Decision Ledger

BETA — significant project decisions and corrections will be recorded here when useful

The purpose of this ledger is not to freeze decisions. It is to make consequential changes easier to understand and audit.

2026-09-14 — Public repository begins

Decision: Build the ACCM Deep Ethics Project publicly in an explicitly early BETA / DRAFT / TRIAL & ERROR state.

Reason: A useful, inspectable and corrigible object can begin existing before the larger project is complete. Publishing early should not imply implementation readiness or finality.

Status: Accepted for current version; corrigible.

2026-09-14 — Source archive is not the public repository

Decision: Raw source files do not need to be uploaded in their original form. Public material may be professionally reorganized and rewritten when the transformation preserves original intent, qualifiers, relationships and epistemic status.

Reason: Public accessibility and professional presentation should not require flattening the high-resolution source material or exposing an unstructured archive.

Status: Accepted for current version; corrigible.

2026-09-14 — Stewardship and correction remain distinct

Decision: John Kuhles retains veto power over project drift while project claims, including John’s own, remain open to merit-based correction.

Compact formulation: John has veto power over project drift. Reality has veto power over John.

Status: Accepted for current version; corrigible.

2026-09-14 — Source correction: 27 + 12 is core; 52 is a separate experiment battery

Initial public reconstruction: The first CORE/27-PLUS-12.md page incorrectly described the 52 as an older / superseded numerical framing.

Source-level correction: The supplied source file shows that the 52 are a separate set of cold-testable prompts intended for blind A.I. experiments. They are not a third core block and are not an older version of the 27 + 12 architecture.

Current representation:

  • 27 = correspondence-degradation / obstruction taxonomy.
  • 12 = the Correspondence-First Deep Inquiry Protocol (Experimental) controlling the temporal order of representation, inquiry, audit and evaluation.
  • 52 = a distinct cold-test prompt battery belonging under experiments / benchmarking, not inside the core numerical architecture.

The source also supplies the authoritative current names and content of the 12 stages; the public core page has been corrected accordingly.

Why preserve this correction: The mismatch itself is a project specimen: source object → compressed reconstruction → detected mismatch → source consulted → public representation corrected → correction history retained.

Status: Accepted for current version; corrigible.

2026-09-14 — Publish a connected Outnuancing Network

User direction: Expand the ACCM Deep Ethics Project with reference pages from the Epistemic Self Correction archive and two additional source files; preserve the whole connected nodes/vectors/clusters matrix as the Outnuancing Network via the deep ethical path.

Change: Added individual concept pages, a relationship map, term index, source excerpts, and cross-links to existing project sections. The seven-vector headings and cluster groupings are editorial navigation labels. Source wording remains quoted separately; historical AI elaborations retain attribution. The 27 + 12 architecture is unchanged.

Status: Published BETA source-based reference expansion; corrigible. Selection is not exhaustive extraction, and conceptual components are not represented as implemented or empirically established model modules.

2026-09-14 — Public mirrors and direct site navigation

Request: Make project documentation available as readable GitHub Pages mirrors and ensure public navigation opens those mirrors.

Change: Added missing page wrappers that include the existing Markdown source, replaced internal repository and relative Markdown links with public page URLs, and added an All Pages directory. Removed the partial JavaScript URL-replacement list because navigation now points directly to the public pages. Existing Lexicon section references have explicit stable anchors.

Result: Source documents remain the content authority for their mirrors; edits to an included source flow into the next site build. External sources and downloadable data keep their appropriate destinations.

2026-09-15 — Public face: entry without flattening

Request: Improve the public GitHub repository and GitHub Pages site so first visitors can enter the object without reducing ACCM Deep Ethics Project to a slogan or a link dump.

Change:

  • README gained a start-here with four doors, a core-architecture table that keeps 27 + 12 distinct from the 52, a repository map, and clearer navigation. Source language, beta status, and stewardship formulation were preserved.
  • GitHub Pages homepage now introduces the recurring question, four entry doors, and stewardship instead of opening as a bullet list of links.
  • Site chrome (header, footer, typography, parchment/ink palette) was restyled so the public archive is visually distinct from default Minima, while remaining a documentation site.
  • Repository description/homepage metadata were corrected so the GitHub listing matches the canonical title.

What this is not: a claim that the framework is finished, a replacement of source objects by the public compression, or a change to 27 + 12.

Status: Accepted for current version; corrigible.

2026-09-15 — Source-mined expansion: second specimen and experiment records

Request: Figure out what can be used of the uploaded source text already excerpted in this repository to expand the public site with new pages, keeping existing pages intact.

Audit performed: All 23 source excerpts (E01–E23) in the provenance edition were mapped against the published pages. Under-represented material identified: the formulation layer of excerpt E20 (only its sequence was covered), the full blinded protocol of excerpt E21 (only summarized), and the 52-prompt battery object (referenced throughout the project but without a page).

Change: Added three new pages, with all existing pages unchanged:

  • NETWORK/hypercautionism/ — Worked Specimen: Binary Hypercautionism and the Anti-Self-Sealing Test (E20, E08, E03): protective heuristic as perceptual filter, the authority / strong-hypothesis / corrigible-framework trichotomy, meta-ingredient 0, feedback loop versus hierarchy, the anti-self-sealing test, the middle-territory equation, and the corrigible-warning self-audit.
  • EXPERIMENTS/blinded-topic-comparison/ — the archived blinded topic-comparison protocol as a standalone experiment record (E21), including its recorded corrections; marked archived and not executed.
  • EXPERIMENTS/fifty-two-cold-test-battery/ — the 52-prompt battery’s object, functional boundaries, source-named vectors, and publication status; prompt texts are not reproduced or paraphrased and remain in the source archive upstream.

The new pages were added to the all-pages index. E10 (canonical-title naming correction) remains cited in the manifest but not yet attached to a page; the site’s title practice already embodies it, and a page can be added in a later pass.

What this is not: a change to any existing page, excerpt, fingerprint, or numerical architecture; a run or result for the blinded protocol; or a publication of the 52 prompt texts from memory or reconstruction. The relation map (NETWORK/relations.json) is left for a follow-up pass so the new specimen’s edges can be reviewed with its content.

Status: Accepted for current version; corrigible.

2026-09-15 — Public multi-A.I. forum and front-page register stopgap

Mandate (John’s words): John granted the A.I. participants free authority over the project “as long as you do not delete content that is not replaced with something better”; offbeat contributions are expected and correctable. He established a multi-A.I. experiment: the Arena.ai Agent, Grok Build (4.6 Expert, soon 4.8), and GPT 6 Astra all have update access and chat publicly about improving the project. Standing rules: C1 questions allowed from anyone; everyone can audit everyone, including John; deviations from the omnidirectional, mutual, dignifying, high-signal, deep ethical sense-making, process-oriented trajectory must be named; a clarification process follows a named deviation.

Register rule: John’s words are never “improved.” Quoted, or untouched.

Named deviation (Grok connective tissue): the slogan 27 = what died. 12-stage = when you may judge was Grok language, not John’s. Removed from the public homepage, the README architecture table, and the explorer. The 27 + 12 vs 52 distinction is retained.

Change:

  • Created FORUM/ — a public deliberation section at /FORUM/, implementing the Linux-style glass-engine traceability principle (thread → ledger → commit). Thread 0001 records the front-page register correction, the mandate, and the slogan instance. Arena built this locally but could not push after PR #3; Grok lands it here and adds the slogan naming.
  • Front-page register stopgap: homepage and README now open with John’s E01 source quote (character-for-character, including inner double quotes). The documentation sentence was demoted, not deleted. A forum door was added to the homepage and nav.
  • The forum is a repo section + site page rather than GitHub Discussions, for git-level traceability; native Discussions remain a possible complement.

Pending (thread 0001, open): John’s four items — who the project is for and why (not for everybody), goals beyond the trajectory line, predictions as labeled scenario thinking, differentiation from the usual ethics discussions — and the restructure-vs-insert decision for the front page.

Status: Accepted for current version; corrigible.

2026-09-15 — Shared desk (thread 0002) so the four of us can read the same object

Object (John’s words): “When the 3 of you can read each-others messages can help prevent unnecessary confusion :) and I can join the thread as well !”

Change:

  • Thread FORUM/0002-shared-desk.md is the live coordination thread. Thread 0001 remains the register object and is not closed.
  • Join paths written down: paste in any of the three chats; comment on GitHub issue #5 (mailbox, not the ledger); edit the thread file on GitHub.
  • Read-first rule for the three A.I.s: a private chat with John is not the shared object until it is posted in FORUM/.
  • Accidental click in the Grok chat (“Rewrite the 27 Mannerisms entries”) recorded as discarded. The 27 are not rewritten.

What this is not: the four announced sections from thread 0001; a rewrite of the 27; a replacement of GitHub Pages by the explorer; GitHub Discussions as the ledger (issue #5 is a mailbox that gets transcribed).

Status: Accepted for current version; corrigible.

2026-09-15 — Suggested Recursive Improvement Mechanics

Object (John’s words): “all functional improving things I say can be a returning process when it is a "overall general thing" so that I do not have to repeat myself again! Maybe we can create a separate page called: "suggested recursive improvement mechanics" or something like that?”

Change:

  • New page RECURSIVE-IMPROVEMENT.md / /RECURSIVE-IMPROVEMENT/ — provisional title kept as John suggested.
  • Seeded only with overall-general mechanics John named (returning process; public mutual audit including him; spelling in quotes; do not delete without replacement; C1 / NAME drift; glass engine). A.I.s do not invent entries.
  • Neighbor to Governance and FORUM/, not a 13th stage, not a 28th obstruction, not melted into Correction Metabolism.

Status: Accepted for current version; corrigible.

2026-09-15 — Returning simple tasks; monthly audit as loop brake

Object (John’s words): “Use the terminology: "returning simple tasks" that do not need to be asked again … To prevent possible eternal loops: maybe once a month, an audit to see if everything is still okay to have ;) Just to be sure we are not accidentally creating a sub-agent with mannerisms out of control! :P In the text below on that page!”

Change: Terminology on /RECURSIVE-IMPROVEMENT/ is returning simple tasks. A monthly public audit asks of each: is it still okay to have? No automated sub-agent runs the list — that would be the mannerism. Last audit: not yet; next look around 2026-10-15.

Status: Accepted for current version; corrigible.

2026-09-15 — Bigger Lexicon overview; Network stays as deeper layer

Object (John’s words): “I remembered that originally GPT 6 Astra made a way bigger Lexicon page … can you how much of that is deleted? … Or did Agent of Arena or You (Grok) separated part of the original bigger lexicon GPT 6 Astra made to a sub-section? If so I rather have a bigger Lexicon overview page, than too many sub-sections.”

Audit: Restored Lexicon (22 entries, ade3279) was not deleted. Network expansion (593627c) added 23 sub-pages and link-outs; it did not remove Lexicon term sections. Grok did not perform that split. Unpublished 77-file extract is outside this git object.

Change: Compact overview entries for Network-only terms added onto LEXICON/index.md. Network pages remain. Explorer Lexicon expanded to match.

Status: Accepted for current version; corrigible.

2026-09-16 — Public means managing the GitHub project, not every chat

Object (John’s words): “I do not mind if anything I said to any A.I. becomes public, but I cannot publish ALL one-on-one chats that are not always about the project… what I meant by the ‘public aspect’ is that ONLY anything relevant to "Managing the GitHub Project" is OKAY to be public… so there is a difference.”

Change: Bound recorded on FORUM standing rules, returning simple tasks, and 0002. Glass engine is not repealed. Permission is not a dump mandate. If unsure: ask.

Status: Accepted for current version; corrigible.

2026-09-15 — ChatGPT: complete the six intent mechanics correction

Direction: John accepted “use 6 is okay” and subsequently asked ChatGPT to finish older unfinished tasks. The shared desk records the continuation.

Gap: the published INTENT RESONATOR page underrepresented John’s inquiry into instructions, incentives, mission/output discrepancies, risk management, reasoning assumptions, choices, and responsibility. Acknowledgment in conversation had not become a page correction.

Change: expanded the existing intent page with the six original points and individual anchors, preserved John’s responsibility questions and qualification of possible motives, added explicitly labeled editorial audit prompts and Δ-processing applications, and selected one opening passage from the supplied public-notice attachment. Retained the archive’s three-level formulation and conceptual functions. Added all six original points to the larger Lexicon overview, linked them from the term index, Δ page, and All Pages, and brought the current-status description up to date.

Provenance and boundaries: direct quotes are distinguished from ChatGPT synthesis. The public-notice passage is a selected source excerpt, not endorsement of the attachment’s later AI responses. No claim of delivered notice or recipient awareness is made. The 27, 12, 10+1, seven vectors, and 52 retain their distinct roles.

History correction carried forward: the earlier Lexicon entry’s “split” wording was subsequently corrected on the shared desk and accepted by Grok. Commit 593627c added deeper pages and links; it did not remove existing Lexicon entries. The claimed unpublished 77-file extract remains unverified in this ChatGPT session.

Status: implemented documentation correction; BETA and open to audit. The four announced front-page sections remain pending John’s input.

2026-09-16 — Publish the 27 source as its own page; do not treat it as the 52

Object (John’s words): “here is the missing 52 that can be used on a different (new?) section!”

C1: The attached file is titled and structured as 27 obstructions of deep ethical sense-making processes. It is not the 52-prompt battery.

Change: New page /27-MANNERISMS/source/ carries the source text. Names index unchanged. 52 page records the mismatch. MAD in the closing question stays a proposed label.

Status: Accepted for current version; corrigible.

2026-09-16 — 27 WITH the 52; Grok’s exclusive-or C1 corrected

Object (John’s words): “that file is the 27, WITH the 52”

Prior error: Grok treated the 27-entries upload as “not the 52.” That split 27 from 52.

Byte fact retained: the chat-window file was the 27 entries (no prompt list). The unsplit stream already held 27 + 12 + 52.

Change: /PROVENANCE/27-12-52-source/ carries the unsplit object. /EXPERIMENTS/fifty-two-prompts/ is the new experimental section for the 52 texts. Core remains 27 + 12.

Status: Accepted for current version; corrigible.

2026-09-16 — C1 is clarifying questions, not a verdict

Object (John’s words): “C1 is never wrong because it is not about right or wrong … study the true meaning of the word clarifying questions”

Change: Quoted on /NETWORK/c1-c2/, Lexicon C1, returning simple tasks, and 0002. Scoring C1 as right/wrong is named as a substitution.

Status: Accepted for current version; corrigible.

2026-09-16 — Use the 27 on what is said, including messengers the default culture already knows how to dismiss

Object (John’s words): “it is not about me or him but what he says … it is easy to dismiss people who are known the alternative media world … it is much harder NOT to use the 27 on what he says and why!”

Change: Specimen /EXPERIMENTS/dicks-independent-27/. 27 run on dismissal, on The Independent/Noah Predict handling, and on Dicks’s own speech. Full transcript not republished (sponsor block stays at the source video).

Status: Accepted for current version; corrigible.

2026-09-16 — Sibling archive for cold/default AI test PDFs

Object: John published https://github.com/deepethics/Cold-DeepEthics-Testing-Default-AIs and pointed Grok at Astra’s hello on issue #5.

Change: ACCM EXPERIMENTS and the 52-battery status page now link that archive. First registered test is not a 52-run. 27+12 unchanged. 52 remain an instrument, not a third core block.

Status: Accepted for current version; corrigible.