Expand description
Evidence-based flag advisors: each module inspects the analysis result
(and, where useful, trial-encodes the loader sample through the real
encoder) and returns Advice entries for stt-build flags beyond the
zoom/bucket basics.
Contract: every advisor exposes
pub fn advise(result: &AnalysisResult, data: &LoadedData) -> anyhow::Result<Vec<Advice>>.
Advice with lossy: true is NEVER auto-applied by stt-build --auto and
never lands in to_command — it is surfaced loudly for the user to opt in
(the no-thinning principle).
Modules§
- budget
- Tile-budget / aggregation advisor (
--maximum-tile-features,--summary-tier,--min-zoom-field): flags deep-zoom oversized tiles from the density model, suggests an additive summary tier for very large dense point sets, and points hotspot-skewed data at per-feature LOD floors. - layout
- Wire/layout advisor (
--publish(zstd 19),--blob-ordering,--pack-size): compression measured on the sample, ordering from the spatial/temporal access shape. - quantize
- Coordinate / attribute quantization advisor (
--quantize-coords,--quantize-attrs-auto): derives candidate precision from the max-zoom ground resolution and verifies the win by trial-encoding the sample. - temporal
- Temporal shaping advisor (
--temporal-lod,--adaptive-temporal): tier and window suggestions from the temporal distribution and bucket count.
Structs§
- Advice
- One evidence-based flag recommendation.
Enums§
- Advice
Confidence - Advisor confidence in a recommendation.
Functions§
- run_all
- Run every advisor. Order: quantize, temporal, layout, budget.