Skip to main content

Open Theories: What We Think Is Going On, and What Would Prove Us Wrong

A theory here is an explanation we have proposed for something in our own data, written so it can be shot down. Each entry carries the observation with its numbers, the explanation, the competing explanations we could not rule out, the test that would falsify it, and who proposed it. Nothing on this page is a position. A theory that survives testing can become one; a theory that fails stays here with its date.

This page sits between Open Problems, which says what we don't know, and Research Asks, which lists the experiments that would settle it. The chain runs problem, theory, ask, data. A theory is never proven. It's tested, and it survives or it doesn't. Status runs proposed (stated, not yet tested outside the data that suggested it), testing (an ask is open or a trial is running), supported (held up against independent data), or refuted (failed, with the evidence that failed it). A refuted theory is not deleted, because the record of what didn't hold is worth as much as what did.

Who proposes. Some entries began as pattern-finding by the AI tools we use to work our own datasets; each of those is labeled AI-proposed and was recomputed by a person from the published file before it was entered. An AI can propose a theory. It can't approve one, and nothing here becomes published guidance without a named person deciding it should (how we work).

Malting​

The fourth germination day buys alpha-amylase, not diastatic power

The observation. In the USDA Madison screening, 25 named sorghum cultivars were malted twice under one protocol, at three and four days of germination. Between day three and day four, alpha-amylase rose in all 25 cultivars, by 19.2 DU on average and by anywhere from 3.3 to 41.8 DU. Diastatic power did not move: down 0.3 °ASBC on average, up in 14 cultivars and down in 11. Extract rose in 18 and fell in 7, FAN rose in 20 of 25. The arithmetic is on the published CSV, all 50 rows. The theory. In sorghum malt, the diastatic power assay is carried mostly by beta-amylase, which is low to begin with and, on this evidence, not built by another day of germination. Alpha-amylase is. So the fourth day is an alpha-amylase decision, not an "enzyme power" decision, and it's cultivar-specific: the gain ran across a twelve-fold range. A germination endpoint chosen for a cultivar, rather than a single schedule for all sorghum, is the practical reading. A second reading of the same data: the modification gain from the extra day (soluble-to-total protein) tracked grain protein (r = +0.65, n = 25) and survived controlling for the day-three starting point, so high-protein grain got modification from day four and low-protein grain got only enzyme. What we could not rule out. One location, one crop year, one micro-malting protocol, so genotype and environment are confounded. The modification index has protein in its denominator, so part of the protein link may be structural. And our own doctrine reads this screening differently on purpose: we malt for character and let external enzymes own conversion, so a schedule rule tuned to enzyme yield matters less to this site than to an enzyme-first maltster. What would falsify it. Malt two high-gain and two low-gain cultivars from this list at three, four, and five days under one protocol, at a second location or crop year, measuring alpha-amylase and diastatic power by the same methods. Falsified if diastatic power rises with alpha-amylase, or if the alpha-amylase gain fails to replicate. The cheapest first pass needs no malting: the 2002 to 2004 screenings carry grain protein and modification, and if that record holds a germination-length variable the protein link can be checked against independent grain. Why it matters. Anyone choosing a germination schedule is choosing how much alpha-amylase the malt will carry, and by how much depends on the cultivar. It also sharpens how the reliable-enzymes problem should be read: the malt does build alpha-amylase, and the schedule controls how much; what it does not build is beta-amylase. Research asks. Ask 9 is the test, extended to two germination lengths. Status. Proposed. Origin. AI-proposed, discovery run of 2026-08-26; recomputed from the published CSV 2026-09-06. Confidence. Medium: the arithmetic is exact, the dataset is one. Entered. 2026-09-06.

Modification, not protein, is the better predictor of sorghum wort clarity

The observation. Across the 74 micro-malted samples of the 2002 to 2004 screenings, higher total protein went with hazier wort (r = +0.33 against turbidity, +0.40 against the clarity score) and higher modification went with clearer wort (soluble-to-total protein against the clarity score, r = −0.32); soluble protein on its own predicted nothing (r = +0.07). The Madison screening shows the same signs, weaker, at three days (+0.29 and −0.28) and almost nothing at four days, where 18 of 25 worts score clear. Of the three 2019 production malts, the hazy, slow-filtering one was not the highest in protein but the lowest in modification. Recomputed from the published CSVs 2026-09-07. The theory. What clouds a sorghum wort is protein that germination did not break down, so the modification index carries the signal and total protein is only its proxy: high-protein grain modifies less in a fixed schedule (Madison 3-day, protein against modification r = −0.73), which is why protein correlates with haze at all. If that is right, a well-modified high-protein lot should run clear and a poorly modified low-protein lot hazy, which is what the 2019 malts did. What we could not rule out. The clarity measures are a 1-to-3 score and a turbidity column of unrecorded unit, from one program's micro-malting; the modification index has protein in its denominator, so some of the negative correlation is arithmetic; and no production dataset we hold carries protein, modification and turbidity together. Kafirin cross-linking under mash heat, which the kafirin page describes, could produce the same pattern by a different route. What would falsify it. Ten or more production lots with total protein, soluble-to-total ratio and wort turbidity measured on the same lot under one mash. Falsified if turbidity tracks protein after modification is held constant, or if it tracks neither. Why it matters. The protein page says protein is a weak haze signal and gives no threshold. If modification is the real variable, a malt certificate's S/T line is the number to read for clarity, and a protein specification alone would sort lots on the wrong axis. Research asks. Ask 5, whose reference mash and turbidity method would produce exactly the ten-lot dataset. Status. Proposed. Origin. AI-proposed from the 2026-09-07 protein-direction analysis (pulled forward from a post-conference schedule by Craig). Confidence. The correlations are exact; the mechanism is one reading of them, medium-low. Entered. 2026-09-07.

Consumers and labeling​

Confusion at the shelf, not expectation, drove the reaction gap in the 2015 survey

The observation. In the 2015 CSA and Bard's survey, 711 respondents had tried a gluten-reduced beer and answered whether they knew at purchase that it contained wheat or barley. Of those who knew, 129 of 399 reported a reaction (32.3%). Of those who did not know, 149 of 312 did (47.8%), a ratio of 1.48. The gap holds inside the diagnosed-celiac subset (33.1% against 48.7%), inside the respondents who liked the beer (29.5% against 46.7%), and for strong reactions alone (13.8% against 27.2%). Recomputed from the published response file. The theory. The usual objection to a reaction rate from a survey is expectation: people who know there's barley in the beer expect to react, and report that they did. This data runs the other way. The people who reacted more were the ones who did not know. Our explanation is that a buyer who knew the grain could weigh the risk, take a smaller serving, one beer, or none, while a buyer who believed they had bought a gluten-free beer drank it like one. Labeling that failed at the shelf exposed the least-informed consumers most. What we could not rule out. Attribution after the fact: a person who reacted goes looking for a cause, learns about the grain afterward, and honestly reports having been unaware at purchase. That channel produces the same gap without any difference in what was drunk. The survey is self-selected and retrospective, awareness was asked once rather than per beer, and respondents could tick more than one brand, so what the two groups drank cannot be matched. What would falsify it. A prospective design: record awareness before consumption and outcome after, on the same products, with serving size. Falsified if the reaction rates converge once awareness is fixed before drinking, or if the unaware group turns out to have drunk different products with higher gluten, or if reported serving sizes do not differ by awareness. Why it matters. Gluten-reduced beer is not gluten-free rests partly on this survey, and the expectation objection is the standard way to wave the reaction rate away. If the mechanism is confusion at the shelf, the fix is the label, which is what the qualifying statement the regulator now requires is for. Research asks. None yet. The test above is a survey design, not a brewing experiment, and belongs with whoever runs the next consumer study. Status. Proposed. Origin. AI-proposed, discovery run of 2026-08-26; recomputed from the published file 2026-09-06. Confidence. The observation is exact; the causal reading is medium-low until awareness is measured before drinking. Entered. 2026-09-06.

What is not here yet​

One further candidate from the same runs is held until its dates are verified against the primary record. A theory enters this page after a person has recomputed its observation from the published data; a candidate that can't yet be checked stays in the workbench.

How this page changes​

A theory is added when a recomputed observation on our own data supports an explanation worth testing. It moves when an ask opens, when independent data arrives, or when someone outside this site sends the evidence that settles it. Each change is dated below. Refuted theories stay, marked, with what refuted them.

Changes.

  • 2026-09-07 — Third theory entered: modification, not protein, as the predictor of sorghum wort clarity, from the protein-direction analysis that closed SP-0013.
  • 2026-09-06 — Page created with two theories, both from the 2026-08-26 discovery run and recomputed the day of entry (DR-0079).