experiment_logs

---

iso_doc_type: "Report" iso_viewpoint: "DeploymentView" type: "operations" title: "Experiment Tracking & Telemetry Logs" description: "ISO 42010 DeploymentView / ISO 15289 Report documentation detailing the MLflow experiment taxonomy, metric schemas, and artifact layout." tags: ["iso42010", "okf", "report", "experiment_logs", "mlflow"] timestamp: "2026-10-01T17:50:00Z" generated: "agent:wiki-iso-documentation-standard" verified: "true" last_verified_commit: "073b5fd"

Experiment Tracking & Telemetry Logs

Purpose: Detail the telemetry schema, parameter tracking conventions, and artifact storage hierarchies maintained inside MLflow.

---

1. Experiment Schema & Naming Conventions

All runs are organized hierarchically under defined MLflow experiments:

mlruns/
└── <experiment_id>/
    └── <run_id>/
        ├── params/               # Model hyperparameters & split settings
        │   ├── max_depth
        │   ├── n_estimators
        │   └── test_size
        ├── metrics/              # Time-series & scalar evaluation results
        │   ├── train_mse
        │   ├── test_mse
        │   ├── test_rmse
        │   └── test_r2
        ├── artifacts/            # Serialized models and attestation files
        │   ├── model/            # MLmodel binary bundle
        │   ├── explanations/     # SHAP summary plots & feature rankings
        │   └── signatures/       # SHA-256 InferSigner receipts
        └── meta.yaml             # Git commit SHA, author, and timestamp

---

2. Standard Metric Logging Schema

Metric Key	Mathematical Definition	Evaluation Target	Logged In Job
mse	Mean Squared Error	Minimization	TrainingJob, EvaluationsJob
rmse	Root Mean Squared Error	Minimization	TrainingJob, EvaluationsJob
mae	Mean Absolute Error	Minimization	TrainingJob, EvaluationsJob
r2	Coefficient of Determination	Maximization ($ o 1.0$)	TrainingJob, EvaluationsJob
trial_score	Validation score per trial	Optimization objective	TuningJob

---

3. Engineering Lab Notebook Experiments Catalog

Experiment ID	Title / Focus	Deliverable / Artifact Location	Primary Findings
EXP-2026-008	Deterministic Verification vs. LLM-as-a-Judge Bias	evidence/evaluations/llm-judge-vs-pytest-benchmark.json · Lab Notebook Notebook	LLM judge had 44% False Negative rate (11/25 defects missed); Pytest harness detected 100% of defects.

---

Related: Lab Notebook EXP-2026-008 · Production Operations · Tactical Design · Master Index