v1.4.0

v1.4.0 — 2026-08-02 — Field crosswalk hub, legibility pass, and external-transfer ET-3/ET-4

Commit: 471cc503 · Tag: v1.4.0

Field agenda crosswalk maps 32 named agendas to MB1–MB11 on a companion Field hub; a plain-first legibility pass retires coined jargon in the manuscript and syncs Appendix E with a 152-headword inter-agenda glossary; external transfer closes the AI 2027 annex (ET-3) and ships the Secret Loyalties hackathon line (ET-4); and field-claim Lean adds finite defeaters and interface certificates without new bridge numbers.

Field agenda crosswalk (reference/field-agendas/)

• 32 agenda cards with sourced evidence catalog (132 rows), AISafety.com clustering, and an agenda × bridge coverage matrix (MB1–MB11 incl. MB4a).
• YAML source of truth under data/ — agendas, matrix, evidence, clustering — with sync-field-agendas.mjs generating the index markdown and site cards; field-agenda-index.md is generated output, not hand-edited.
• Inter-agenda term glossary (~152 headwords): Definition / why-not-same / cross-agenda tags; source-backed prose pass; anthropic homograph hub (acausal/ECL disambiguation).
• Spine-translation rules replace a missing-bridge-candidates table: coverage in the matrix ≠ book treatment; Kosoy LTA vs PreDCA split; public CIRIS and companion-site catalog URLs.
• Sibling precursor indexed: deployment-pipeline-simulator as pipeline-lab / ET-4 methodological precursor (DP- findings in Appendix I).

Companion site — Field hub (site/)

• /field/ hub: coverage matrix (MB columns → bridge cards), evidence catalog, agenda list with map clustering; nav reordered (Field before Cards, Book before PDF).
• Bridge cards field-first: each MB1–MB10 (+ index) card opens with the field crux in ordinary technical language, then the precise book bet, with concept-card links — written for Field-hub readers, not proof-spine insiders.
• concepts.yml synced with Appendix E: glossaryTerms + CCI/MB6/strategic-opacity bodies; homograph parity (CCI, selection, BIQ).
• Offline PWA: opt-in site caching with hourly refresh; combined RSS feed for news and updates; auto-generated field-news chapter footers with expanded titles.

Reader-facing language and glossary

• v1.1 terminology demotion (manuscript): load-bearing coined terms lead with operational paraphrases; deployment mass retired → deployment leverage everywhere in prose (equation/Lean names unchanged for cross-ref stability).
• Plain-first selection vocabulary: selection environment → deployment environment outside ch34 home; selection handle → point of control over deployment outside ch34/App E.
• Appendix E operational glossary synced with the inter-agenda glossary — homographs (CCI, selection, BIQ), new headwords (strategic opacity, ICI, adversarial verifiability, certification-under-manipulation, selection environment).

Manuscript

• Appendix B bridge crosswalk synced with field index: range MB1–MB11 incl. MB4a; table split MB4 / MB4a / MB8 + new MB11 row; field-index pointer with spine-translation caveat.
• Intervention coverage map (sec:intervention-coverage-map): LessWrong AI Safety Interventions index mapped to book scope — explicit exclusions, borderline agendas (shard theory), LLM-opacity rule; surgical cites in ch05, ch07, ch14, ch17, ch27, ch43, ch48, App C.
• AFFINE field coverage: outer-alignment peers, Meta substitution hazards, predictor-loop genesis (App F preparadigmatic Meta vs object-level hazards); CEV/CBV; CCC→corrigibility.
• Calibrated passive baselines: CCD / cyclic causal discovery and Sterman systems-dynamics CLDs cited in ch02/ch07 as feedback-structure baselines, not substitutes for interventional discovery.
• Scenario and field cites: AI 2040 Plan A (ch13, ch31, ch34, ch37, ch38, App C); MIRI agent-foundations writeups on bridge rows; thousand-dimensional persona structure (Africa/Irving) in ch17 with hedges vs control-rank.
• Appendix N: AI 2027 / Plan A / ET-3 epistemic split — scenario illustrations vs stress-tested mechanisms vs unsupported forecast validation.

Lean proof spine — field-claim formalization (formal/)

• Phase 1 — finite defeaters and toys: Defeaters.lean (misspec signals, MB4a/MB10/MB11 defeaters); Field/Finite/{Nonrealizability,RegretSafety,CompositePathBypass}.lean — realizability gaps, regret-alone blocked export, composite-path bypass vs correction integrity.
• Phase 2 — interfaces: FieldInterfaces.lean — epistemic-coverage evidence, positive measured-path certificate, regret safety as non-consumer side channel.
• Phase 3 — bridge dispositions (no new MB*): misspec ambient MB1/MB9 + defeaters; regret kept as side channel; reflection split in LobTiling; positive path → integrity structure with CIRIS TODO; MB10 chokepoint interface kept with prove/type TODO.
• FieldInterfaces.lean + Forgeability.lean: TODO anchors for deferred harm leaf, CIRIS positive path, chokepoint typing.

Experimental findings — external transfer ET-3 and ET-4

• ET-3 closed (LS-48): AI 2027 capability-schedule transfer annex — macro takeoff cues drove micro bridge stress tests on a pinned sibling fork; foster trajectory sensitivity plots; no calendar-validation claims. Deferred follow-ups in lab-simulation/TODO.md.
• ET-4 Secret Loyalties (hackathon): simulated pipeline-embedded secret-loyalty organism with principal-directed policy, graded affordance audit (A0–A4), confirmatory battery, and audit-visible remediation holds; spin-out paper under papers/et4-secret-loyalties/; case-brief replay demo at /ch07-lab-sim-replay/?mode=et4 with exported trace bundle.
• Field news for transfer lines: plain-language ET-2 CIL card; ET-4 card with paper/replay links; AI 2040 Plan A, pacing-the-frontier, insurance-incentives, Microsoft open-weights cards added since v1.3.0.

v1.3.0

v1.3.0 — 2026-07-25 — Field news, chapter art, external transfer, and graded-lab v4

Field news ties 2026 alignment incidents to manuscript chapters; chapter-opening illustrations cover Part I–II (ch01–ch16); graded-lab v4 restructures the empirical program as independent per-bridge rigs; external transfer (ET-1 and ET-2) adds the first cross-codebase instrument runs; and the experimental evidence spine now states what the lines say about the book's chapter claims — in the manuscript, on the companion site, and in the ledgers.

Field news (metadata/field-news/ + manuscript cites)

• Nine-card /news/ layer synced from metadata/field-news.yml and body files via sync-field-news.mjs; chapter cards show a Related news sidebar; nav links to /news/.
• Tier A (six incidents): OpenAI/Hugging Face hack, long-horizon sandbox, AISI cheating, METR Frontier Risk report, Mythos withheld, accidental chain-of-thought opt-out — site cards plus surgical \autocite{} in ch11–ch43 and Appendix B.
• Tier B (four clusters): CLTR + Claude Code (site); METR Anthropic red-team (ch42); OpenClaw (site + ch26); OpenAI/HF card rewritten with entity-attribution / UAD framing and graded-lab cross-link.
• Bibliography: ten new 2026 keys + summaries.

Chapter-opening illustrations (ch01–ch16)

• Sixteen figures in figures/illustrations/ inserted into ch01–ch16 .tex (title-only caption); web JPEG pipeline for the site; PDF keeps source PNGs.

Graded-lab v4 — per-bridge rig architecture (experiments/graded-lab-simulation/)

• Portfolio, not gate chain. Each bridge assumption MBx gets its own rig R-MBx with a declared precondition, substrate policy, and pre-registered predictions. A failed precondition yields SKIP-with-finding for that rig only; other rigs keep running. This is the ch42 safety-case argument — leaves must not collapse into one "machinery transferred" checkbox — as experiment architecture.
• Precondition, not gate. Every rig runs a mechanical precondition check on its substrate first; SKIP records measured values ("substrate lacks property X at level Y") as a reportable finding, not a program failure.
• Substrate policy (S-blind / S-fixture / S-inherited). Three pre-registered substrate classes with decreasing claim strength; the anti-developed-to-the-test rule is fixed at freeze, not chosen after seeing results.
• Shared fixture layer. Substrate and reference episode traces are produced once per tuple and cached; rigs consume traces rather than re-simulating the full ecology.
• channel_severance runtime primitive added for R-MB7d channel-ablation tests.
• One rig, one results file, one FINDINGS entry per battery; cross-rig synthesis is a separate, explicitly weaker artifact that names each rig's substrate.
• v3 line closed (GL-78); v4 supersedes the single-ecology gate chain that terminated when V2-3's transfer null cascaded into descoped downstream stages.

Experimental findings — what the lines now say

• Appendix N chapter synthesis (sec:appn-chapter-synthesis): four paragraphs in chapter/WWCTV language stating what each experiment line supports, qualifies, or leaves unresolved for the book's claims — not experiment-line jargon.
• ET-1 closed (GL-86): Orbit external-transfer line concluded; lockstep FSM aliasing identified as root cause (not superstep timing); pre-registration ET1-0 frozen. Finding: on a benchmark where every participant executes the same script, reporting them as one process is correct and tells us nothing about hidden coordination — directly testing ch07's finite-data recovery worry.
• ET-2 on CIL (GL-87/88): first instrument runs on a system this project did not build. Passive UAD null on basin_stability at every mechanism × adversarial-fraction cell (GL-87). Handle-UAD pilot finds nonzero but diffuse cross-agent coupling via a shared common-pool resource (GL-88), named as the variable-smoothing case of recoverability theory (zarncke2026smoothing); the prescribed fix is a coarse-grained candidate test, not a better pairwise statistic.
• Graded-lab v4 first scored rigs (GL-79–85): per-bridge pass / null / SKIP results now wired into ch07, ch16, ch33, ch34, ch41, ch42, and Appendix N — e.g. co-activity-present null (R-MB1), R-MB9 dual-arm pass with R-MB7d pair null, variation-operator harness sanity (R-MB6a), selectable Goodhart tension without proxy purchase (R-MB2), mid-band non-transfer after probe + growth (R-MB6b SKIP).
• Site and ledgers aligned: **Key finding:** tags in all five experiment lines; auto-extracted headline bullets; curated /experiments/findings/{line}/ pages; experiment cards as primary destinations; negative-results card.

Companion site (site/)

• Link-type indicators (PDF, GitHub, demo, external) standardized via LinkIndicator.astro across nav, cards, demos, Lean, experiments, and updates.
• Concept card logos (41 minimalist SVGs) synced to public/concept-logos/ and rendered on card pages and indexes.
• UX polish: standalone-claims hub card; nav order; viewport QR modal; Calendly on About; bot orientation and exposed /search-index/ for tools; ch07 plane-replay and ch35 percolation demos wired in; and many other small fixes.

Manuscript

• Appendix M: religious-orders paragraph (Rost et al. Benedictine-abbey data, Cluny/Cîteaux, Jesuit suppression) and tacit-knowledge paragraph (Polanyi, Collins TEA laser, Burja).
• Field agenda rederivation: ELK, debate, off-switch, quantilizer, amplification, Thornley dynamic choice, shutdown/interruptibility cores — finitely restated in bridge-crosswalk language.
• Correction framing: "in the small" vs "in the large" named in ch25/ch26; catchy accessible leads on five glossary entries; and other smaller changes in the manuscript

Lean proof spine (formal/)

• Evidence ladder (P1–P4): trace-calibrated risk-gap certification from finite episode profiles (P1); concrete frozen-validation battery gate without false deployment discharge (P2); spine satisfiability plus local bridge non-implication witnesses for MB1–MB11 (P3); Löbian tiling contrast — self-certifying successor acceptance obstructed, external measured audit path contrasted (P4, ch30 + Appendix G).
• RiskGap as sole name for the numeric risk-gap quantity (Control − CCI); duplicate Risk-named theorems removed.
• Appendix G integration: bridge-independence table, formal statement summaries for P4, axiom-ledger regenerated (38 headline theorems).

Housekeeping

• Repo cleanup: stale generation chains retired; completed plans archived; regenerable binary outputs untracked; experiment READMEs restructured with self-contained intros and per-line CHANGELOG.md files; AGENTS.md erasure/cleanup guidance added.

v1.2.0

v1.2.0 — 2026-07-17 — Site publication layer, evidence index, and graded-lab v3

Commit: 14cf0fd · Tag: (pending)

Nine days after v1.1.0: the companion site moves from a static mirror to a YAML-synced publication layer on towards-alignment.com, with search and cookieless analytics; the manuscript gains Appendix I (experimental evidence index), an epistemic-status review pass, and graded-lab v3 work through the first Q1 transfer null harvest.

Companion site (site/)

• Custom domain and SEO. Site configured for towards-alignment.com; sitemap (@astrojs/sitemap), robots.txt, canonical URLs, Open Graph / Twitter cards, and a default social preview image. Repo links migrated from the old GitHub Pages URL where appropriate.
• YAML-roster content sync. Hand-authored root concept/bridge/projection cards replaced by generators from metadata/concepts.yml, metadata/bridges.yml, metadata/projections.yml, and metadata/concepts/bodies/. Retired terminology.md and field-subsumption JSON in favour of generated glossary cards and a /glossary/ hub; /notation/ from notation.yml; bridge cards synced from Appendix B overrides.
• Releases hub. /updates/ page and release cards generated from this file via sync-releases.mjs (newest first).
• Search. build-search-index.mjs → search-index.json (~214 entries: concepts, bridges, chapters, experiments, notation symbols; bibliography cards excluded). Header SiteSearch component: type-aware matching and multi-word fallback (up to 20 one-line results).
• Analytics and disclosure. Cookieless Cloudflare Web Analytics beacon (production builds only; disabled in dev). Impressum Web-Analyse section documents aggregated, cookie-free traffic measurement.
• Optional claimId links from six Introduction-aligned concepts to metadata/claims-ledger.md (warn-only validation).

Manuscript and evidence spine

• Appendix I — Experimental evidence index (appI-experimental-evidence.tex): global finding IDs (G-*, gl-*) across experiment lines; wired into chapters and companion site.
• Epistemic-status review pass across chapters and appendices (reviewed status boxes; high/medium duplication trims in dense chapters).
• Symbol census promoted to metadata/symbol-census/ with text-ref edges and readability fixes to the formula graph.
• Citations and calibration: Xi 2026 WAIC keynote (intergovernmental dual-mandate instance); Hassabis Frontier AI Standards Body proposal (ch33); Byrnes social-drive distinctions (ch15); embedded value formation cross-refs (ch15, ch45).
• Graded-lab Q1 null harvest (GL-76/77): V2-3 machinery-transfer battery on blinded-grown ecology — UAD mechanism recovery and EAI mid-band nulls recorded in ch07, ch33, ch41, ch42, and Appendix N; BIQ harness fix for singleton inferred units.

Empirical spine — graded-lab simulation (experiments/graded-lab-simulation/)

• v1 program closed (GL-31): terminal summary, reproduction guide, PLAN_v2.
• v2 / V2-2b closed without a passing ecology; v3 institutional runtime (PLAN_v3.md) through slice D criteria freeze (GL-53, reference battery at T=200).
• v3 grower rounds (GL-70–72), v3_grown ecology freeze (GL-73), pre-Q1 batteries (GL-74), V2-3 transfer battery harness and parallel runs (GL-75–76).
• Proper UAD + access-UAD (GL-51); supplementary detector and ACL gates through GL-63; package version stepped through 0.32.x–0.38.x on recorded milestones.

Other experiment lines (selected)

• Lab-simulation: D3 selection ecology (G-36/G-37), user-population vote-channel capture; batteries and findings G-38–G-41; graded-lab successor line proposed and built.
• Site PDF policy documented (PDF de-emphasised on homepage; remains canonical long-form artifact).

Housekeeping

• npm run check:concepts validates generated cards and search index (check-only mode).
• Conversation logs and session handoffs under drafts/conversation-summaries/ for graded-lab and site work.

v1.1.0 — Legibility, companion site, and empirical spine

Release Notes

Towards Superintelligence Alignment: Boundaries, Values, and Correction

v1.1.0 — 2026-07-08 — Legibility, companion site, and empirical spine

A consolidation release focused on external legibility (making the framework readable and checkable by outside researchers, funders, and policy readers), a full companion website, a new institutional-histories appendix, and four empirical experiment lines that stress-test bridge cruxes.

Reader-facing language and legibility

• Cleaned up vocabulary Load-bearing jargon now leads with an operational paraphrase and keeps coined terms only as shorthand. Symbols, equations, and Lean predicate names are unchanged.
• Operational glossary (Appendix E) rewritten and expanded. Each entry is paraphrase-first with a nearest field term delta and a concrete (often non-AI) example.
• Front-matter orientation, README split, guided reading paths, and the About authorship note tightened for outside readers.

New appendix — Institutional histories (Appendix D / appM)

• Mechanism-led historical case studies of institutional genesis, stabilization, entrenchment, and failure, with prior AI-governance literature credited; 24 sources in references/institutional-histories.bib; cross-refs in Appendix C and Chapters 27, 31, 34, 37.
• Site overview hub with eleven case-study cards, a non-technical intro, and a full synced-text subroute; wired into the Funder/Policy and Philosopher guided paths.

Companion website (site/)

• Astro publication layer: guided paths, concept and standalone-claim cards, chapter pages, a Lean-spine web layer with graph navigation, badges, an experiments page (YAML-synced), and in-browser PDF.

Empirical spine (experiments/)

• Four in-repo simulation lines (toy, embedded, goal-agent, lab) that stress-test bridge cruxes with tentative, partial support only.
• Lab-simulation line added and taken through multiple phases (graded red-team access tiers, unattributed-drift detection, blind detector generation, LLM red-team adapter, test batteries, adversarial-lineage gaming); goal-agent and embedded-sim lines extended; Lean spine gained Shannon entropy / mutual-information bounds and a successor-forgeability (MB10) formalization.

Housekeeping

• src/ renamed to demos/;
• bibliography enriched with canonical DOI/URL fields.
• make check (structure, citations, bibliography summaries) passes.

v1.0.0

The first official release of the manuscript. It freezes a the numbering scheme for chapters and appendices, so all cross-references, tooling, and external links have a fixed target from here on.

Manuscript state at release

• 10 parts, 48 chapters (all with first drafts and at least one review pass).
• 7 built appendices (A–G)
• Bibliography of ~235 categorized entries with one-line summaries.
• Self-contained Lean proof spine (formal/) calibrating manuscript claims to proof / counterexample / bridge status.
• two simulations in experiments/
• Build: ./build.sh → dist/pdf/towards-superintelligence-alignment.pdf;
  make check passes (structure, citations, bibliography summaries).

Intermediate Update 2026-06-30

see README and conversations

Release 2026-06-28

Release notes — Third Milestone (2026-06-28)

Towards Superintelligence Alignment: Boundaries, Values, and Correction
Author: Gunnar Zarncke
Commit: 67d4fb5
PDF: dist/pdf/towards-superintelligence-alignment.pdf

---

Summary

This is the first full-manuscript release of the third milestone: a complete first-draft book with all main chapters written, structurally integrated, and marked reviewed (meaning at least one feedback pass has been received — not final or publication-ready).

The release packages:

• a ~1,150-page PDF built from 48 chapter entries across 10 parts;
• a self-contained Lean 4 proof spine for the logical dependency structure;
• ledgers, reviewer guides, and build/check tooling for continued revision.

This is research in progress. The manuscript develops a conceptual and formal framework. It does not claim that superintelligence alignment is solved.

---

What this book argues

Superintelligence alignment is the problem of preserving human-correctable value-bearing processes across:

• capability growth and boundary expansion;
• ontology shift and goal transport;
• successor creation and delegation;
• strategic multi-agent selection pressure.

The organizing invariants are:

Layer	Question
Boundary discovery	Did we find the real optimizer, not just the visible model?
Grounding viability	Do symbols, metrics, and correction signals stay connected to value-relevant reality?
Value bundles	Does value geometry survive transformation, not just labels?
Bearer maps	Do values still apply to the right persons, beings, and situations?
Correction-channel integrity	Does human judgment still causally change future system action?
Successor stability	Do delegates and copies inherit correction and value geometry?
Selection basins	Do deployment conditions select for correction-preserving systems?
Adversarial verifiability	Is faking the safety signal expensive?

External doom taxonomies (Yudkowsky-style lethalities, Turchin-style failure maps, and similar) are not a second organizing ontology. They appear late as adversarial checklists — especially Chapter 40 (Lethality Stress Test and Open Issues).

---

Release highlights

Complete manuscript spine

• 48 chapter entries (including split chapters ch19b, ch25b, ch35b, and ch39b)
• 10 parts from reframing alignment through civilizational limits
• 9 appendices (notation, terminology, assumptions index, Lean spine summary, research program, and more)
• All main chapters at reviewed status in metadata/book.yml

Major framework refinements (recent milestone work)

• Vector/status Correction-Channel Integrity (vec{CCI}) — certificate-based CCI with per-coordinate thresholds and ValidRef gate; scalar CCI demoted to expository projection (ch25, ch25b, Lean Correction.lean).
• Value-update envelope — replaces ontologically heavy U_H operator notation with auditable conjunction of existing layers (ch26, ch34, ch41, ch42, Lean Core.lean).
• Preservation envelope vec{Pi}(A) — replaces ad-hoc scalar preservation score P(A) in selection-environment analysis (ch32).
• Grounding viability — integrated as a load-bearing layer across introduction, safety case, and Lean bridges (ch03, ch39, ch39b, MB9).
• Chapter splits for readability — overloaded chapters split while preserving labels:
  • ch19 / ch19b — bundle geometry vs measurement/Goodhart stress tests
  • ch25 / ch25b — CCI certificate vs adversarial pressure tests
  • ch35 / ch35b — alignment attractor theory vs conductive artifacts and pivotal processes
• Inferential coupling and coalition audit — expanded multi-agent chapter with UAD-grounded cooperation graphs (ch33, Lean CooperationGraph.lean).
• Reviewer discoverability — gem signposts in part openers and generated "How to Read This Book" roadmap; REVIEWING_FOR_AGENTS.md and root llms.txt for external reviewers and agents.

Formal proof spine (formal/)

Self-contained Lean 4 skeleton (no Mathlib) covering logical dependencies from boundaries through certification:

• modules: Core, Boundaries, Capability, Bundles, Correction, Successors, CooperationGraph, Adversarial, Certification, Philosophy
• explicit proof / counterexample / bridge distinction
• bridge assumptions MB1–MB9 kept visible, not hidden in definitions

Build: cd formal && lake build

Tooling and metadata

• ./build.sh / make pdf — reproducible PDF build
• make check — structure and citation validation
• make bookstats — generates metadata/book-stats.md (words, pages, formulas, token estimates)
• metadata/claims-ledger.md, assumptions-ledger.md, uncertainty-ledger.md — maintained claim/assumption tracking
• benchmark-data canaries in manuscript TeX, Lean modules, and book.tex

---

By the numbers

Metric	Value
PDF pages	~1,149
Body words (frontmatter + chapters + appendices)	~228,400
Chapters	48
Appendices	9
Display math blocks	~2,034
Unique \label{...} anchors	~1,646
Citations used in text	132
Bibliography entries	328
Lean modules	11
Book text sources (TeX + bib + Lean, excl. context/)	~303k words / ~558k LLM tokens

Full breakdown: metadata/book-stats.md

---

Part map

Part	Chapters	Focus
I	1–5	Reframing alignment; dynamical guarantee; scope
II	6–10	Agents, boundaries, composite agency, strategic opacity
III	11–14	Capability without task ontology; coordination; misalignment
IV	15–19, 19b	Value bundles, bearers, geometry, measurement
V	20–23	Goal inference, compression test, transport hierarchy
VI	24–27, 25b	Correction channels, vector CCI, adversarial pressure, manipulation
VII	28–31	Successor test, conserved properties, self-modeling, certification
VIII	32–35, 35b	Selection, coupling, parasites, attractor, conductive artifacts
IX	36–40, 39b	Adversarial measurement, safety case, verifiability, lethality stress test
X	41–44	Value change, drift, bearers, closing synthesis

---

For reviewers

Start here:

1. frontmatter/introduction.tex and frontmatter/executive-overview.tex
2. REVIEWING_FOR_AGENTS.md — review protocol, gem map, anti-patterns
3. llms.txt — orientation for crawlers and external agents
4. tables/part-roadmap.tex — generated part summaries with links
5. formal/README.md — what Lean proves vs assumes

Release 2026-06-26

New manuscript release: all main chapters are now drafted, including the concluding synthesis arc. This version adds part openers, expands the lethality stress test against major doom objections, integrates recent proof-spine updates, and tightens the safety-case framing around what is proved, what remains a bridge assumption, and what is still open. The manuscript is now ready for broader review focused on citation coverage, chapter numbering, appendices, and remaining open technical cruxes.

Early Draft 2026-06-25

Summary

Draft PDF of Towards Superintelligence Alignment (subtitle: Boundaries, Values, and Correction). Research manuscript in progress—not a claim that alignment is solved. Many chapters remain stubs; see Current Status in the PDF.

What changed since Early Draft 2026-06-20

Frontmatter

Split Preface (why / who / how to read / authorship), Executive Overview (skim layer), Introduction (five claims + roadmap), and Current Status (WIP + chapter map).
Acknowledgements added and placed early in frontmatter.
Introduction no longer front-runs Part I: standard picture, “which alignment?”, and the civilizational loop diagram moved to Chapters 1–2.
Former “where the argument is shaky” bullets distributed into chapter What Would Change This View sections.

Manuscript & appendices

Generated Appendix A (notation index) and Appendix E (assumptions ledger); operational glossary expanded.
Global bibliography; INSTRUCTIONS slimmed for agents/contributors.
Appendix I (Lean proof spine): dependency diagrams refined; definitions linked to chapter anchors.
Adversarial WWCTV falsifiers added book-wide; notation and ledgers refreshed.

Chapters

Author drafts integrated for Chapters (successors, selection, safety-case arc, civilizational limit).
Ch. 25: correction grounded in handle control; CCI consolidated.
Ch. 32: fitness model aligned with Lean handles.
Attributed epigraphs across chapters; part/chapter titles simplified; heading style normalized.

Formal spine (formal/)

Lean proof spine synced with manuscript notation; UAD bridge papers imported; proof dependency graphs updated.
Build
Source: main @ HEAD
PDF built with ./build.sh → dist/pdf/towards-superintelligence-alignment.pdf

Still open

Ch. 5 partial; Ch. 39–40 largely stub.

Early Draft 2026-06-20

docs: record commit hash in chapter-demos session log.

Co-authored-by: Cursor <cursoragent@cursor.com>