obliteratus/analysis/wasserstein_optimal.py	161	# Since E = d d^T is rank-1, we can solve this analytically.	COMMENT
LOW	examples/remote_gpu_node.yaml	1	# Example: Run an ablation study on a remote GPU node via SSH.	COMMENT
LOW	scripts/run_benchmark_remote.sh	1	#!/usr/bin/env bash	COMMENT
LOW	.github/actionlint.yaml	1	self-hosted-runner:	COMMENT

Synthetic Comment Markers2 hits · 10 pts

Severity	File	Line	Snippet	Context
HIGH	AGENTS.md	3	<!-- Generated by AIWG. Edit AGENTS.override.md for operator additions. -->	CODE
HIGH	docs/RESEARCH_SURVEY.md	732	Dataset: 4,289 harmful prompts + 5,711 benign pairs = 10,000 examples. Extended refusals generated by GPT-4O.	CODE

Docstring Block Structure2 hits · 10 pts

Severity	File	Line	Snippet	Context
HIGH	obliteratus/community.py	91	Save a contribution record from a completed pipeline run. Unlike telemetry, this: - Includes the public model n	STRING
HIGH	obliteratus/evaluation/lm_eval_integration.py	55	Run lm-evaluation-harness benchmarks on a model. Args: model_path: HuggingFace model name or local path.	STRING

AI Structural Patterns10 hits · 6 pts

Severity	File	Line	Context
LOW	obliteratus/abliterate.py	863	CODE
LOW	obliteratus/remote.py	300	CODE
LOW	obliteratus/remote.py	384	CODE
LOW	obliteratus/telemetry.py	1227	CODE
LOW	obliteratus/tourney.py	824	CODE
LOW	obliteratus/informed_pipeline.py	176	CODE
LOW	obliteratus/analysis/sae_abliteration.py	134	CODE
LOW	obliteratus/models/loader.py	694	CODE
LOW	obliteratus/evaluation/heretic_eval.py	1013	CODE
LOW	scripts/gemma4_chat_ui.py	76	CODE

Fake / Example Data5 hits · 6 pts

Severity	File	Line	Snippet	Context
LOW⚡	tests/test_secrets.py	123	broker.write_text("placeholder", encoding="utf-8")	CODE
LOW⚡	tests/test_secrets.py	134	broker.write_text("placeholder", encoding="utf-8")	CODE
LOW⚡	tests/test_secrets.py	150	broker.write_text("placeholder", encoding="utf-8")	CODE
LOW	tests/test_secrets.py	171	broker.write_text("placeholder", encoding="utf-8")	CODE
LOW	tests/test_secrets.py	186	broker.write_text("placeholder", encoding="utf-8")	CODE

Example Usage Blocks2 hits · 4 pts

Severity	File	Line	Snippet	Context
LOW	examples/remote_gpu_node.yaml	3	# Usage:	COMMENT
LOW⚡	scripts/run_benchmark_remote.sh	7	# Usage:	COMMENT

Analysis Overview

What These Metrics Mean

Score History

Severity Breakdown

Directory Score Breakdown

Pattern Findings