A Universal, Accession-Maximized SNP Panel for Cannabis sativa: Meta-Analysis of Twenty-Two Published Marker Sets, a Universal SNP Panel and the Case for Multi-Allelic Anchors

Cannabis genetics has produced a decade of SNP panels that cannot speak to one another, and we show that the reason runs deeper than competing coordinate systems. Building a first liftOver chain between the field's two principal assemblies — Purple Kush ASM23057v4 and CBDRx/cs10 — over ten multi-allelic loci, validated with CrossMap at 93.6–99.8% flank identity, we find that the two assemblies share no chromosome numbering: Purple Kush chromosomes 1, 2, 6 and 9 correspond to cs10 chromosomes 5, 6, 7 and 8 respectively, and not one maps to the cs10 chromosome of its own number. Because both assemblies use the same vocabulary, any comparison made between them on chromosome labels alone has been comparing different chromosomes. The chain reconciles their apparently conflicting placements of the terminal cannabinoid synthase cluster, and establishes that none of the ten loci falls within the capture space of CannSelect, the field's largest genotyping platform — a deployment question unanswerable before the chain existed. We inventoried 22 published marker sets, preprints and databases spanning 4,099 accession records (a conservative unique union of ~3,300–3,800 accessions), retrieved or recomputed marker lists for every set that could be obtained — including probe-by-probe parsing of the 18,040-probe CannSelect v1.0 target-enrichment design — and tested them against each other at the exact-position level. Overlap was zero in six of eight panel-to-panel tests. Critically, this holds within a single reference assembly: Cull & Joly (20 SNPs), HASCH (1,504 targets) and CannSelect (18,040 probes, ~1.03 Mb of capture space) are all specified on CBDRx/cs10, yet CannSelect captures only 10 of 1,504 HASCH SNPs (0.7%) and none of the 20 Cull & Joly SNPs. Positional disjointness is therefore not merely an artefact of competing assemblies but a consequence of independent discovery populations and thinning rules operating within the same coordinate frame. We correct two coordinate assignments in the prior literature. The 344 structural SNPs of Jin et al. (2021) are on the Purple Kush assembly ASM23057v4, not cs10, verified against GenBank, which is what made the chain necessary. And the Henry et al. (2020) KASP panel, which we re-place marker by marker from its original assay flanking sequences, is not in a separate coordinate family as previously described but in cs10 — reported under chromosome numbers that are the ordinal index of the cs10 RefSeq accession, so that seven of its ten chromosome labels are wrong. Of its 21 testable nuclear markers, eleven reproduce at their published position at 96.2–100% identity and confirm that key without exception; nine do not reproduce there at all and resolve elsewhere in cs10 at 93.0–99.4%, and we supply their verified positions; one lies in a repeat family and is left provisional. The marker names themselves prove to encode draft-assembly scaffold and offset, which supplies an independent check: two markers 554 bp apart in the draft are 554 bp apart in our cs10 placement. Three markers carry near-identical paralogues elsewhere in the genome, a specificity caveat for anyone running the assays. Relabelled and re-placed, the panel becomes a fourth cs10 panel and behaves like the others: one of 1,504 HASCH markers falls inside its verified amplicon intervals, none of Cull & Joly's 20 do. That single HASCH marker, four CannSelect capture regions and the Henry marker SVIP9 converge within 200 bp on chromosome 5 — the only such convergence in the inventory. Beneath the positional divergence the panels converge functionally on the THCAS/CBDAS synthase region, where Henry's VSSL_A250D KASP marker and Kannapedia's THCAS c.749C>A tag the identical inactive-synthase Ala250Asp allele, and where even CannSelect's curated gene set centres on the cannabinoid synthases. A universal panel therefore cannot be assembled by concatenating published lists and must be built as modules in each laboratory's native coordinate system, anchored by functional loci that translate across assemblies. We propose a 60-marker Tier-1 universal core in four modules, every marker published and genotype-validated, with Tier-2 high-density backbones for NGS-capable laboratories. The Henry 2020 KASP core alone achieves P(collision) = 4.5 × 10⁻⁷ across 420 accessions; adding ten multi-allelic anchors takes it to 2.1 × 10⁻¹⁴. We report a caveat not visible in the source data: five of those ten anchors lie within a single 2.45 Mb block of chromosome 1, violating the independence assumed by the chain rule, with a sensitivity estimate of approximately 1.8 × 10⁻¹¹ under the conservative assumption that the block behaves as one locus. We then validate Module A in silico on its full public genotype matrix (420 accessions, 22 markers): principal components recover the published population structure, leave-one-out cluster assignment reaches 84.0% against a 37.1% baseline, and identity markers and synthase anchors prove complementary rather than redundant. But the same data show that the theoretical P(ID), which we reproduce exactly, understates reality by about 280-fold: 11 pairs of distinctly named cultivars share a complete 22-marker genotype where 0.04 collisions were expected, because clonal propagation, linkage and bottlenecked germplasm violate the independence the chain rule assumes. Probability of identity should therefore be measured on real germplasm rather than calculated. A community liftOver chain between cs10, Purple Kush, Pink Pepper and the MGC contigs remains the single highest-value next step for the field.