feat: scholarly codification of NSAM critical review (DocBook 5.2 + XSLT + RELAX NG)

[metavacua]

Summary

• Adds papers/ai_and_agents/principled-agent-architectures-review/ with a full scholarly document set for the critical review of "Principled Agent Architectures: A Neuro-Symbolic Blueprint Beyond Large Language Models"
• Canonical source is DocBook 5.2 XML; HTML5 and LaTeX are XSLT-derived outputs, never primary
• Deep research provenance: 105 agents · 1,307 tool uses · 2,247,194 subagent tokens (2026-06-23)

Contents

File	Purpose
src/00-metadata.xml	Dublin Core Terms + Schema.org ScholarlyArticle (XIncluded by articles)
src/01-original-paper.xml	Full NSAM paper encoded as DocBook 5.2 XML
src/02-critical-review.xml	6 verified findings, 11 refuted claims, 4 open questions, 5 priority revisions
src/03-repository-audit.xml	7-repo relevance audit with cross-repo synthesis
src/bibliography.bib	23 BibTeX sources (Logic-LM, Mamba, FNet, PCP, Scallop, A-NeSI, PyReason, …)
xsl/html5.xsl	XSLT 1.0: DocBook → HTML5 with DC <meta> + Schema.org JSON-LD
xsl/latex.xsl	XSLT 1.0: DocBook → LaTeX article class with longtable/booktabs
schema/critical-review.rnc	RELAX NG Compact: extends DocBook 5.2 with finding role/condition
scratch/formulas.md	Curry-Howard table, NSAM typing judgments, benchmark tables, Kautz taxonomy
scratch/notes.md	Session provenance, citation gaps, open questions, portfolio arch diagram
Makefile	make validate html latex — xmllint + xsltproc pipeline
README.md	Build instructions, findings table, repo audit table

Key Findings (adversarially verified)

Finding	Verdict
Logic-LM 39.2% figure strips dataset/model context	Confirmed with caveats
Mamba linear scaling valid; "5x throughput" unsupported (rejected 0-3)	Split
FNet 80% speedup accurate; 92–97% GLUE retention contested (rejected 1-2)	Split
Curry-Howard for agent planning anticipated by Proof-Carrying Plans (PPDP 2020)	Confirmed
Scallop/A-NeSI claims accurate; A-NeSI gradient estimator provably biased	Confirmed with caveats
PyReason features confirmed; "exact yet scalable" vs. peers unsupported	Split

Repository Audit

verumorphism (HDL calculus + proof/refutation threads + REST API) and larql-to-sparql (vindex + Wikidata/DBpedia semantic grounding) already implement both halves of the NSAM architecture — they are simply not yet wired together.

Build

cd papers/ai_and_agents/principled-agent-architectures-review
make validate   # xmllint against DocBook 5.2 RELAX NG
make html       # xsltproc → generated/*.html
make latex      # xsltproc → generated/*.tex

🤖 Generated with Claude Code