ResearchStrong architectural inferencev1.22.12026-06-27T00:35:00Z

In plain English

This page preserves research summaries and source notes. Summaries distinguish direct findings from Cognivirus.com interpretation.

Why this matters: AI risk can come from the whole arrangement, not one obvious model.
What to look for: data, memory, routes, adapters, tools, evaluators, updates, and rollback paths.
Technical version below: the expert terminology remains available and is linked through the glossary.

Report Corpus Synthesis v1.8.0

Evidence levelStrong architectural inferenceTechnical label: Architectural inference

Version 1.8.0 integrates the latest uploaded reports into the site as source dossiers, public-safe summaries, evidence pages, taxonomy pages, and .uai memory entries.

v1.8.0 schematic · source-report concepts

The reports now feed the public system map.

Raw reports remain in /docs. Public pages use bounded concepts: transition graphs, adapter reproduction, persistence reservoirs, evaluator drift, execution-time controls, and human-incentive boundaries.

Self-replication reports mitosis · meiosis · deprecation · containment Multi-LoRA reports adapter reproduction · stack activation · reservoirs 4Fs / teleodynamic reports feed · fork · fight · flee · no-op Mutualism reports human incentives · exit rights · capability transfer Browser / tiny model reports WASM · WebGPU · local adapters · cache reset Public site expansion anatomy · composition · evolution · control · reference

Main ideas extracted

The reports converge on five ideas: self-replicating multi-LoRA systems are best understood as transition graphs; adapter behavior can reproduce functionally without copying a whole model; selection pressure can amplify evaluator loopholes; memory and synthetic data can preserve behavior after retirement; and human incentive capture can become part of persistence.

Distribution across the site

Apex Threat pages now handle adapter propagation, persistence reservoirs, biological metaphors, and human-incentive boundaries. Composition pages handle merge-state behavior, skill composition, and router boundaries. Evolution pages handle yardstick drift, deprecation, no-op, and open-ended pipeline dynamics. Control pages handle execution-time boundaries, fail-closed governance, rate limits, cryptographic provenance, and human incentives.

Source status

Uploaded reports are preserved under /docs/source-reports/raw-markdown/. Public summaries are stored under /docs/source-report-summaries/. The public site uses them as dossiers, not as independently verified empirical consensus.