Claims & Evidence
How well-supported is a specific scientific claim — as distinct from how often it gets cited. Each claim is scored across eight evidence dimensions, and its status is derived by an inspectable rule, not a single truth score.
What this is, and isn't
A 9-claim pilot across six fields. Four claims were scored against outcomes the field had already settled — including two where the right answer is “this is solid,” so the rubric gets caught if it manufactures doubt. The other five are questions that remain genuinely unresolved. All passed their stated bar.
That is not five demonstrations that this finds unknown gaps. Only one unresolved claim was approached without substantial prior knowledge of the literature — a prior disclosed in writing before each pilot either way — so discovery of genuinely unknown gaps rests on a single case, and will until this process can read full texts rather than relying on well-documented debates. These pages report what the rubric found on these 9 claims, nothing about how often it would be right elsewhere.
Weak links found so far
6 of 9 tracked claims trip at least one named weak-link pattern — highly-cited-but-untested, conflicting evidence, an untested theoretical assumption, or an influential paper later repeatedly contradicted.
This describes what these four patterns found across the 9 tracked claims. It does notdescribe this system as demonstrated to discover unknown gaps in general. One unresolved-claim pilot passed without a prior answer key; that is evidence the method can work, not evidence about its hit rate. Only patterns A, C, D, and F are implemented — B, E, and G need dependency-graph machinery that doesn't exist yet (docs/BASIC_ROADMAP.md Phase 11).
Moderate alcohol consumption provides a net cardioprotective effect compared to abstention
Moderate increases in the legal minimum wage do not cause significant employment losses among low-wage workers
Accumulation of amyloid-beta is the primary causal driver of Alzheimer's disease pathogenesis
Fundamental particles are excitations of one-dimensional strings, requiring additional compactified spatial dimensions
Self-control failures occur because self-control draws on a limited, depletable resource
Implicit regularization by stochastic gradient descent is the primary explanation for why over-parameterized neural networks generalize
Open questions
6 questions across 6 claims, each motivated by a specific weak-link finding — not a ranked priority list.
Accumulation of amyloid-beta is the primary causal driver of Alzheimer's disease pathogenesis
Unknown — conflicting evidence- Is amyloid-beta accumulation the primary causal driver of Alzheimer's, or a downstream marker of the real cause -- given repeated Phase 3 trial failures directly targeting it?
Fundamental particles are excitations of one-dimensional strings, requiring additional compactified spatial dimensions
Speculative- What observable, even in principle, would distinguish string theory from its competing frameworks, given this claim has zero empirical rows of any kind?
Moderate alcohol consumption provides a net cardioprotective effect compared to abstention
Unknown — conflicting evidence- Does the apparent cardioprotective effect of moderate alcohol consumption survive Mendelian-randomization-quality causal designs, or is it explained by the "sick quitter" confound in the original observational studies?
Moderate increases in the legal minimum wage do not cause significant employment losses among low-wage workers
Unknown — conflicting evidence- Does the employment effect of a minimum-wage increase depend on local labor-market monopsony power, which would explain why methodologically credible studies reach opposite conclusions?
Implicit regularization by stochastic gradient descent is the primary explanation for why over-parameterized neural networks generalize
Unknown — conflicting evidence- Is SGD's implicit regularization the primary mechanism behind generalization in over-parameterized networks, or one of several redundant mechanisms (architecture, data augmentation, explicit regularization) that jointly suffice?
Self-control failures occur because self-control draws on a limited, depletable resource
Unknown — conflicting evidence- Does ego depletion exist as a real, smaller-than-originally-claimed effect, or was the original effect size substantially inflated by publication bias -- as the multilab preregistered replication failure suggests?
Self-control failures occur because self-control draws on a limited, depletable resource
Unknown — conflicting evidenceAdopting expansive postures causes hormonal changes (lower cortisol / higher testosterone) and increased risk tolerance
PreliminaryDistributed practice produces better long-term retention than massed practice
EstablishedModerate alcohol consumption provides a net cardioprotective effect compared to abstention
Unknown — conflicting evidenceModerate increases in the legal minimum wage do not cause significant employment losses among low-wage workers
Unknown — conflicting evidenceAccumulation of amyloid-beta is the primary causal driver of Alzheimer's disease pathogenesis
Unknown — conflicting evidenceHeavily over-parameterized neural networks generalize well to held-out data despite having enough capacity to fit random labels
EstablishedImplicit regularization by stochastic gradient descent is the primary explanation for why over-parameterized neural networks generalize
Unknown — conflicting evidenceFundamental particles are excitations of one-dimensional strings, requiring additional compactified spatial dimensions
Speculative