Home / stas00 / ml-engineering
AI Slop Code RIsk Monitor

stas00/ml-engineering

Machine Learning Engineering Open Book

5.5 Moderate pattern signal View on GitHub

Maintenance and Governance Review

This report summarizes scanner evidence for stas00/ml-engineering, a Python project with 19,445 GitHub stars. SynthScan examined 29,980 lines of code across 110 source files, recording 127 pattern matches distributed across 14 syntactic categories. The legacy adjusted pattern signal is 5.5 (Moderate pattern signal); it is not an authorship estimate.

Use the findings to ask whether maintainers can explain intent, verify dependencies, simplify unnecessary scaffolding, and focus review effort. Risk Payload v1 adds ownership, supply-chain, over-engineering, and review-effort vectors when available. These are prompts for human verification, not probabilities, security findings, or measured maintenance costs.

5.5
Adjusted Signal
5.5
Raw Pattern Score
100%
Time Factor
2026-10-08
Last Push
19.4K
Stars
Python
Language
30.0K
Lines of Code
110
Files
127
Pattern Hits
2026-10-12
Scan Date
0.05
HC Hit Rate

Risk Review · v3

Use each vector to guide manual verification of the underlying findings.

Cognitive Ownership Risk 12.5/100 · LOW

Can maintainers explain intent, assumptions, and trade-offs? 61 finding(s).

Hallucinated Dependency & Supply Chain Risk 7.9/100 · LOW

Are imports declared and package/API names verified? 43 finding(s).

Agent Over-Engineering Ratio 7.9/100 · LOW

Are wrappers, guards, and repeated structures necessary? 40 finding(s).

PR Review Velocity Impact 25.8/100 · MEDIUM

Which changes need focused human review? 127 finding(s).

What These Metrics Mean

Adjusted Signal
Legacy pattern score normalized per 1,000 scanned lines and multiplied by the repository-level time factor. It is not an estimate of AI authorship.
Raw Pattern Score
Sum of severity- and context-weighted pattern findings before applying the time factor.
Time Factor
A coarse multiplier based on repository last-push metadata. It does not identify when individual code was written.
Pattern Hits
Count of configured pattern findings. Review each location and surrounding code before acting.
HC Hit Rate
High- and critical-severity findings per scanned file. Severity is a configured detector weight, not proof of a vulnerability or authorship.
Lines / Files
Scan coverage denominators used by the legacy normalized score.

Pattern-Signal History

This chart tracks the legacy adjusted pattern score across scan runs. Changes can reflect source changes, scan coverage, or detector/scoring versions; they do not measure AI adoption or prove that a team refactored code. Compare runs only when their scoring versions are compatible. The dashed line tracks the raw pattern-hit count.

Severity Breakdown

Groups findings by configured severity weight. Severity is not a confidence estimate, security rating, or proof that a package is missing. Verify imports against the project's manifests and inspect each finding in its local design context.

CRITICAL 0HIGH 6MEDIUM 11LOW 110

Directory Score Breakdown

This horizontal bar chart shows where configured pattern scores accumulate across directories. Treat the bars as locations to inspect, not as evidence of authorship or code quality. Use the file-level findings to decide whether a directory needs a focused ownership, dependency, complexity, or review check.

Pattern Findings

The scanner identified 127 distinct pattern matches across 14 syntactic categories. Each entry below represents a discrete location in the source code where the engine recorded a detector match to inspect in context. 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 configured finding weight (CRITICAL / HIGH / MEDIUM / LOW). The Context column identifies whether the match occurred inside executable code, an inline comment, or a string literal — legacy comment-context matches receive a ×1.5 score weight. The ⚡ bolt icon marks clustered matches: three or more patterns within a 10-line window, each receiving an additional ×1.5 density multiplier under the legacy scoring model.

Unused Imports40 hits · 40 pts
SeverityFileLineSnippetContext
LOWtraining/fault-tolerance/slurm-status.py13CODE
LOWtraining/fault-tolerance/slurm-status.py14CODE
LOWtraining/fault-tolerance/slurm-status.py15CODE
LOWtraining/fault-tolerance/slurm-status.py16CODE
LOWtraining/fault-tolerance/slurm-status.py21CODE
LOWtraining/fault-tolerance/fs-watchdog.py16CODE
LOWnetwork/benchmarks/torch-dist-bench.py106CODE
LOWnetwork/benchmarks/torch-dist-bench.py122CODE
LOWstorage/fio-json-extract.py13CODE
LOWdebug/tiny-scripts/general-pmd-ds-unpack.py21CODE
LOWdebug/tiny-scripts/general-pmd-ds-unpack.py22CODE
LOWdebug/tiny-scripts/general-pmd-ds-unpack.py24CODE
LOWdebug/tiny-scripts/general-pmd-ds-unpack.py25CODE
LOWdebug/tiny-scripts/general-pmd-ds-unpack.py26CODE
LOWdebug/tiny-scripts/general-pmd-ds-unpack.py27CODE
LOWdebug/tiny-scripts/general-pmd-ds-unpack.py28CODE
LOWdebug/tiny-scripts/general-pmd-ds-unpack.py29CODE
LOWdebug/tiny-scripts/general-pmd-ds-unpack.py30CODE
LOWdebug/tiny-scripts/m4-ds-unpack.py24CODE
LOWdebug/tiny-scripts/m4-ds-unpack.py26CODE
LOWdebug/tiny-scripts/m4-ds-unpack.py27CODE
LOWdebug/tiny-scripts/m4-ds-unpack.py28CODE
LOWdebug/tiny-scripts/m4-ds-unpack.py29CODE
LOWdebug/tiny-scripts/m4-ds-unpack.py30CODE
LOWdebug/tiny-scripts/m4-ds-unpack.py31CODE
LOWdebug/tiny-scripts/m4-ds-unpack.py32CODE
LOWdebug/tiny-scripts/general-pmd-synthetic-testing.py52CODE
LOWdebug/tiny-scripts/general-pmd-synthetic-testing.py53CODE
LOWdebug/tiny-scripts/general-pmd-synthetic-testing.py56CODE
LOWdebug/tiny-scripts/general-pmd-synthetic-testing.py58CODE
LOWdebug/tiny-scripts/general-pmd-synthetic-testing.py59CODE
LOWdebug/tiny-scripts/cm4-synthetic-testing.py52CODE
LOWdebug/tiny-scripts/cm4-synthetic-testing.py56CODE
LOWdebug/tiny-scripts/cm4-synthetic-testing.py58CODE
LOWdebug/tiny-scripts/cm4-synthetic-testing.py59CODE
LOWdebug/tiny-scripts/mt5-make-tiny-model.py28CODE
LOWdebug/tiny-scripts/mt5-make-tiny-model.py29CODE
LOWdebug/tiny-scripts/mt5-make-tiny-model.py30CODE
LOWdebug/tiny-scripts/mt5-make-tiny-model.py32CODE
LOWdebug/tiny-scripts/mt5-make-tiny-model.py33CODE
Over-Commented Block19 hits · 19 pts
SeverityFileLineSnippetContext
LOWtraining/tools/printflock.py1# If you have ever done multi-gpu work and tried to `print` for debugging you quickly discoveredCOMMENT
LOWtraining/tools/multi-gpu-non-interleaved-print.py1#!/usr/bin/env pythonCOMMENT
LOWtraining/checkpoints/torch-checkpoint-shrink.py1#!/usr/bin/env pythonCOMMENT
LOWtraining/performance/benchmarks/numa/numa-set.sh1#!/usr/bin/bashCOMMENT
LOWnetwork/benchmarks/all_gather_object_vs_all_gather.py1#!/usr/bin/env pythonCOMMENT
LOWstorage/mmap-io-bench.py1#!/usr/bin/env python3COMMENT
LOWtesting/testing_utils.py1# I developed the bulk of this library while I worked at HFCOMMENT
LOWcompute/accelerator/benchmarks/mamf-finder.py101COMMENT
LOWcompute/accelerator/benchmarks/mamf-finder.py721 )COMMENT
LOWcompute/accelerator/benchmarks/mamf-finder.py1501 start = stepCOMMENT
LOWcompute/accelerator/benchmarks/mamf-finder.py2281 finish()COMMENT
LOWdebug/underflow_overflow.py1# Copyright 2020 The HuggingFace Team. All rights reserved.COMMENT
LOWdebug/tiny-scripts/openwebtext-10k.py1# coding=utf-8COMMENT
LOWdebug/tiny-scripts/general-pmd-ds-unpack.py1# unpack the desired datasets records into a filesystem-based subdir structure which can then beCOMMENT
LOWdebug/tiny-scripts/oscar-en-10k.py1# coding=utf-8COMMENT
LOWdebug/tiny-scripts/m4-ds-unpack.py1# unpack the desired datasets records into a filesystem-based subdir structure which can then beCOMMENT
LOWdebug/tiny-scripts/c4-en-10k.py1# coding=utf-8COMMENT
LOWdebug/tiny-scripts/fsmt-make-super-tiny-model.py1#!/usr/bin/env pythonCOMMENT
LOWdebug/tiny-scripts/mt5-make-tiny-model.py1#!/usr/bin/env pythonCOMMENT
Excessive Try-Catch Wrapping18 hits · 18 pts
SeverityFileLineSnippetContext
LOWtraining/emulate-multi-node.md149except Exception:CODE
LOWtesting/testing_utils.py46except Exception:CODE
LOWtesting/README.md797 except Exception as e:CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py286 except Exception:CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py296 except Exception:CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py308 except Exception:CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py1132 except Exception:CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py1552 except Exception as e:CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py1565 except Exception as e:CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py1575 except Exception:CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py1591 except Exception:CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py1602 except Exception:CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py1613 except Exception:CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py1622 except Exception:CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py1651 except Exception:CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py1713 except Exception as e:CODE
LOWdebug/torch-distributed-gpu-test.py110except Exception:CODE
LOWdebug/code/see-mem-usage.py22except Exception:CODE
Modern AI Meta-Vocabulary7 hits · 16 pts
SeverityFileLineSnippetContext
MEDIUMtraining/fault-tolerance/README.md188The section on [Crontab Emulation](../../orchestration/slurm/users.md#crontab-emulation) discusses how to implement cronCODE
MEDIUMcompute/cpu/README.md53footnote: A [Nebius agentic-RL writeup](https://nebius.com/blog/posts/in-agentic-rl-faster-tokens-are-not-enough) (2026-CODE
MEDIUMcompute/accelerator/README.md920- [SN50](https://sambanova.ai/blog/introducing-the-sn50-rdu-purpose-built-for-agentic-inference) - fifth generation, annCODE
MEDIUMcompute/accelerator/README.md1025Crescent Island is an inference-oriented Xe3P GPU with customer sampling in H2-2026. Intel's card is 160GB LPDDR5X (ODMsCODE
MEDIUMcompute/accelerator/benchmarks/mamf-finder.py2293 # These two bounds are internal guardrails, not tuning dials, so they're not exposed on the CLI:COMMENT
MEDIUMorchestration/README.md25## Other orchestration frameworksCOMMENT
MEDIUMinference/README.md379It's the process of giving the pre-trained model additional information that wasn't available during its training. For eCODE
AI Response Leakage3 hits · 15 pts
SeverityFileLineSnippetContext
HIGHtraining/model-parallelism/README.md172[ZeRO++](https://arxiv.org/abs/2306.10209) solves the 2nd limitation by introducing Hierarchical Weight Partition for ZeCODE
HIGHinference/README.md330One other much simpler solution for [input-grounded tasks](#input-grounded-tasks), is to use [ngram prompt lookup decodiCODE
HIGHinference/README.md330One other much simpler solution for [input-grounded tasks](#input-grounded-tasks), is to use [ngram prompt lookup decodiCODE
Hyper-Verbose Identifiers17 hits · 14 pts
SeverityFileLineSnippetContext
LOWtraining/tools/main_process_first.py76def main_process_by_path_first(path):CODE
LOWtraining/tools/main_process_first.py103def global_main_process_first():CODE
LOWtraining/fault-tolerance/slurm-status.py44def send_email_alert_job_not_scheduled(job_name):STRING
LOWtraining/fault-tolerance/slurm-status.py63def check_running_on_jean_zay():STRING
LOWtraining/fault-tolerance/fs-watchdog.py54def check_running_on_jean_zay():STRING
LOWtraining/fault-tolerance/fs-watchdog.py114 def analyse_partition_idrquota(partition_name, partition_flag, alert_bytes_threshold, alert_inodes_threshold):STRING
LOWnetwork/benchmarks/torch-dist-bench.py350def latency_samples_per_trial(calls):CODE
LOWtesting/testing_utils.py128def require_torch_non_multi_gpu(test_case):CODE
LOWtesting/testing_utils.py203def require_bnb_non_decorator():CODE
LOWtesting/testing_utils.py308def parameterized_custom_name_func_join_params(func, param_num, param):CODE
LOWtesting/testing_utils.py323 def test_determinism_wrt_rank(self, num_workers, pad_dataset):STRING
LOWtesting/testing_utils.py777 def get_auto_remove_tmp_dir_str(self, *args, **kwargs):CODE
LOWtesting/testing_utils.py954def pytest_terminal_summary_main(tr, id):CODE
LOWtesting/README.md339def test_example_with_multi_gpu():CODE
LOWtesting/README.md581 def test_something_involving_local_locations(self):CODE
LOWtesting/README.md592 def test_something_involving_stringified_locations(self):CODE
LOWdebug/NicerTrace.py122 def localtrace_trace_and_count(self, frame, why, arg):CODE
Deep Nesting13 hits · 12 pts
SeverityFileLineSnippetContext
LOWtraining/fault-tolerance/slurm-status.py118CODE
LOWtraining/performance/benchmarks/torch-compile-bench.py23CODE
LOWnetwork/benchmarks/torch-dist-bench.py359CODE
LOWtesting/testing_utils.py954CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py1340CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py1758CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py2173CODE
LOWcompute/accelerator/benchmarks/mamf-finder.py2210CODE
LOWcompute/accelerator/benchmarks/mamf-finder-all-gpus.py221CODE
LOWdebug/NicerTrace.py73CODE
LOWdebug/underflow_overflow.py214CODE
LOWdebug/tiny-scripts/general-pmd-synthetic-testing.py124CODE
LOWdebug/tiny-scripts/general-pmd-synthetic-testing.py134CODE
Empty Error Handlers2 hits · 10 pts
SeverityFileLineSnippetContext
HIGHnetwork/benchmarks/torch-dist-bench.py125except ModuleNotFoundError:CODE
HIGHcompute/accelerator/benchmarks/mamf-finder.py76except ModuleNotFoundError:CODE
Cross-Language Confusion1 hit · 8 pts
SeverityFileLineSnippetContext
HIGHdebug/tiny-scripts/cm4-synthetic-testing.py19# perl -le 'qx[touch texts_$_.null] for map {sprintf "%02d", $_} 0..38'COMMENT
Decorative Section Separators2 hits · 6 pts
SeverityFileLineSnippetContext
MEDIUMinsights/when-to-upgrade-gpus/bench_decompose.py54# -----------------------------------------------------------------------------------------COMMENT
MEDIUMinsights/when-to-upgrade-gpus/benchmark.py41# -----------------------------------------------------------------------------------------COMMENT
Slop Phrases1 hit · 3 pts
SeverityFileLineSnippetContext
MEDIUMstorage/mmap-io-bench.py21# IMPORTANT: edit the paths to the target folders to reflect the actual NVME, Network fs (e.g. Lustre), and you can add COMMENT
Fake / Example Data2 hits · 2 pts
SeverityFileLineSnippetContext
LOWtraining/fault-tolerance/slurm-status.py28TO_ADDRS = ["bigscience@googlegroups.com", "foo@bar.com"] # wants a listCODE
LOWtraining/fault-tolerance/fs-watchdog.py22TO_ADDRS = ["bigscience@googlegroups.com", "foo@bar.com"] # wants a listCODE
Modern Structural Boilerplate1 hit · 1 pts
SeverityFileLineSnippetContext
LOWinsights/when-to-upgrade-gpus/benchmark.py116def set_seed(seed: int) -> None:CODE
AI Slop Vocabulary1 hit · 1 pts
SeverityFileLineSnippetContext
MEDIUMtesting/testing_utils.py397 # environment, so to make the test robust a combined entry of both streams is availableSTRING