What the dataset covers
The dataset holds structured pomological records for 285 cultivars across 10 species. A record carries ploidy, how many pollen partners the cultivar needs, whether it can donate pollen at all, published parentage and sport relationships, and — where a crop has a genotype model — its S-alleles.
Coverage is uneven by design rather than by accident: the crops were commissioned in batches, and a crop with no data reports that it has none instead of falling back on a guess.
| Crop | Cultivars | Ripening windows | Incompatibility model |
|---|---|---|---|
| Almond | 5 | none | named-pairs |
| Apple | 115 | 111 | named-pairs |
| Apricot | 30 | none | sparse |
| Cherry | 30 | none | s-genotype |
| Hazelnut | 4 | none | named-pairs |
| Peach | 30 | none | named-pairs |
| Pear | 30 | 30 | named-pairs |
| Pecan | 6 | none | named-pairs |
| Plum | 30 | none | sparse |
| Walnut | 5 | none | named-pairs |
The rule engines
Ploidy and pollen sterility
Most fruit cultivars are diploid. A triploid — Bramley's Seedling, Jonagold, Blenheim Orange — carries three chromosome sets and produces effectively sterile pollen. It needs a partner, and it cannot repay one.
That is why a triploid requires two compatible diploid partners rather than one: the two diploids must pollinate the triploid and each other, because the triploid cannot pollinate either of them. A block of a triploid plus one diploid leaves the diploid without a partner, and it is the diploid that fails.
Pollen-donor status is tracked separately from ploidy, because ploidy alone does not catch every case. Male-sterile diploids exist — Magness, Waite and Ubileen among pears accept pollen freely but never donate it — and a model keyed only to ploidy would offer them as partners.
S-allele compatibility
Sweet cherry uses gametophytic self-incompatibility: each cultivar carries two S-alleles, and pollen carrying an allele the mother also carries is rejected in the style. Compatibility is therefore computed, not looked up — no shared allele gives a full cross, one shared allele a partial one, and two shared alleles a complete block.
This matters more than it sounds. For cherry a partner list is a derived output rather than stored data, so it stays correct as cultivars are added — a stored list silently goes wrong the moment the catalogue grows. The one exception is the S4′ mutation bred from the Stella line: cultivars carrying it are universal donors, accepted by every cultivar regardless of shared alleles.
Parentage, sports and named incompatible pairs
Apple and pear have no deployed genotype model to compute from, so incompatibility is asserted from published records instead, along three routes:
- Parent and offspring. A cultivar cannot be reliably pollinated by a direct parent. Jonagold is a Golden Delicious × Jonathan cross, so Golden Delicious is excluded as its partner on parentage — quite separately from Jonagold also being a triploid.
- Sports. A sport is a mutation of the same tree, genetically identical to its source and therefore useless as its partner. Tracked separately from parentage because the certainty differs: a sport shares 100% of its source's alleles, an offspring 50%.
- Named intersterile pairs. Crosses a source states outright do not work, held as explicit pairs with the reason attached.
Siblings are deliberately not excluded. Two cultivars sharing both parents inherit alleles probabilistically rather than identically, so blocking them would remove genuinely usable partners. Parent-to-offspring and sport-to-source are the only two relationships certain enough to block on.
Chill models
Winter chill is not measured on one scale, and the figures are not interchangeable. The schema records which model a figure was measured in, because a number carried across models is simply wrong:
- Hours below 7.2 °C — the simplest count, and the one most commonly quoted in nursery and extension literature. Treats every hour below the threshold as equal.
- Utah chill units — weights hours by temperature and subtracts for warm spells, which makes it better suited to climates with mild winter interruptions.
- Dynamic model chill portions — models chill accumulation as a two-step process that can be reversed before it is fixed. The most robust of the three in warm climates.
The schema supports all three. Coverage today is 17 cultivars of 285, and every one of those figures is recorded in a single model: hours-below-7.2c (17). Cold hardiness has a validated schema recording an absolute minimum survivable temperature with an empirical source, and no cultivar carries a record yet. Neither field is used to gate advice while coverage is this thin.
How the ripening figures were obtained
This section describes ripening data only. Ploidy, S-alleles, bloom groups and parentage carry no retrieval field, so nothing below should be read as a dataset-wide quality measure.
A required source string turned out to be a weak guard. It is trivially satisfied by a plausible-looking citation, so it demands attribution while having no way to demand verification — and commissioned research produced authoritative-looking sources without opening them, one report saying so outright. Every ripening window therefore records the route by which its figure arrived:
- transcribed0 records
- Read directly from the named primary source. The only value that means somebody opened the primary document.
- quoted33 records
- Read from a secondary source quoting or compiling primary research, without opening the original.
- converted108 records
- Transcribed from a different reference frame and converted onto ours — the source stated a real observed date, measured against a different anchor cultivar or region.
- reconstructed0 records
- Derived from established sequence knowledge rather than read from a document.
This is not a quality ranking. A reconstructed figure drawn from a strong sequence consensus can be better than a transcribed one taken from a nursery catalogue. It is a statement about the retrieval path, so a reviewer knows which rows a spot-check would actually change.
Read plainly, 141 of 285 cultivars carry a ripening window, and none of them was read direct from a primary document — the transcribed tier is empty. The figures are converted from other reference frames or quoted from compilations. We publish that rather than round it up.
How completely each crop is modelled
A separate axis from the one above, and a different question. Retrieval path asks how a figure was obtained; this asks whether the crop has a model at all. The distinction decides what a grower is told — a data gap invites them to look a cultivar up elsewhere, a model gap says the tool cannot answer for that crop yet.
- named-pairs
- No genotype model for the crop. Incompatibility comes from published parentage, sport relationships and explicitly named intersterile pairs.
- Almond, Apple, Hazelnut, Peach, Pear, Pecan, Walnut
- s-genotype
- Compatibility is computed from published S-genotypes. Partner lists are a derived output, so they stay correct as cultivars are added.
- Cherry
- sparse
- Partial published data. The tool can identify reliable donors within a pool but cannot rule every pairing in or out.
- Apricot, Plum
References
The dataset's backbone is the pomological literature and the national collections: the National Fruit Collection at Brogdale for accession identity and phenology, Bultitude (1983) and Morgan & Richards (2002) for cultivar description and parentage, the Brooks & Olmo Register of Fruit & Nut Varieties for release records, and peer-reviewed HortScience cultivar releases for disease resistance. Below is every named reference actually cited by a record, generated from the data rather than curated by hand.
- Agriculture Canada Harrow
- Agriculture Canada Kentville
- Aomori Station
- Australian Department of Agriculture
- Beckerman (2006) Purdue Extension BP-132-W
- Bell & van der Zwet (2003), USDA-ARS Kearneysville
- Brooks & Olmo, Register of Fruit & Nut Varieties
- Bultitude (1983)
- Cornell Fruit Resources
- Crosby et al. (1994) HortScience 29(7)
- East Malling Research
- Hedrick (1921)
- HortScience
- IVT Wageningen
- Iowa AES Bulletin 207
- Janick et al. (1988) HortScience 23(6)
- Janick et al. (2006) HortScience 41(2)
- Lamb et al. (1979) HortScience 14(6)
- Lamb et al. (1985) HortScience 20(4)
- Long Ashton Research Station
- Michigan AES
- Morgan & Richards (2002)
- Morioka / NZ DSIR joint release
- NJAES Rutgers
- NYSAES Geneva
- National Fruit Collection (Brogdale)
- Ohio AES
- PRI (Purdue–Rutgers–Illinois) release bulletin
- Peterson (1973)
- Pillnitz Dresden AES
- SDSU Fruit Bulletin
- Střížovice AES, Czech Republic
- Summerland AES (Agriculture Canada)
- Tohoku Station
- USDA Beltsville release records
- USDA-ARS / Ohio AES release bulletin
- USDA-ARS release bulletin
- University of Idaho
- University of Minnesota HRC
- Virginia Cooperative Extension
- WSU Extension
- WSU Mount Vernon trial evaluations
- WSU release bulletin
- Western Australia Department of Agriculture
- Williams et al. (1981) HortScience 16(6)
- Württemberg research station
A further 28 records cite breeder release, plant patent and plant variety rights records; nursery and trade catalogues; our own pomologist review; trial and extension records naming no retrievable document — real sources, but not named literature, so they are not listed above.