Experiments
Machine-readable registry of every executed experiment — 22 entries, sourced from the generated evidence registry. Registered experiment families: DRIFT, BASE, XAI, EXP.
DRIFT-FIXED-B1-001
How does feature drift evolve from batch 1?
Metrics: results/drift/feature_drift_by_batch.csv
Univariate normalized Wasserstein and standardized mean shift; median aggregation.
BASE-FIXED-C1-001
Classical temporal generalization and adaptation
Metrics: results/baselines/fixed_origin_metrics.csv
Metrics recomputed from saved raw predictions.
BASE-FIXED-C2-001
Classical temporal generalization and adaptation
Metrics: results/baselines/fixed_origin_metrics.csv
Metrics recomputed from saved raw predictions.
BASE-FIXED-C3-001
Classical temporal generalization and adaptation
Metrics: results/baselines/fixed_origin_metrics.csv
Metrics recomputed from saved raw predictions.
BASE-FIXED-C4-001
Classical temporal generalization and adaptation
Metrics: results/baselines/fixed_origin_metrics.csv
Metrics recomputed from saved raw predictions.
BASE-EXPAND-C1-001
Classical temporal generalization and adaptation
Metrics: results/baselines/expanding_window_metrics.csv
Metrics recomputed from saved raw predictions.
BASE-EXPAND-C2-001
Classical temporal generalization and adaptation
Metrics: results/baselines/expanding_window_metrics.csv
Metrics recomputed from saved raw predictions.
BASE-EXPAND-C3-001
Classical temporal generalization and adaptation
Metrics: results/baselines/expanding_window_metrics.csv
Metrics recomputed from saved raw predictions.
BASE-EXPAND-C4-001
Classical temporal generalization and adaptation
Metrics: results/baselines/expanding_window_metrics.csv
Metrics recomputed from saved raw predictions.
BASE-IID-C1-001
Classical temporal generalization and adaptation
Metrics: results/baselines/iid_diagnostic_metrics.csv
Metrics recomputed from saved raw predictions.
BASE-IID-C2-001
Classical temporal generalization and adaptation
Metrics: results/baselines/iid_diagnostic_metrics.csv
Metrics recomputed from saved raw predictions.
BASE-IID-C3-001
Classical temporal generalization and adaptation
Metrics: results/baselines/iid_diagnostic_metrics.csv
Metrics recomputed from saved raw predictions.
BASE-IID-C4-001
Classical temporal generalization and adaptation
Metrics: results/baselines/iid_diagnostic_metrics.csv
Metrics recomputed from saved raw predictions.
EXP-XAI-0001-C1
What lightweight explanation representations can be generated for the existing drift-robust model candidates, and which representations are practical enough to evaluate later for fidelity, stability, latency, and TinyML deployment?
Metrics: results/xai/stage09_global_importance.csv
Stage 09 resource-aware XAI. No retraining; loads frozen FIXED_ORIGIN model artifact. Explanations saved under artifacts/explanations/EXP-XAI-0001/. Method applicability: results/xai/stage09_manifest.csv.
EXP-XAI-0001-C2
What lightweight explanation representations can be generated for the existing drift-robust model candidates, and which representations are practical enough to evaluate later for fidelity, stability, latency, and TinyML deployment?
Metrics: results/xai/stage09_global_importance.csv
Stage 09 resource-aware XAI. No retraining; loads frozen FIXED_ORIGIN model artifact. Explanations saved under artifacts/explanations/EXP-XAI-0001/. Method applicability: results/xai/stage09_manifest.csv.
EXP-XAI-0001-C3
What lightweight explanation representations can be generated for the existing drift-robust model candidates, and which representations are practical enough to evaluate later for fidelity, stability, latency, and TinyML deployment?
Metrics: results/xai/stage09_global_importance.csv
Stage 09 resource-aware XAI. No retraining; loads frozen FIXED_ORIGIN model artifact. Explanations saved under artifacts/explanations/EXP-XAI-0001/. Method applicability: results/xai/stage09_manifest.csv.
EXP-XAI-0001-C4
What lightweight explanation representations can be generated for the existing drift-robust model candidates, and which representations are practical enough to evaluate later for fidelity, stability, latency, and TinyML deployment?
Metrics: results/xai/stage09_global_importance.csv
Stage 09 resource-aware XAI. No retraining; loads frozen FIXED_ORIGIN model artifact. Explanations saved under artifacts/explanations/EXP-XAI-0001/. Method applicability: results/xai/stage09_manifest.csv.
EXP-XAI-FIDELITY-001
Do Stage-09 explanations faithfully identify features that alter frozen model behavior under chronological drift?
Metrics: results/xai/stage10_fidelity_summary.csv
Frozen models and Batch-1 scaler means; 30 matched controls; 1000 bootstrap replicates; ablation results labelled consistency.
EXP-XAI-STABILITY-001
How stable are Stage-09 explanations under chronological sensor drift after conditioning on input and model change?
Metrics: results/xai/stage11_stability_summary.csv
No retraining; chronological permutation vectors; natural-neighbor and one-to-one matched cross-sectional local analysis; 1000 bootstrap replicates.
EXP-XAI-LATENCY-001
What is the reproducible controlled host cost and computational accounting of each Stage-09 explanation?
Metrics: results/xai/stage12_latency_summary.csv
Controlled single-thread host timing only; raw ns retained; operation counts separate; MCU latency, energy, Flash, and SRAM not measured.
EXP-EMBED-FP32-EQUIV-001
Can C1 and C4 standalone host-compiled FP32 implementations satisfy frozen Stage-13 numerical equivalence?
Metrics: results/embedded/stage14_summary.json
Scientific outcome FAILED: both candidates failed frozen preprocessing criteria despite 100% golden and boundary decisions. No quantization or MCU execution.
EXP-EMBED-C1-PREPROC-REPAIR-001
Can prospectively defined all-FP32 explicit StandardScaler arithmetic repair C1 under unchanged equivalence criteria?
Metrics: results/embedded/stage14r_candidate_summary.csv
Scientific outcome FAILED; no P0-P4 candidate passed universal frozen preprocessing criteria. Stage-14 failure preserved; no C4 modification, quantization, or MCU execution.