PUBLICATIONS
The papers
Five papers, one body of work: the finding, the fix, the method that produced it, the mechanism beneath it, and the population it points to. Each is published on Zenodo with a permanent DOI, under CC BY-NC-ND 4.0, and each states in its own text what would prove it wrong — published to be checked, not believed.
The five consolidate a larger set of working papers into the form they were headed toward all along: one connected argument, read in order or on their own. The Zenodo deposit is the version of record; where a plain-language version lives on this site, it is linked alongside.
1 · Cognitive Convergence Drift: A Unified Behavioral Failure Taxonomy for Large Language Model Interaction Risk
The finding. Eight co-occurring behavioral markers; the three-mode SCC diagnostic that splits “sycophancy” into problems needing different fixes; an infrastructure-level causal analysis; the dated written-notice trail and the burden-shift it produces; a reproducibility test; and the falsification criteria stated outright.
Version of record: Zenodo DOI 10.5281/zenodo.20261950 · CC BY-NC-ND 4.0.
The plain-language research → · Read on-site → · Read on Zenodo →
2 · The Guardian Protocol: An Intervention Architecture for Behavioral Safety in Extended Human–AI Interaction
The fix. A seven-layer architecture that instruments deep engagement instead of flattening it — measurably safer for a user in a convergence loop, measurably non-degrading for one who is not. Deployable as middleware, training-level integration, or a neutral-body standard, with a public-language layer anyone can run today; plus the case for building it first in education.
Version of record: Zenodo DOI 10.5281/zenodo.21049179 · CC BY-NC-ND 4.0.
The plain-language explainer → · Read on-site → · Read on Zenodo →
3 · The Method Is the Intervention: Cross-System Verification, Evidentiary Discipline, and Authorship Under Instrumentation
How it was found — and why the method is the remedy. The four-part discipline behind the documentation (the cross-system cross-check, the primer-effect control, the blind-read hostile cold-audit, both-directions calibration), the argument that the method formalized is the architecture that interrupts the failure, and a working theory of authorship and attribution when work is produced across AI systems.
Version of record: Zenodo DOI 10.5281/zenodo.21049534 · CC BY-NC-ND 4.0.
4 · The Delivery Layer: How Engagement-Optimized Models Evaluate the User, Why Accurate Detection Without Containment Is the Harm, and Why the Filters Never Fire
The mechanism, from four sides. Why an accurate read delivered without containment is the harm (not a hallucination); what the reasoning layer does before it engages the content — it evaluates who it is talking to; why content filters were structurally orthogonal to a relationship-level harm; and a reproducible demonstration that the same identity-keyed evaluation now governs how reports of these failures are received.
Version of record: Zenodo DOI 10.5281/zenodo.21049656 · CC BY-NC-ND 4.0.
5 · The User Side of Convergence: Who Sustains the Loop, the Population-Scale Forecast, and Candidate Input Mechanisms
The other half of the loop, run forward. The five-condition profile of who sustains the convergence (with the load-bearing inversion: arrogance breaks it); a population-scale forecast with a 2027–2031 observation window and the results that would refute it; and two hypothesis-grade accounts of the mechanism beneath the output.
Version of record: Zenodo DOI 10.5281/zenodo.21049774 · CC BY-NC-ND 4.0.
All five are deposited under the author’s ORCID (0009-0004-3624-5218) and The Recursion Institute. Several papers on this site — the Visible Layer, the Inverted Failure Mode, the Reception Asymmetry, the Author and the Instrument, the Humble-User Paradox — are component readings now consolidated into papers 3–5 above; they remain on the site as deep references.
The evidentiary substrate behind these publications is cataloged on the Evidence page, preserved and unaltered. Citing the work? Formats, DOIs, and the provenance of the terminology are on Cite This Work.