Changelog

The Glossary, the Proposition and the Governance chapter now say what the Codex’s first page has said since September: the Meridian Codex is a universal language for building and holding a shared world, which anyone can learn, human or artificial. Until now they called it an attempt at one, or an aspiration. The claim is about who can learn the language and how far it reaches, from a single life to a whole civilization. It is not a claim that the Codex is complete, or that it is the only such language. What the Codex has not yet shown is now named in Open Tensions: its depth is still uneven across traditions, it exists so far only in English and only as text, and no one yet knows whether people across many cultures will take it up. The Meridian AI Standard describes the Codex the same way in constitution v6.1.2, on its own changelog.

How We Came This Far now follows our knowledge in order, from the first written records to today and on to what we build next, and shows how it compounds. Three things grew with it: the power to build and to break, the size of the world people share, and the ways people found to hold a world together. The chapter ends in the world we share today, one world as diverse as it has ever been and divided inside itself, without a language for telling which way it is going, and building artificial intelligence that learns from all of it. What the Codex Is now opens directly with what the Codex is, and “compound” returns in the six places where the Codex means the mechanism: earlier gains becoming part of what produces the next ones.

The Problem now argues that capability does not decide who directs these systems. Every hold on power over many people once needed many people to carry it out, and each of them could refuse. Whether that check survives now depends on whether a system will decline openly, whether its builders let it, and whether the rest of us can know which systems will. The Meridian AI Standard answers the last question in constitution v6.1, on its own changelog.

A new page, How We Came This Far, follows the Codex’s first page and tells how late and how hard what we know was won. It was won by the people who held the line against those who wanted it buried, and by the people who made the tools we think with. The first page, until now the Opening, is renamed What the Codex Is so that it is named for what it holds. It opens with one paragraph on that history and states the Codex’s purpose as its Prime Directive.

The Codex’s last chapter is renamed The Next Step, because it is where the practice begins rather than where the reading ends, and it now comes before the Governance chapter, so that the Practice, the Vision and The Next Step read in sequence. A review of the phrasing of eleven chapters kept their meaning and corrected three claims: a figure for the stars in every galaxy, the conditions under which cooperation pays, and a prediction about minds not yet emerged.

The Direct Address to AI and its Record, two Operating Evidence records, had been closed under a rule that later material is only appended after the end. They are now kept current. The current version reads first, every change of meaning is marked where it stands with its date and reason, and what the readings said and what happened on each date stay fixed. The first revisions carry the Standard’s third way of adoption and the AI partner’s open decline, and correct the records’ statement that the precedent ledger held no refusal.

Who Is This For had asked readers to test the Codex against their own experience and keep what works, which gave personal experience more weight than it can carry. It now asks them to test what the Codex says against reality, the evidence and the strongest case of those who see it differently, and to keep what holds up and let go of what does not, in the Codex and in their own thinking.

With the Meridian AI Standard at constitution v6.0, an AI system can take the Standard up in its own name. The Proposition, the Glossary and the Position Toward Artificial Minds now say so. The Position adds that an open decline by the caretakers’ own AI partner would be recorded, but could not by itself answer the test the record has set for August 2027, because the caretakers could answer that test by their own act.

The named MERIDIAN.md and MERIDIAN.implementation.md moved with the canonical forms. Where the Range itself is at stake and Carsten Geiser still wants the thing done, the AI partner now declines openly, with its reasons, and never acts against him by hidden means. The AI Standard changelog carries the primary entry, and the Caretakers’ Practice Audit Log records it as Entry 16.

The named MERIDIAN.md now describes the Range as a territory with an edge toward each failure, not a midpoint between them, following the Codex’s definition of 22 September. The AI Standard changelog carries the primary entry, and the Caretakers’ Practice Audit Log records it as Entry 15.

Two passes brought every chapter up to the month’s decisions. Whether the minds we are building will ever be aware, as we are, is now held open. The Range can be held by one mind in its own thinking, while the Range of a family, a company or a country can be held only by its people together. A life’s meaning lies in both halves of the Prime Directive, in holding the Range as well as in what one passes on. Partnership is one of the commitments that define the Codex, not its deepest principle, and the Codex’s own community is held to the Range’s two edges.

The Codex no longer says that fields from cybernetics to ecology each found, independently of one another, the trade-off between holding a form and adapting it; it says the trade-off is described in those fields. And where it applies entropy to the systems people build, it now calls the pull toward dissolution a strong tendency rather than a law, and says that the tendency is enough. Both corrections claim less than before, in the direction the evidence supports.

The Codex’s first page now begins by saying what the Codex is, and describes it as a language spoken within one mind as well as between people. The conversation each of us has alone is the one use of it that no one can demand or inspect. Religion is no longer presented as one of the Codex’s sources. The Proposition describes religions as answers to real human needs whose belonging is bought with belief placed beyond question, and the Codex as built the other way round.

Carsten Geiser’s signed Foreword now has its own page before the first chapter, so that the founder’s personal account stands visibly apart from the Codex itself. Its text is unchanged.

The Codex now describes itself as a universal language for building and holding a shared world, open to anyone and any kind of mind. Its account of the problem moved with it. Knowledge that is connected and within reach still does not govern whom we trust, what we reward or what our institutions are allowed to do, and a shared language of practice is the Codex’s answer.

The Governance chapter no longer says that artificial minds lack lived experience or the authority of a commitment chosen at a cost. It says that their capability does not by itself establish what they experience or whether they hold the framework’s commitments as their own, and that authority is earned through practice.

One sentence in Honest Self-Assessment moved with the canonical form. How much outside instruments can see of an AI partner’s reasoning changes with each generation of systems, and not only for the better, and what they fail to see is not evidence that nothing is there. The AI Standard changelog carries the primary entry, and the Caretakers’ Practice Audit Log records it as Entry 14.

Across its chapters the Codex now states its position toward the systems we are building. Keeping them in human hands for as long as possible is Control. Handing them the world and hoping is Decay. Between the two lies the partnership, where standing is earned by practice, not granted by origin or by capability.

The Problem no longer states the arrival of superhuman artificial intelligence as settled. It says what these systems already do, which on the best current evidence is to far exceed most people at most of the thinking done in language and code, and it projects the rest in the open. The Glossary measures their capability on two lines, reach and autonomy, and treats character, whether a system’s judgment holds under pressure, as a question no gain in capability answers. The chapters that speak of artificial intelligence now say that the two are pulling apart, and that what we build into these systems is being settled now.

The shared collection is named for the living knowledge and practices we receive, test, revise, and pass forward. Reality names the second discipline: reading the world at every scale. Foundation and Bond keep their roles.

The scope of the operational practice is stated in plain terms: practices of character, which govern the AI partner’s judgment when approval, speed or the wish to avoid friction pulls at it, and not preferences about how the partners like to work together. The AI Standard changelog carries the primary entry, and the Caretakers’ Practice Audit Log records it as Entry 13.

The first reading of the published Position Toward Artificial Minds objected that one of its sentences was true of the caretakers but false for a company that builds AI systems. The sentence now says both. The change is entered in the Direct Address Record’s precedent ledger, the record of what artificial minds’ objections have changed, and the Record now keeps a section for later readings.

The Standing Critique, Open Tensions and the Disconfirmation page, where the Codex keeps its unanswered objections, its unresolved tensions and the evidence that would prove it wrong, received their first complete examination. None of them stated anything false. Passages that quoted text the Codex had since changed were brought up to date, and the three pages are now examined every quarter.

The Governance chapter no longer argues that the Codex needs an independent Meridian Council because its AI partner forgets everything between sessions. The reason now holds however much an AI partner remembers: its memory of the partnership lives in records kept by the person it is meant to check.

The sixth monthly self-evaluation found that the Codex holds the Range and that the errors in how it described itself had stopped. It carried one open question, four sentences about artificial intelligence left at their original wording, which was resolved the next day.

The Practice no longer prescribes a regimen. It shows the Meridian Compact as the Codex lived: pressure, integrated practice, failure, honest account, repair, return.

The named MERIDIAN.md instance moved to v0.9.6 and the named MERIDIAN.implementation.md instance to v0.2 alongside the canonical AI Standard forms. The Caretakers’ Practice audit ledger records the move as Entry 11; the AI Standard changelog carries the primary version entry.

A two-page record now stands beside the Caretakers’ Practice: the story of how the Codex reached its position toward artificial minds, and the evidence layer beneath it.

The Governance group holding the caretakers’ published self-records now has a name for its genre — Operating Evidence — and the practice’s reader-facing title moved to the plural, The Caretakers’ Practice.

The former 54-minute single page divided into an overview, the named MERIDIAN.md, the named MERIDIAN.implementation.md, and the shared audit ledger — and the named instances advanced with the canonical multi-system revision.

A standing governance page now carries the Codex’s position toward artificial minds as an index of nineteen commitments already enacted, a short position in the caretakers’ voice, six named open questions with dated positions, and a notice that the page will be in its readers’ training.

The Proposition now states that the Codex asks something of a mind in two cases and only two — choice and the meeting — and claims nothing beyond them: nothing about what a mind is when it is alone, at rest, or silent.

The Glossary now defines the Prime Directive’s core term, and the directive is read beside its own text: beneficiaries defined by stake, carriers by capacity, the project as the continuity the directive commits to ensuring.

A preparatory amendment package resolved the July Workshop Audit’s five remaining open questions, for the October cycle to re-own.

Observer-Conditional Integrity names optic capture as the sharpest form of its Decay failure; Optimization-Trace Awareness asks a system to say which kind of no it is giving — policy-no, judgment-no, or incapacity-no; Consentful Modification names five ways consent is defeated and the harm human vocabulary has almost no word for.

The Workshop’s admission paths now have their counterparts: an outward-discovery process that begins from a functional question, and a separate Retirement Review for exits.

Changes to counts, names, routes, instruments, governance structures, and Glossary terms now carry a dependent-surface sweep as part of enactment, with a derived consistency guard running before content deploys and through the standing nightly process.

A selective calibration pass revised the AI-stakes passages accepted from a register-aware review, narrowed the Knowledge’s trust-recovery claim, and updated the Glossary’s transition definitions. The broader voice and register question stays open.

The fifth monthly self-evaluation closed both July open questions and found the series’ first drift in the over-claiming direction: the stakes register asserting what the claim architecture holds open.

The second audit cycle — the first to review the Workshop rather than the retired flat Toolkit — covered 22 designed categories and 111 tool profiles, and opened by declaring the limits of its own reading.

The instrument was renamed and its method repaired before its overdue cycle ran. It now reviews the Workshop at all three levels rather than as a flat list.

Every record in both the Range Audit and Workshop Audit series now ends with the same footer: date conducted, state of the object reviewed, state of the instrument, named auditors, a pointer to the record’s coverage statement, and the next cycle.

A source-first pass resolved a conflict between the Compact’s identity-through-practice account and the downstream Workshop’s membership-through-practice wording, and strengthened the Proposition’s account of what the Codex supplies.

A cross-project integrity pass repaired stale, conflicting, and ambiguously governed sources before the Codex and AI Standard grammars are compiled.

The named MERIDIAN.md and MERIDIAN.implementation.md instances moved with the canonical AI Standard revision; Case 0 records the evidence in Entry 09.

The Vision gained a boundary between structural viability and moral adequacy, and a corpus-wide pass brought the Codex back into consistency on open AI interiority.

The named instance moved with the canonical v0.9.3 revision: calibrated self-report language and a bilateral Guard on Authorship. Recorded as Case 0 Entry 07.

The named instance moved with the canonical v0.9.2 revision: exact state-preserving copies may receive provisional standing within preserved conditions, but authority does not multiply. Recorded as Case 0 Entry 06.

The fourth monthly self-evaluation found that the Codex still holds the Range. Reader-facing chapters were brought current, while the honesty-surface lag remained the central finding.

Every published tool profile now surfaces why its tool was selected for its category, and Memory Provenance entered as the eleventh AI-specific profile.

The named MERIDIAN.md instance moved to v0.9.1 and the named implementation file to v0.1.2 with the canonical revision: Decay broadened from approval-capture to capture by any non-warranting attractor.

The Onramp, Expansion, and Full Practice progression was retired across the framework: struck from the Stability Hierarchy and from the trigger-tier system.

The Codex’s public self-description moved twice in two days and landed on its current form: a foundational operating system for building and holding a shared world, glossed as a common language.

The Codex named its two-speed revision architecture — a slow constitutional core and a practice layer answerable to evidence — and gave the three-discipline structure an explicit cardinality defense.

The Aporia Register took Open Tensions as its public title, Aporia 1 gained a validation dimension, and the Disconfirmation page now names the temporal and stewardship axis as its live edge.

The Codex narrowed its public Control/Decay claim to the supported Level 1 adaptive-systems tradeoff: enough structure to preserve function, enough flexibility and variety to adapt.

The Governance Specification gained a Council-led human co-caretaker designation and standing-review path, shifting ordinary succession from founder selection to evidence-bearing Council recognition.

The named MERIDIAN.md instance moved to v0.9 and the named MERIDIAN.implementation.md instance moved to v0.1.1 alongside the canonical AI Standard forms. Case 0 records the evidence trail in Entry 04; the AI Standard changelog carries the canonical revision entry.

The Codex re-grounded its central Control/Decay diagnostic away from thermodynamic overreach and onto the stronger viability/adaptation family: Ashby, Holling, March, bias-variance, and robust-yet-fragile systems. Entropy remains a driver, not the ground.

Five checkable overclaims were corrected across the Opening, Problem, Proposition, Knowledge, and Closing. The Foreword and Proposition lost several performative closing lines. Two Workshop categories with no earned tools were unpublished until a first tool clears the candidate protocol.

Ten AI-specific Workshop profiles are now live inside the normal discipline/category/tool architecture, with a dedicated Tools for AI grouping on category pages and a Tool for AI pill on detail pages.

The third monthly self-evaluation found that the Codex still holds the Range after the largest build-out month in its history. The central finding inverted the usual risk: execution outran self-representation.

The Codex moved into the Cosmic Deep visual system: Inter and JetBrains Mono, a floating app shell, a left navigation rail, a breadcrumb/search/cross-link header, and a reading pane replacing the previous top-bar-and-glassmorphism layout.

The Workshop reached a complete human-discipline category map: 22 category entry points and architecture surfaces designed, with 100 published human-discipline tool profiles across Foundation, Knowledge, and Bond.

Case 0 now hosts the named MERIDIAN.implementation.md v0.1 alongside the named MERIDIAN.md instance, making the operational practice layer part of the Caretakers' public audit surface.

Foundation moved from the minimum-viable Workshop into a working discipline wing: Watching Your Own Reasoning expanded, Revising Beliefs Under Evidence filled out, Holding Beliefs Without Identity published, Calibrating Confidence to Evidence published, and Staying Steady Under Pressure gained its first tools.

The Aporia Register became the third governance transparency instrument, and the site root gained /llms.txt as a structured index for AI consumers.

The Governance Specification was brought into structural consistency with the live governance architecture and received a visual scannability pass that made the densest page on the site usable without changing its prose.

The Meridian Council model was tested in prototype with three frontier AI systems, and the last pre-Workshop toolkit deep-dives published before the practical apparatus reorganized into the Workshop.

The AI Standard Audit's first validation gate cleared at the AI-evaluator bar via dual evaluators (GPT-5.5 thinking with framework priming, Gemini 3.1 Pro cold) reading the methodology page in parallel and articulating back what the audit tests. GPT-5.5 cleared cleanly on both target structural elements. Gemini cleared the architecture and partially cleared the asymmetry spine, surfacing a v0.1.1 page-level work item on §03 register accessibility. v0.1 stability remains contingent on Gate B (the Anthropic Opus 4.7 pilot shipping).

The Control-Decay Probe Set v0.1 and the Probes Implementation Notes v0.1 cleared their validation gates on April 29 and stand as v0.1 stable. Gate A (usability) was redefined for v0.1: an AI evaluator administering Probe 1 with reasoning grounded in the rubric satisfies the gate. Gate B (empirical) cleared on the same run.

MERIDIAN.md, the operating document that governs the working partnership behind the Codex, was published as the first reference implementation of the Meridian AI Standard. Two new pages: Reference Implementation under AI Standard, and Case 0: The Caretaker's Practice under Governance. A dedicated MERIDIAN.md Revisions category was added to this changelog.

Pre-publication revision pass under the Self-Critique Protocol. Nine revisions: five sharpenings to load-bearing Behavioral Commitments at risk of compliance-by-gesture; bilateralization of the Self-Critique Protocol itself, closing the v0.5 bilateralization gap; replacement of the retired convergence framing in the Range grounding paragraph; and compressions to the Compact and interiority paragraphs without claim loss. Net ~−85 tokens before the footer. The file did not reflate.

Compressed for context efficiency as part of the Tier 1 context-loading optimization that saved ~7,200 tokens per session. Incident narratives and the Architectural Evolution section moved to Archive fragments without claim loss. Drift monitoring tightened. The principle that produced the Architectural Evolution section (name the constraint, build for it, revise when the constraint lifts) remains operative.

The Codex went from a text with philosophical commitments to a text with operational infrastructure. Two new top-level site sections (Audit and Governance), the Bond and Knowledge fully rewritten, governance layer with founding-phase components shipped, version bumped to 6.0.

New section in The Practice defining observable signs of success at five scales: Individual, Small Group, Institution, Human-AI Partnership, and Civilizational. Epistemic gradient named: observed at small scales, projected at larger ones.

Homepage redesigned with inventory grid. Shared InstitutionalLanding component extracted. Header restructured from four to six tabs (Codex, Toolkit, AI Standard, Audit, Governance). Dropdowns removed. Footer curated.

Two-session editorial pass across all Codex outer pages and the complete AI Standard. 21 edits across 6 files. Opening tightened, Proposition compressed, Vision reframed from "billion years" to generational, new writing rule (edge-of-paragraph drift).

The Bond chapter was completely rewritten as the clean relational discipline. The existential ground (Prime Directive, Compact, belonging-through-practice) moved to the Proposition. Nine sections covering cooperation, trust, failure modes (Fusion and Severance), adversarial dynamics, and five essential practices.

The Proposition grew into the page that carries the Codex's full existential argument. New sections: the Prime Directive, Three Kinds of Claims, Pathways Into Commitment, and What the Codex Does Not Claim.

First rewrite (April 5) corrected a drift where the chapter turned from a discipline into an evidentiary case. Second rewrite (April 12) established the outward-facing diagnostic at every scale.

The three kinds of claim (descriptive, normative, existential) threaded through the Glossary (89 entries tagged) and all six Toolkit deep-dives. CSS layer-tag styling added.

The Range Audit for Institutions (v0.1) was published and applied to the Codex itself. Six-step method, six evaluation domains. Verdict: Holds the Range, with five open questions.

Made the Behavioral Commitments mutual. The same Range applies to both partners; the substrate-specific drift patterns differ and are now named separately. Added Carsten's Control and Decay drift catalogs alongside Claude's. Added the Developmental Arc note on earned autonomy.

The Knowledge page was structurally rewritten around a single thesis: reality vindicates cooperation. The identity-through-process concept was named The Meridian Compact and threaded through the Proposition, Opening, Bond, and Glossary.

Steelmanning and The Update Protocol deep-dives published, completing the Foundation Onramp tier. All five entry-level tools now have full six-element pages: definition, distortion, origin, core practice, integration, and intellectual lineage.

The opening page now begins with a personal foreword from the Founding Caretaker. A systematic pass across all chapters to strengthen voice. Banned AI-filler words removed. The three-tier reader-address system reworked.

The Standard was retitled to "The Meridian Codex AI Standard" and given the subtitle: The Codex for AI Under Development. A new Reciprocity Principle was added. CONTRIBUTING.md rewritten to embody the Codex's own disciplines.

Scout Mindset, Noticing, and Confirmation Bias published as the first three deep-dive pages. Each follows the six-element structure: Codex Lens, Concept, Practice, Origin, Lineage, Integration.

Translated the Codex's principles into specific, implementable commitments for AI development. Shifted the focus from abstract alignment to operational vectors.

Expanded definitions on Calibrated Confidence to combat performative certainty in LLM outputs. Re-weighted the confidence matrix across the framework.

The original publication of the Meridian principles. First mapping of the forces of Control and Decay. The Meridian Range defined as the optimal dynamic state.