Skip to content

Latest commit

 

History

1,871 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

The ASI Stack

The ASI Stack: A Governed Systems Architecture for Advanced AI, with ASI as the Stress Case is Corben Sorenson's living technical book about building advanced AI as an auditable system rather than an unconstrained model.

This repository contains the canonical Quarto manuscript, its dynamic structure, public-safe source and evidence metadata, executable validation, and Lean proof workspace.

Choose the product you need

Product Best for Open
Narrative book A human-oriented route through the architecture. Narrative route
Architecture reference Interfaces, invariants, protocols, tests, proofs, and implementation horizons. Architecture index
Evidence registry Claims, support states, sources, tests, proofs, releases, and residuals. Evidence registry
Original paper library Corben's digest-bound papers and architecture sources, with links to the chapters they informed. Read the original papers

Projection boundaries are defined in docs/product_contracts.md.

The canonical current book is the live 89-chapter site. Use its AI view for the complete research scaffold or Human view for the cleaner prose projection. Human view is a reading aid, not a reviewed reader-release manuscript.

The mutable root site and /latest/ are the canonical current publication surfaces and contain all 89 manifest chapters. Versioned tags and GitHub Release assets are immutable historical snapshots. v2.3.0 remains the latest completed immutable HTML release. The Post-v2.3 Evidence Competence, Transfer, and Publication Roadmap is the sole active roadmap, with machine authority in roadmap_records/post_v2_3_maintenance_transfer_and_publication_status.json. All 89 live chapter-core claims remain at argument.

60-Second Trust Surface

Current canonical metrics (generated from machine records): 89 manifest chapters; 495 public-safe records; 89 chapter-core claims; 89/89 chapters externally positioned; 0 promoted core claims; 115/115 accepted transitions identity-resolved (25 direct, 61 subclaim, 29 proxy; 0 parent movements).

What this is: a public living-book research program and evidence system for governed advanced-AI systems, using ASI as the extreme stress case.

What this is not: not a validated ASI implementation, not a deployed safety system, and not a benchmark-proven architecture.

The inventory has 495 public-safe records; 89/89 chapters are externally positioned with 0 explicit external-baseline exceptions. External positioning is not exhaustive literature synthesis. All 89 chapter core claims remain at argument; the core-claim disposition ledger records 89 per-chapter core-claim dispositions, 22 accepted no-change transition dispositions, 67 accepted no-promotion dispositions, and 0 promoted core claims. The 25 accepted non-core upward evidence transitions are recorded in the non-core evidence ledger, alongside 61 accepted blocks_promotion decisions and no chapter-core promotion.

Appendix C is the claim/support-state ledger. Appendix G separates Corben-authored and local-project sources from Appendix H, which records external literature. The Corben paper library publishes 53 exact, digest-bound original manuscripts as HTML reading pages and links each one to its current chapter assignments; publication exposes lineage but does not validate or promote the papers' claims. docs/chapter_external_grounding_status.md reports current per-chapter coverage. Novelty positioning is not proof of novelty. No independent external review is claimed; see the review ledger.

Contribution-level prior-art positioning is recorded in docs/defended_contribution_prior_art_positioning.md and checked by scripts/validate_defended_contribution_prior_art.py; positioning is not proof of novelty.

Evidence-laundering failure cases and their non-promotion boundaries are recorded in docs/evidence_laundering_prevention_case_studies.md and checked by scripts/validate_evidence_laundering_case_studies.py.

Historical chapter merges, redirects, and retained URL lineage are summarized in docs/chapter_consolidation_sequence.md and docs/chapter_consolidation_url_history_policy.md; the full history is checked by scripts/validate_chapter_consolidation_sequence.py. The current 89-owner publication-consolidation plan is the P7.1-EM section of the active roadmap, with its exact machine contract in roadmap_records/post_v2_3_maintenance_transfer_and_publication_status.json. The independent human-reader edition is specified by the canonical 26-unit writing outline; the older 22-unit product spine remains historical candidate custody.

The compact public contract is checked by scripts/validate_trust_surface.py.

Start Here

Need Canonical owner
Read the book Living book
Change parts or chapter order book_structure.json
Draft chapters or plan proofs docs/book_outline.md
Execute current work docs/post_v2_3_maintenance_transfer_and_publication_roadmap.md and its machine status
Audit claims and evidence Appendix C and proofs/proof_manifest.json
Read Corben's original papers papers/index.qmd and the paper links in Appendix G
Understand repository ownership docs/repository_map.md
Ingest or update the book docs/living_update_workflow.md

Earlier roadmaps, review packets, probes, and release receipts are historical records, not competing instructions.

Work Locally

Prerequisites: Python 3, Quarto, Node.js for browser checks, and Lean/Lake for formal proofs.

python3 scripts/sync_scaffold.py
python3 scripts/sync_proof_manifest.py
python3 scripts/validate_book.py
quarto render --to html
quarto render papers --to html

book_structure.json is the only ordering authority. Add, move, merge, or remove chapters there, then regenerate the scaffold; chapter numbering and navigation update automatically. Do not hand-edit _quarto.yml.

Do not report a theorem as proven unless lake build passed, a test as passing unless it ran, or a source claim unless the source was actually reviewed. Proof artifacts, citations, and implementation code do not silently promote chapter-core support states.

The registered Phase 5 harness set is defined in experiments/phase5_harness_registry.json, summarized in docs/test_harness_status_ledger.md, checked by scripts/validate_phase5_harness_registry.py, and executed through scripts/run_phase5_harnesses.py.

Artifact Discipline

The normal loop is HTML-first and content-first. EPUB, PDF, DOCX, and audio generation are major-version release work only, after the manuscript is approved for that gate. Audience and strip policy lives in editions/release_profiles.json.

Avoid disposable tracked files. Update the canonical owner for a fact or decision; use ignored build/ or /tmp output for diagnostics; batch expensive validation; and remove regenerated local artifacts when their review is complete. A standalone report belongs in Git only when it is a durable evidence, governance, or release record with a named consumer.

See CONTRIBUTING.md for the internal change gate and the repository map for storage classes.

Rights

At exact tag v2.3.0, rights are routed per file: cleared author-owned prose and figures are CC BY 4.0, cleared software-like artifacts are Apache 2.0, and excluded paths receive no grant. Later drafting states remain all-rights-reserved unless another exact release ledger says otherwise. The manuscript is not an invitation for unsolicited contributions. See LICENSE.md, NOTICE.md, and CONTRIBUTING.md.

About

A governed systems architecture for advanced AI, with ASI as the stress case—living book, reference, and evidence registry.

Topics

Resources

Contributing

Stars

3 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages