Patrick D. McCarthy
A paper series deriving the architecture of intelligence from a single generative constraint: bounded active context — a hard limit on the number of propositions simultaneously available for inference.
| Paper | Title | |
|---|---|---|
| 0 | All Software Is a Graph |
| Paper | Title | |
|---|---|---|
| 1 | Graph Structure Is Necessary for Managing Consistency and Availability Under Bounded Context | |
| 2 | Convergence of Control Architectures to Escalation Under Bounded Context | |
| 3 | Knowledge as Normalization: Consistency Under Partition Separates Shared Propositions from Action | |
| 4 | Evaluation-Driven Descent, Encoding Permanence, and the Structural Invariants of Intelligence |
| Paper | Title | |
|---|---|---|
| 5 | Genome as Projection: Coupling Channels Predict Cancer Survival and Suggest Drug Response Principles | |
| 5b | Not Incidental: Developmental Phenotype Predicts Adult Cancer in HR-Repair Carriers | |
| 6 | Coupling-Channel Structure Predicts Cancer Survival | |
| 7 | Graph Consequences of DNA Serialization | |
| 8 | The Cost of Computation: Shared Molecular Machinery Between Learning and Cancer | |
| 9 | Presence Without Performance |
| Paper | Title | |
|---|---|---|
| 10 | The Genome as Knowledge Graph: Governance, Scaling, and the Architecture of Multicellular Life | |
| 11 | The Ratchet: Evolution and Cancer as Opposing Failures of the Same Mechanism |
| Paper | Title | |
|---|---|---|
| 12 | Why Different Folds: Book Thickness, the Anti-Index, and the Functional Contact Graph of the Human Genome |
Paper 12 applies the book thickness theorem from graph theory to chromatin folding, showing that the genome's functional contact graph requires multiple tissue-specific folds. Co-essentiality clustering (DepMap, 14,129 genes) independently recovers the cancer channel architecture from Papers 5-6 at the same rate as Reactome's curated pathways (9.0x vs 8.5x enrichment, statistically indistinguishable). Three cancer failure modes — broken node, broken edge, broken fold — map to LOF, missense, and GOF mutations respectively.
Interactive Jupyter notebooks that reproduce the empirical results using public data. Designed to run in Google Colab or locally.
| Notebook | Paper | Description | Data needed |
|---|---|---|---|
| paper5_reproduce.ipynb | 5 | Channel count vs mutation count survival analysis across MSK-IMPACT cohorts | cBioPortal (auto-download) |
| paper6_reproduce.ipynb | 6 | Channel structure survival prediction | cBioPortal (auto-download) |
| paper7_reproduce.ipynb | 7 | Graph position consequences of DNA serialization | cBioPortal (auto-download) |
| paper10_reproduce.ipynb | 10 | Genome as knowledge graph | cBioPortal (auto-download) |
| paper12_why_different_folds.ipynb | 12 | Book thickness, co-essentiality clustering, tissue-specific folds, three failure modes | DepMap (~430 MB, auto-download) + committed derived data |
The Paper 12 notebook is the most comprehensive — it walks the deductive chain from graph theory through polymer physics to cancer biology, generating all figures computationally. Pre-computed Hi-C results from 21 tissues (Schmitt 2016) are included as derived data; the raw Hi-C (~467 GB) is not required.
Companion essays for a general audience, each grounded in one or more papers from the series.
| Essay | Paper | Core argument |
|---|---|---|
| Separating and Bounding Concerns | 1–3 | Factoring is mandatory under bounded context |
| Transformers Are Dinosaurs | 3 | CAP theorem forces the graph; LLMs are the availability layer |
| Stop Building Ralph Loops | 4 | Learning that doesn't build durable structure isn't learning |
| Cancer Is an Org Chart Problem | 5 | Channel count = levels of org chart compromised |
| Shadows on the Wall | 6 | The right simplicity explains more than the wrong complexity |
| Every Story Has a Spine | 7 | Language and DNA are both serial recordings of graphs |
| What if Cancer Is Rational? | 8 | The machinery that enables learning is the machinery cancer exploits |
| Sam Altman Can Eat My Ass | 0 | The software engineers are coming for you |
The knowledge-graph/ directory contains ~150 YAML nodes with full provenance — definitions, theorems, equivalency claims, novel results, emergent predictions, and references. Every node carries (attribution, evidence, derivation) triples. Edge relations include derives_from, evidences, grounds, overlaps_with, generalizes, and others.
See publications/research/CONFIDENCE_CHAIN_PAPER.md for the formal epistemic provenance framework and knowledge-graph/NOVELTY-ANALYSIS.md for cross-framework comparison.
The analysis/ directory contains empirical studies:
- Mutation survival atlas — per-gene per-cancer-type hazard ratios
- Channel survival — Kaplan-Meier by coupling channel
- Mutation interactions — gene-pair interaction effects on survival
- Metastatic tropism — organ-specific mutation patterns
- Escalation entropy — information-theoretic measures of escalation chain disruption
- Treatment-channel matching — drug sensitivity aligned to escape channel
- Protective variants — hypothesis: coupling-channel-strengthening GOF variants enrich in high-risk individuals who never develop cancer
- Co-essentiality clustering — DepMap CRISPR gene effect → functional contact graph → channel recovery
- Hi-C tissue-specific folds — GM12878, K562, IMR90 (Rao 2014) + 21 tissues (Schmitt 2016)
- Book thickness computation — edge counts at multiple granularities → predicted fold count
- Mutation spectra by channel — IntOGen driver mechanisms (LOF/GOF) × channel × age
- SEER age distributions — cancer type incidence by age → three failure mode classification
- Inference layer (
inference/) — document embedding, clustering, and keyword extraction pipeline - Annotated formulas (
docs/annotated/) — every formula in the paper series explained
references.bib — BibTeX file shared across all papers.
© Patrick D. McCarthy. All rights reserved.