Content
Citation(s) & Footnote(s)
Book(s)Status: Article analysis retained; all ten separately published data workbooks now read and integrated. Publication use remains qualified by the limitations below. The prior page-by-page review covered No access.
Analytical note
Core summary
A controlled Texas study finds that corn variety, growing environment, and their interaction affect new-make bourbon yield and trained-panel sensory measurements.
Author argument
Corn is not a neutral starch carrier; genotype and agronomy can leave measurable chemical and sensory differences before barrel aging.
Evidence assessment
Multi-site crop trial, controlled processing, trained-panel lexicon, HPLC, GC-MS/O, and statistical modeling. Primary research with disclosed distillery employment.
Researcher synthesis
A rigorous account of terroir must separate raw-material effects measured in new make from romantic claims about finished-whiskey regional character, then ask which differences survive maturation and blending.
Limitations
Thirty experimental samples; three commodity hybrids; new make only; participating distillery employed two authors.
Open questions
Which detected differences persist after equivalent barrel aging? How reproducible are effects across years, distilleries, yeast strains, and larger agricultural samples?
Sources
- Source: No access
- Citation:
Corn Variety and Texas Terroir — Chicago
Supplementary data assessment — September 8, 2026
The data contain ten location-by-variety treatments with three batches each. Hansford contains Terral only. Intermediate fermentation measurements are irregular and incomplete by design; baseline and final measurements cover all 30 batches.
Do not count analytical replicates or repeated timepoints as independent farm treatments. Normalize variant spellings only in a separate join; retain source values.
The supplied chemical matrices label their values Area. Milled corn contains 52 compound columns and new make 68, each with 30 observations. Numerous zero entries and duplicate-like names remain in the originals.
Peak area is not a calibrated concentration, odor-activity value or safety assessment. Do not merge differently labeled compounds or interpret zeros as universal chemical absence without method support.
The workbook distinguishes post-mashing and post-fermentation data. Four post-mash rows lack carbohydrate measurements; some totals contain a space rather than a number. Ethanol is explicitly labeled percent by weight.
Do not replace blanks with zero or convert ABW directly to ABV. Reconcile ppm carbohydrate headings against methods and calibration before using absolute values. Preserve missing rows16,17,23,29.
Thirty yield formulas divide recovered ethanol milliliters by 448 grams and agree with their cached values. The separate spirit-proof observations range from 117.732 to 128.238.
Arithmetic agreement does not validate the underlying measurement or reproduce the statistical models. These experimental proofs and yields are not a commercial blending target or universal distillation specification.
The sensory matrices contain batch-level consensus descriptors, not individual consumer ratings. S9 totals sum the selected aroma columns and partition them using S10 good/bad categories; all 90 stored sum formulas agree with independent arithmetic.
Blended, Alcohol and trigeminal attributes are outside the total. Medicinal is classified good while woody and buttery are bad; these are study-specific new-make assumptions, not rules for the user's blend. The paper's favorable soapy wording conflicts with S10.
The direct corn-benzaldehyde/favorable-aroma relationship is nonsignificant across all samples (R=0.2837); excluding one observation gives R=0.362, p=0.0536, n=29. This meets the authors' exploratory 10% threshold, not 5%.
Retain both reported results. The data support hypothesis development, not proof that adding a benzaldehyde-rich ingredient improves a finished commercial blend. Maturation and consumer preference were not tested.
All ten original workbooks recovered and preserved together on the Source. Every populated cell was indexed; whole-column numeric/missing-value checks and all 120 formula checks were performed. This does not reproduce the REML models or validate causal claims. Individual-row review of all ten worksheets is complete. Twelve added evidence/citation bodies and their Source/Literature Note links have been checked. Match datasets by location, variety, and batch: S2 Hill/DynaGro rows 24–26 are batches 2, 3, 1 and Hill/Terral rows 30–32 are 3, 1, 2; S8/S9 are sorted 1, 2, 3. A row-position join would mispair measurements. Page-layout settings remain unverified.
Independent full-source audit — September 27, 2026
All 30 article pages, 66 reference entries, Figures 1–4 and Tables 1–14 were read; all relevant figure/table panels were inspected in page renders. All populated rows and cells in the ten separately published S1–S10 workbooks were read, including formulas and category definitions. Fresh publisher copies were retained unchanged locally; the existing PDF and ten-workbook ZIP already preserved on this Source were retained without duplicate uploads. Current downloads have recorded hashes; byte identity with the earlier Notion ZIP and the cloud placeholder has not been re-established. This is a complete examination of the article and its ten supplied supplements, not a replication of its statistical models or a review of all 66 cited works.
What the experiment actually measured
One 2016 crop year, three commercial hybrids across three Texas farms plus Terral at Hansford produced ten location–variety treatments and 30 processing batches. These are not 30 independent farms. Location bundles soil, weather, irrigation, planting date, population and rotation; field randomization/replication is insufficiently described to isolate geography from management. Treating factors as random does not establish broad representativeness.
The laboratory recipe used 448 g corn, 1,750 g water, added enzymes and proprietary yeast, followed by roughly five days of fermentation, freezing/thawing, and two distillations. It was 100% corn experimental unaged spirit. Grain combinations, cask maturation and consumer liking were not tested. The 550 mL first-run product was low wines, despite Figure 2's new-make label. The spirit run charged 500 mL diluted low wines, discarded 25 mL heads and collected 100 mL hearts; the target dilution ABV is not specified. S6 yield divides ethanol in the 550 mL low-wines collection by original 448 g grain. Only 1.65 L beer was distilled, and sampling removed material: this is not automatically a full-mash commercial recovery or hearts yield.
Seven trained panelists supplied batch consensus aroma ratings, not seven independent consumer observations. New make was assessed by nose at 20% ABV, with 8 mL in covered tulip glasses. The paper's broad flavor/quality language exceeds an aroma-only assessment. Training, reference standards, coding and randomized presentation are strengths; consensus data cannot recover individual panelist repeatability. GC-MS/O used two operators, SPME and selection thresholds; no odor identities were recorded at sniff events. Library matches/retention checks and peak areas do not supply calibrated concentrations, odor-activity values or a mass balance from corn to spirit.
Results and inferential limits
The reported location and genotype differences in yield and some aroma attributes warrant further controlled study. They do not establish a universally superior hybrid, a regional signature that survives aging, or a consumer preference advantage. Farm selection and distillery involvement were disclosed; two authors' employment is relevant context, not grounds to dismiss measurements.
Three final HPLC ethanol observations were excluded because they disagreed with yield measurements. Report both fits: the ethanol residual variance share changes from about 89% to 41.4%, while farm and variety shares rise. Disagreement alone does not prove measurement error. Four missing baseline carbohydrate observations remain missing; S5 labels ethanol as percent ABW and carbohydrate values as ppm, whereas the article's units differ. Do not substitute zero or silently equate ABW with ABV.
Figure 4's direct corn-benzaldehyde/favorable-aroma association is R=0.2837 and nonsignificant with all samples; excluding Perryton/Terral batch 1 gives R=0.362, p=0.0536, n=29. This meets the authors' exploratory 10% threshold, not conventional 5%. Other plotted relationships are R=0.427, p=0.0186 and R=0.5042, p=0.0045, n=30. Multiple screened correlations, post-hoc aroma grouping and lack of held-out validation prevent treating benzaldehyde as a validated procurement marker. A correlation is not an experiment in adding an ingredient.
Numerical and visual audit
- Table 3 labels conflict: values near 7.86 protein and 68.02 starch are percentage-scale NIR predictions, consistent with S1 percent dry-basis headings, not the printed mg/g-style unit. Whole-grain versus ground calibrations give different results and are not independent field replicates.
- Table 7's initial SG rounded to 1.1 obscures the measured values near 1.064; SG is not a direct yeast-growth measurement. S4 contains 123 observations: days 0–5 have 30, 17, 4, 15, 27 and 30 rows. Final times range approximately 118.9–124.8 hours; nominal days are not identical elapsed times. Repeated-measure covariance treatment is not sufficiently described.
- Tables 10/11 report n=60/56 despite 30 processing batches and 26 complete baseline carbohydrate records. The observation basis requires clarification. The prose's all-correlations-significant claim conflicts with starch/ethanol p=0.1751 and 0.1926 and starch/yield p=0.1204.
- Table 13's Alcohol variance percentages total 120%; other rows in Tables 8/13 do not close to 100%. Do not copy these as reliable decompositions without correction.
- Table 14 contains correlation coefficients despite its probability wording. Its stars mean 10%, 5%, 1%; explain that convention.
- Independent supplement arithmetic reproduced 30 S6 division results and all 90 S9 sensory sums. This checks arithmetic, not model validity. S2 ethyl-decanoate mean is 60,421.4 and S8 mean 48,826,484.83 area units; these support the prose's approximate amounts but differ from Tables 4/12 (62,152.3 and 51,696,255). Do not silently choose a denominator or impute a reason.
- S2 Hill/DynaGro batches are ordered 2,3,1 and Hill/Terral 3,1,2; S8/S9 use 1,2,3. Join on location, variety and batch, never row position. Retain original spellings, zeros, blanks and compound labels. Furfural/Furfural 2 and related label variants are not automatically duplicate peaks.
Sensory lexicon audit and Academy use
Tables 1–2 provide useful examples of operational definitions and anchored training. They are not a ready-to-copy learner kit: Anise repeats the alcohol definition; Buttery uses the coconut reference; 190 proof is 95% ABV, not the listed 90%; two purported 60% nose-warming references have inconsistent recipes. Vinegar dilution labels do not establish pure acetic-acid percentage without stock strength. Laboratory chemical references require a reviewed laboratory protocol rather than direct learner DIY transfer.
S9 totals 48 selected descriptors, excluding Blended, Alcohol and four trigeminal attributes. Its 28 good/20 bad grouping is a chosen new-make quality definition: medicinal is good, woody/buttery bad, and soapy is bad in S10 despite favorable prose. Summing these ratings does not create a universal quality or consumer liking scale.
Proposed Academy applications: a farm-to-glass lesson distinguishing genotype, growing conditions, processing and maturation; a sensory exercise separating descriptive intensity from preference; a methods exercise showing how exclusion changes a conclusion; and a data exercise joining batches correctly and distinguishing peak area from concentration. Any procurement trial should repeat across harvests, document field replication, standardize downstream processing, include blinded sensory replication, predeclare exclusions and test maturation/consumer outcomes before claiming commercial superiority. These are proposals, not changes to live courses.
Connections and open questions
Connect the existing terroir idea
