Benchmark design figures

Vector PDF figures of the benchmark design (structure, pipelines, data flow, generative and ML pipelines, statistics). They show no results. Each figure has a Markdown file with the same name that contains a suggested caption and a description for the paper.

#	Figure	Topic
1	fig01_architecture	layered software architecture
2	fig02_stage_pipeline	pipeline stages, outputs, environments
3	fig03_environments	dependency isolation, hand-off via Parquet
4	fig04_data_preprocessing	raw data → windows, leakage boundary
5	fig05_temporal_split	split, embargoes, role of each part
6	fig06_pool_generator_vae	β-VAE pool generator
8	fig08_method_family	the compared methods (16 in min, 17 in mid, 35 in max)
9	fig09_long_generators	Diffusion-TS, Chronos, TimesFM and Timestral baselines
10	fig10_utility_pipeline	TSTR vs TRTR protocol
11	fig11_tasks_models	downstream tasks × models
12	fig12_fidelity_evaluation	fidelity inputs, three references, metric groups
13	fig13_statistical_pipeline	statistical analysis: ranking, comparisons, associations, confirmatory family
14	fig14_hypotheses_map	research questions RQ0–RQ9, C1–C4, evidence, analyses
15	fig15_incremental_recompute	unit fingerprints, caching
16	fig16_artifact_layout	output layout
17	fig17_profiles	run profiles min / mid / max
18	fig18_extensibility	registries and plugins

Regenerating

python docs/figures/src/make_figures.py      # needs Graphviz (`dot`) and matplotlib

The Graphviz sources are written to src/dot/. All figures share one colour code: blue = real data, orange = generative models, yellow = assembly, pink = fidelity, green = downstream ML, purple = statistics, grey = infrastructure and storage. Fills differ in lightness, so they stay distinguishable in grayscale. Fonts are embedded (Helvetica, TrueType).

Numbers in the figures follow the code defaults (marathonts/bench/config.py) and the profile min unless a figure says otherwise. Update the figures when these defaults change.