Research notes

This project grew out of a literature survey of 19 NBA advanced metrics. Those notes live in the repository under docs/research/ and are kept unmodified as a record of what was believed at the start.

They are background, not authority. Three cautions apply when reading them:

1. Some citations no longer resolve. The reliability figure quoted for EPM rests on a URL that now returns 404. Where a claim matters, this project measures it rather than citing it. See Measuring reliability.
2. Many sources are informal. Roughly a fifth of the bibliography is Reddit threads, against two arXiv papers and one working paper as peer-reviewed anchors.
3. The reliability tiers are judgment, not measurement. The methodology note says so itself: assigning the scores "involves some judgment."

The files, in the repository:

File	Contents
01_development_pipeline.md	Data preparation, features, training, validation
02_deployment_stack.md	Serving and monitoring options
03_ai_tools_assistance.md	Tooling survey by project stage
04_expansion_ideas.md	Scope expansion sketches
05_references.md	Bibliography, SRC-139 to SRC-183
06_data_sources.md	Data source comparison
07_data_categorization.md	Categorisation framework
08_data_pipeline_compliance.md	Pipeline phases, rate limits, compliance
nba_metrics_detailed_report.md	Per-metric profiles
nba_metrics_grouped_report.md	Metrics grouped by type
nba_metrics_reliability_report.md	Metrics by reliability tier
nba_metrics_reliability_methodology.md	How the tiers were assigned
nba_metric_sources.md	Bibliography, SRC-1 to SRC-138