PRESS RELEASE
TORONTO, ON, September 2026 — Most Sciences Maintain One Classification of Their Objects. A New Monograph Argues That This Is an Accident of Cost, and Supplies the Measurement That Replaces It.
Anomaly hunting by untargeted cross-partition analysis: Cross-examination of orthogonal classifications. Toronto: Altaspera Publishing. ISBN 979–8–1751–0861–4. https://doi.org/10.13140/RG.2.2.15448.87045
A new monograph from the Institute of Integrative and Interdisciplinary Research proves that a field maintaining a single classification cannot detect its own blind spot — the deficiency is formally indistinguishable from its absence — and develops a method, untargeted cross-partition analysis, that maintains several informationally independent classifications of one corpus and reads their intersections. The method is run, with all code printed, on six public datasets. It recovers the sub-Jovian exoplanet desert at forty bits, isolates the white-dwarf sequence from a stellar table at 2,594 bits, finds the ν₆ resonance boundary in the asteroid belt as a diagonal exclusion invisible in either axis alone, and, on the periodic system, refuses to run and states why.
Monograph: Kriger, B. (2026). Anomaly Hunting by Untargeted Cross-Partition Analysis: Cross-Examination of Orthogonal Classifications. Toronto: Altaspera Publishing. ISBN 979-8-1751-0861-4. DOI: 10.13140/RG.2.2.15448.87045
Chemistry has one periodic system. Biology has one Linnaean hierarchy. Medicine has one International Classification of Diseases. In each case the scheme is treated, in working practice, as a description of how the domain is divided rather than as one projection among many that could have been built.
The monograph argues that the reason is economic rather than epistemic. Maintaining a classification requires that every object be examined, assigned, recorded and taught. Under a fixed budget of expert attention, the number of classifications a community can maintain has historically been one — and a second was never affordable, so the first became unquestionable. Automated classification has reduced the marginal cost of assignment by orders of magnitude, and the binding constraint has moved from labor to data.
Four formal results.
1. A single-axis regime cannot detect its own blind spot. The information about a target recoverable by adding an orthogonal classification is a conditional mutual information. The monograph proves that two models — one in which nothing is missing and one in which an entire orthogonal axis would explain everything — produce identical joint distributions of the observable labels. No statistic computed within the existing scheme can estimate, bound or test the loss. Every standard diagnostic (purity, predictive accuracy, expert agreement) is uninformative as a class.
2. Agreement between independent channels is measurable in bits. Whewell’s consilience is converted into an additive log-likelihood ratio: each independently built classification that places a boundary in the same place contributes evidence, and the total grows linearly in the number of channels. The accompanying correction shows that correlated channels saturate: at a correlation of 0.2, no number of additional agreeing studies is worth more than five independent ones.
3. A domain can support only so many classifications. The sampling ceiling ∑ H∞(Pᵢ) ≤ log₂(n/λ_min) bounds the number of crossed axes a corpus can populate. For 5,659 mineral species it permits three; for roughly 2.1 million described biological species, six. The historically binding constraint was labor, which bound at one.
4. Only the non-separable part of a selection function matters. A detection-efficiency model that factorizes across the axes is absorbed entirely into the fitted margins and changes nothing. A threshold-shaped completeness does change the result — and, in the exoplanet case, makes the anomaly more surprising, not less.
Six computed cases, with the code printed.
The apparatus is run mechanically on public data. On 4,239 transiting exoplanets from the NASA Exoplanet Archive crossed by radius and period, the audit recovers the sub-Jovian desert: three planets observed against 36.8 expected, forty bits against an independence null, rising to 57 bits once transit detectability is modelled. On 24,873 Gaia stars in a 30–50 pc shell, the screening rule rejects colour against absolute magnitude as two coordinates of one relation — and then, with the main sequence absorbed into a fitted null, returns the white dwarf sequence as a single cell at 2,594 bits against a chance maximum of six. On 75,893 asteroids, it locates the ν₆ secular resonance as a diagonal boundary in the plane of semi-major axis and inclination, while demonstrating that the Kirkwood gaps cannot be found by this method at all, because a depletion living in a margin is absorbed before any cell is tested. On 685 Swift gamma-ray bursts it finds nothing at five classes per axis and recovers the known short-hard/long-soft bimodality at 22 bits at three classes — an empirical face for the result that optimal granularity is near three classes per axis. On 1,172 languages, an ignorant null returns twenty-three cells of restated common knowledge and an informed one reduces the queue to a single surviving areal effect. On the periodic system, the ceiling check fires before any table is computed and the run halts.
“A field with one classification is not merely limited — it is self-certifying,” said Boris Kriger, Lead Investigator. “It reports no deficiency precisely when its deficiency is largest, and that is a theorem rather than a complaint. What has changed is the price. Assigning an object to a category used to require a trained person and a morning. A domain that wanted six crossed classifications could not have them in 1960 and can have them now.”
The title names the standard objection.
Anomaly hunting is a term of art in the skeptical literature for a fallacy: trawling data for anything odd and treating the oddity as support for a conclusion held in advance. The monograph adopts the phrase deliberately and answers it procedurally. What counts as anomalous is fixed by a declared null and a threshold recorded, with a hash, before the table is computed; examining m cells costs log₂ m bits, subtracted before anything is reported; a bootstrap measures what the procedure finds in data resampled from the null; and a mechanism is admitted only if it entails consequences outside the cell that prompted it.
Falsifiability.
The monograph states the conditions under which its program should be abandoned: if channel independence cannot be estimated in a domain, the concordance result degrades to a qualitative claim; if crossed audits of well-studied corpora return no anomalies that the single axis had missed, the method is an expensive way of producing tables; and if automated axis generation yields only paraphrases of the dominant scheme, the screening step will pass nothing and the program stalls.
Limitations.
Orthogonality is relative to the corpus and the measure, and catalogues are themselves products of the dominant scheme. Estimation on sparse tables is unreliable near the ceiling. The formalism covers partitions but not overlapping categories, which excludes comorbidity and therefore much of medicine. The framework specifies how to test a candidate axis and not where candidate axes come from; generation remains a creative act, and no unsupervised method can guarantee independence.
An invitation.
The Institute invites taxonomists, astronomers, mineralogists, linguists, clinicians and statisticians to apply the procedure in their own domains, particularly where a canonical classification has stood unchallenged for a generation. The reference implementation, the audited datasets and the full code are printed in the monograph’s appendices and are reproducible from public archives. Researchers interested in running a crossed audit on a corpus, or in estimating channel correlation from documented revision histories, are invited to contact the Institute directly at boriskriger@interdisciplinary-institute.org
Monograph: Anomaly Hunting by Untargeted Cross-Partition Analysis: Cross-Examination of Orthogonal Classifications. ISBN 979-8-1751-0861-4 · DOI: 10.13140/RG.2.2.15448.87045
Parent work: Orders of Orthogonality: Frame Closure and the Displacement That Escapes It. DOI: 10.5281/zenodo.20100022
Data sources used: NASA Exoplanet Archive · Gaia DR3 via VizieR · JPL Small-Body Database · HEASARC (Swift) · World Atlas of Language Structures
Boris Kriger | Lead Investigator
ORCID: orcid.org/0009-0001-0034-2903
Institute of Integrative and Interdisciplinary Research, Toronto
+1 437-552-8807 · boriskriger@interdisciplinary-institute.org
About the Institute
The Institute of Integrative and Interdisciplinary Research (IIIR) is a Toronto-based organization dedicated to solving complex problems through formal precision and cross-domain synthesis. Treating interdisciplinarity as a methodological necessity, the Institute bridges specialized fields to develop coherent theoretical architectures.
Institute of Integrative and Interdisciplinary Research 
Leave a comment