Method, limitations, and research agenda

Method and source-evaluation framework

The search was conducted through 15 July 2026 in English and Spanish. It combined exact-name variants with Peru, Cusco, La Convención, Inkawasi/Incahuasi, genetics, DNA, fingerprinting, accession, agronomy, sensory, processing, auction, price, and producer terms. Automated discovery used OpenAlex, followed by DOI and Crossref verification, PubMed Central, publisher full texts, institutional repositories, WCR, CATIE, Peruvian government pages, auction archives, and direct inspection of traceable trade and seller pages. Exact OpenAlex full-text searches returned no responsive “Inca Gesha” paper and one false-positive “SL9 coffee Peru” result, so absence from that index was treated as a search result—not proof that no study exists OpenAlexIncaGesha2026; OpenAlexSL9Peru2026.

Every candidate remains in data/source_index.csv, including rejected records and a reason for exclusion. The inclusion hierarchy was: peer-reviewed or official primary research and inspectable lab reports; institutional catalogs and government records; technically detailed industry reporting; and commercial or auction pages used only to document what was represented, described, priced, or sold. Search snippets served only for discovery. PDF claims required page locators; HTML claims required headings, paragraphs, table rows, or other reproducible locators plus the access date PruvotWoehl2020; Feran2026; SEY2026Saenz.

Claims were decomposed in data/evidence_ledger.csv and graded for directness, strength, and limitations. A variety claim required a sample-to-reference comparison; a historical route required documentary custody or archival transfer evidence; an agronomic claim required direct observation of authenticated material; a cultivar-specific sensory claim required a design able to separate genotype from environment and processing; and a market-causality claim required a comparison that controlled score, rank, producer, lot size, year, process, and bidder conditions. Commercial sources were never used as scientific proof PruvotWoehl2020; Marie2024; Donnet2008.

For genetics, the paper distinguishes rejection, exact match, cluster membership, and nearest-reference similarity. SSR fragment patterns in tetraploid Arabica are allelic phenotypes rather than whole-genome identities, and their interpretation depends on loci, references, database composition, distance metric, thresholds, and quality controls. The published eight-SSR WCR workflow and the later ten-SSR diversity study provide methodological comparators, but neither can be assumed to be the undisclosed protocol of the private Peruvian job PruvotWoehl2020, pp. 327–333; Montagnon2022, pp. 4–10.

For agronomy and sensory evidence, direct Peruvian studies were separated from regional context and from Panamanian or Nicaraguan comparator studies. The causal model was explicitly multi-stage: genotype interacts with site and season, crop management, cherry selection, fermentation, drying and storage, roast, brew, and assessor context. Controlled work was used to show that genotype can contribute; fermentation, roast, region, and panel studies were used to show why perceived resemblance cannot identify ancestry Marie2024; Ledezma2025; CastilloAviles2024; Koyner2025.

For market history, the unit of evidence was the dated page, lot, label, score, quantity, or price. Auction observations demonstrate realized value but do not estimate the causal effect of a name. Conflicting labels were retained rather than harmonized. Prices at farm gate, FOB, DDP, auction, and retail were not treated as directly comparable. The paper therefore reports transactions and disclosure practices while relying on hedonic and ethnographic research only for the more general proposition that value is assembled from multiple material and symbolic attributes Donnet2008; Fischer2021; Traore2018.

Limitations and unresolved evidence

Last updated: 2026-07-15

This project distinguishes evidence that is absent from evidence that contradicts a claim. Its main conclusions are intentionally asymmetric: the narrative distances the reported tested tree from RD2's undisclosed Geisha comparator more strongly than it establishes exact SL09. It weighs strongly against an unqualified T2722/Panamanian Geisha identification only if that comparator was authenticated T2722 and the submitted sample and custody were sound.

Genetic classification

• The complete RD2 Vision report discussed by Christopher Feran is not public. Public readers cannot inspect its job number and date, sample manifest, number of trees, collection and chain-of-custody records, loci, allele calls or electropherograms, controls and replicates, Geisha or SL09 reference accessions, database version, distance value, classification threshold, or signatures.
• The available account concerns a leaf submission from one reported tree. Even if collection and custody were impeccable, that result cannot classify every tree on the farm, every farm sharing a local name, every commercial lot called Inca Gesha, or a Peruvian population.
• The reported interpretation combines three statements: the fingerprint was previously unseen in the laboratory database, it was very close to an undisclosed SL09 reference, and it was unlike an undisclosed Geisha comparator. Without the underlying data, SL9-like or provisionally closest to SL09 is more defensible than an exact SL9 assignment.
• The public description calls the job an eleven-SSR analysis, but the actual marker panel and protocol could not be independently verified. Published eight- and ten-SSR studies are methodological comparators, not evidence of the private job's method.
• A chromosome-level Geisha genome and WCR's later 45-SNP database do not authenticate the reported Peruvian sample; they use different samples and are not publicly linked to the private result.

Reference material and historical provenance

• Jones's 1956 primary description says SL.9 was selected at Scott Agricultural Laboratories from a block of unknown origin. It does not document a controlled cross, named parent, accession link to T2722, precise Ethiopian collecting site, or transfer to Peru.
• The 1993 CATIE catalog was downloaded and inspected as a scan, but this project did not reliably locate a page proving the often-repeated mapping between T.02730 and SL09. That mapping remains unresolved here.
• A nearest-reference or cluster result is a classification relationship, not a migration history. No archival shipment, nursery, cooperative, monastery, or family-transfer record presently establishes a Kenya-to-Peru introduction route.
• Oral histories and retrospective trade accounts identify leads for further research, but selecting one route because it appears to fit the genetic hypothesis would be circular.

Agronomy and sensory identity

• No peer-reviewed direct agronomic study of genetically authenticated Peruvian Inca Gesha/SL09 material was located. Regional Cusco/Inkawasi evidence concerns farmer-reported Typica, Catimor, or coffee generally and is contextual rather than cultivar-specific.
• No controlled sensory or chemical study of authenticated Peruvian Inca Gesha/SL09 material was located. Commercial tasting notes and competition scores describe lots but cannot identify ancestry or isolate a varietal effect.
• Comparator studies show that genotype can matter and that environment, fermentation, drying, storage, roast, brewing, and panel design can all modify perceived flavor. They support the possibility of sensory convergence, not a prediction of the disputed material's inherent cup profile.
• Morphology, disease response, yield, and elevation performance reported historically for Kenyan SL.9 cannot be transferred automatically to a modern Peruvian near-match without common-garden and multi-environment trials.

Market and chronology

• Auction and seller records establish dated labels, submitted scores, quantities, prices, and descriptions where explicitly recorded. They do not independently validate genotype, and they cannot isolate a causal premium produced by the word Gesha.
• Prices from auctions, retail, farm gate, FOB, and delivered transactions are not directly comparable. Observed value also reflects producer reputation, cup score, rank, scarcity, lot size, processing, harvest, and bidder conditions.
• Several market pages are mutable and may have been edited after publication. Historical snapshots, catalogs, invoices, and original test correspondence could alter the earliest defensible dates in the naming chronology.
• Feran's page displays a publication date of 2 January 2026. Its top update block labels three items 7 January, 9 March, and 1 July 2025, while corresponding lower headings label the same dates in 2026. This internal inconsistency is recorded rather than silently harmonized.

Search and reproducibility limits

• The deduplicated source inventory records 166 considered sources, including 113 exclusions with reasons, but a null search is not proof of nonexistence. Unindexed Spanish-language theses, offline records, unpublished laboratory work, private cooperative documents, and future publications may change the result.
• Some relevant publisher files were unavailable through their PDF endpoints. The project used lawful open HTML, repositories, institutional copies, or recorded the access failure; it did not bypass paywalls, CAPTCHAs, authentication, robots restrictions, or other technical controls.
• Exact quotations and page locators were retained in 69 evidence-ledger rows, but several historical scans required OCR. status=verified means the source, locator, and excerpt were checked; it does not mean the underlying claim was independently proven. Critical OCR-derived passages were checked against page images; OCR remains less reliable than born-digital text.
• Bibliographic URLs and DOIs were checked during the research session, but websites and redirect targets can change after the stated access date.

Evidence needed to resolve the question

The strongest next study would publish a redacted laboratory report and manifest; sample multiple tagged trees per farm and multiple farms per local name; use blinded duplicates and authenticated T2722, historical SL09, neighboring Ethiopian Legacy, and unrelated controls; disclose markers or sequence data, distances, thresholds, and quality controls; and then place verified plant groups in replicated multi-environment agronomic trials. Factorial processing, roast, chemical, trained-panel, and consumer studies would be required separately for sensory attribution. A controlled auction, conjoint experiment, or credible archival natural experiment would be needed to estimate a name premium.