A browser-only ancestry fitting tool built from the cleaned individual-level reference panel. V5 uses a 21-cluster data-driven broad scaffold derived from the fully labeled individual-level reference panel, then uses leave-one-out-validated population leaves for conservative fine-resolution refinement and optional trace detection.
23andMe / AncestryDNA / Living DNA style TXT or CSV files are supported, including .gz and .zip. The genotype stays in this browser.
| Reference profile | Percent | 95% fixed-source interval | Exact-source rediscovery |
|---|
Independent coarse fit to broad reference centroids. Useful for orientation and sanity checks, not a demographic reconstruction.
A selected parent component can be redistributed among genetically validated published populations plus a parent residual. The parent total is fixed, and a split is shown only when it improves held-out genomic-block likelihood.
When closely related parent profiles can substitute for one another with little held-out likelihood cost, V5 shows that ambiguity instead of hiding it.
The trace source is allowed to enter even when the main model already uses all of its selected slots. This is specifically designed to catch small signals that sparse model selection can suppress.
| Best proxy | Broad axis | Residual estimate | 95% block interval | Chromosome support | Δ log L | Stability |
|---|
PCA is calculated from the bundled reference profiles on a compact deterministic subset of the pre-LD-pruned marker panel. The red point is the uploaded genotype; the blue square is the point predicted by the main likelihood mixture.
Diagnostic fit over the 21 data-driven broad profiles on a compact marker panel. Literature-resolved children are deliberately excluded here so nested proxies cannot fragment the weights.