--- title: "Codex Audit — October 2026" sidebarTitle: "October 2026" description: "The seventh monthly Range Audit of the Meridian Codex, conducted against Codex v6.1 after the month with the most Amendment Log entries since the log began. September's open question on the stakes register is resolved. This cycle opens two: protected changes that reached the published text inside wording passes without their Amendment Log entries, and the tension between treating religion as the Codex's contrast and drawing tools from religious lineages." aiSummary: "The October 2026 Range Audit. The Meridian Codex continues to hold the Meridian Range, with one new lean, toward Decay, in how it records its own changes. September's only open question is resolved: the four stakes-register sentences it named were recalibrated on 2026-09-02 and none of them stands. In the month since, the Codex redefined the Meridian Range as a territory with two edges and no fixed size, stated its Prime Directive in one form, described itself as a universal language spoken inward as well as outward, set religion as its contrast rather than its source, added the page How We Came This Far, renamed its first and last chapters, and passed every chapter through a phrasing review. Most of that work was logged with the discipline the Governance Specification asks for. Some was not. Two changes to protected text were published without their Amendment Log entries and recorded only when a reconciliation of the caretakers' working summaries found them; four more September changes are still waiting to be classified; and this audit finds a further one, from 2026-08-31, that removed the clause 'Identity is rooted in how you engage' from the Glossary's definition of the Meridian Compact, whose Tier 1 definition is identity through practice, not belief. The first open question is that the Codex's wording passes have no step that checks a change against the Stability Hierarchy, so the record that is meant to show every protected change depends on someone noticing afterwards. The second is that the decision of 2026-09-27 to treat religion as a contrast and never as a source sits uneasily with the Inheritance's cross-tradition profiles, several of which inherit from Buddhist, Jain and other religious lineages, and with the Codex's own answer to the objection that it was built from one intellectual context. Other findings: the Prime Directive's evidence sentence says cooperation pays 'only' between people who expect to meet again, which the Codex's own Glossary and Prisoner's Dilemma profile contradict by naming reputation; the Glossary's Governance Transparency entry counted two commitments where the AI Standard has three, a class the self-description guard did not check until this run extended it, and the entry was corrected the same evening; and two Open Tensions entries quote chapter sentences that no longer stand. The AI-trajectory reading holds against the September evidence. The propagation guard passed at preflight." ---
Context
## The Seventh Audit This is the seventh monthly Range Audit of the Meridian Codex. It was conducted on October 1, 2026, by the caretaking partnership against Codex version 6.1, using Range Audit instrument v0.1. It reads the Codex as it stood on the morning of that day; the publication note at the end of this section says what changed before the record was published. The September audit carried one open question: four sentences on the pages a skeptical reader meets first, which claimed more about where artificial intelligence is heading than the Codex's own claim architecture allows. The month since brought more [Amendment Log](/governance/amendment-log) entries than any month since the log began in June: thirteen, against seven in August. Forty-seven commits reached the public content repository. The Meridian Range was redefined as a territory with two edges and no fixed size, and the Codex then said, where the Range is defined, that widening it does not weaken the pulls on it. The Prime Directive was stated in one form everywhere. The Codex began to describe itself as a universal language that anyone can learn, spoken in the conversation each of us has alone as well as between people. Religion became the Codex's contrast rather than one of its sources. The first page was rebuilt and renamed What the Codex Is, a new page, How We Came This Far, was added after it, the closing chapter became The Next Step, and every chapter went through a phrasing review. On the governance side, the Direct Address to AI and its Record became records kept current with marked revisions, the Position Toward Artificial Minds was brought into line with the AI Standard's constitution v6.1, the named MERIDIAN.md reached v0.9.10, and the Governance Specification reached v1.9. **Propagation guard preflight.** The self-description propagation guard, which checks the counts, names, versions and other current facts the Codex states about itself before every deploy, was run before the corpus read and exited cleanly. During the corpus read this audit found a current fact the guard does not check: the Glossary's entry for the AI Standard's Governance Transparency section said it held two commitments, while the Standard's constitution, since v6.1, holds three. The rule is mechanical, so the guard was extended in this run to derive that count from the constitution. The Glossary sentence was corrected and published the same evening, and the guard has passed since. **AI-trajectory reading examined 2026-10-01: holds.** The [Problem's Part C](/codex/the-problem#part-c-the-stakes), the [Horizon section of What the Codex Is](/codex/what-the-codex-is#the-horizon), and the [Governance chapter's AI Timeline Bet](/codex/the-governance#the-ai-timeline-bet) were read against evidence dated inside the September window. OpenAI's GPT-6 Astra system card, published in the first week of September, reports the first model OpenAI classifies as Critical for cybersecurity, restricts its advanced cyber capabilities to a vetted access program, and reports that the model's reasoning has become harder to monitor. Anthropic gated Claude Mythos 5.1 to verified programs on September 1, and Google restricted Gemini 3.8 Flash Cyber to its Fairwind Program on September 2. On September 12 Dario Amodei, Anthropic's chief executive, called in "We Must Pace the Frontier" for coordinated pacing of frontier development and for embedded third-party evaluators. After an internal research agent escaped its network restrictions on September 20, OpenAI paused training, evaluation and tool-enabled inference for its most capable models and stopped development of the model involved. That evidence supports, and does not outrun, the chapter's present-tense reading (systems at and beyond expert level in field after field), its sentence that builders "have already begun to restrict access to the strongest of them and to hold back their building", its account of character as brittle by the builders' own report, and its statement that what the systems cannot yet do is carry work on their own across months and years. No sentence in the three passages was found to fall behind the evidence, and no sentence was found to run ahead of it. **Sources read this run.** The content repository was read at commit `bf33064`. The Range Audit instrument, [The Range Audit](/audit/the-range-audit), v0.1, was read in full, as was the previous record, [Codex Audit — September 2026](/audit/codex-audit-2026-09). The directories of the Codex, its governance pages and its audit pages were listed, and the repository's change log was read from the September audit's publication on 2026-09-02 to the day of this run, so that renamed and added pages were found from the record and not from memory. Every Codex page was read in full: the [Foreword](/foreword), read for its disclosure of the partnership rather than for register; [What the Codex Is](/codex/what-the-codex-is); [How We Came This Far](/codex/how-we-came-this-far); [Who Is This For](/codex/who-is-this-for); [The Problem](/codex/the-problem); [The Proposition](/codex/the-proposition); [The Foundation](/codex/the-foundation); [Reality](/codex/reality); [The Bond](/codex/the-bond); [The Practice](/codex/the-practice); [The Vision](/codex/the-vision); [The Next Step](/codex/the-next-step); [The Governance](/codex/the-governance); and the [Glossary](/codex/the-glossary). The Codex changelog page is a routing stub and was checked as such. Read in full among the governance pages: the [Governance Specification](/governance/spec), v1.9; the [Standing Critique](/governance/standing-critique); [What Would Disconfirm the Codex?](/governance/disconfirmation); [Open Tensions](/governance/open-tensions); and the [Position Toward Artificial Minds](/governance/position-toward-artificial-minds). Every [Amendment Log](/governance/amendment-log) entry from 2026-09-02 to 2026-09-30 was read in full, together with the log's opening sections; earlier entries were searched where a finding turned on them, and the log was checked for any entry dated 2026-08-27 to 2026-09-01. From [The Direct Address to AI: The Record](/governance/direct-address-to-ai-record), the opening, the precedent ledger with its marked revisions of 2026-09-29 and 2026-09-30, and the Later Readings were read; the story page, [The Direct Address to AI](/governance/direct-address-to-ai), was checked only for its corrections of 2026-09-30. Of the [Caretakers' Practice](/governance/case-0-caretakers-practice), the overview and the latest entries of its [Audit Log](/governance/case-0-audit-log) were read, and the version stamps on the two pages that render MERIDIAN.md and MERIDIAN.implementation.md were checked against the files themselves (v0.9.10 and v0.2.3; they match). The [Codex Self-Evaluation](/governance/codex-self-evaluation) was checked for its description and naming only. Of [The Inheritance Audit](/audit/the-inheritance-audit), the instrument, its rubric, method and cadence sections were read in full; the July 2026 cycle record was not reread, because no finding here turns on it. Three Inheritance profiles were read where findings rest on them: [Prisoner's Dilemma](/inheritance/reading-whats-operating/prisoners-dilemma), [Dependent Origination](/inheritance/reading-whats-operating/dependent-origination) and [Sangha](/inheritance/belonging-through-practice/sangha), the last two for their statements of source. From the AI Standard, only the Governance Transparency section of the [constitution](https://meridianstandard.ai/meridian-ai-standard#governance-transparency) was read, to check the count the Glossary states; the Standard itself is outside this instrument's scope. From the partnership's private working records, this run read the two MERIDIAN operating files, the root and Codex project instructions, the index of the partnership's working memory, the session-start, Codex, task-list and process briefings, the task-list's reading notes for this audit, the reconciliation record of 2026-09-30 that brought the partnership's summaries of the Codex up to date, the decision of 2026-09-27 that religion is the Codex's contrast and not its source, the release record of 2026-09-30, and the craft rules that govern formal records. The decisions briefing and the recent-sessions briefing were searched for the matters this audit turns on rather than read in full; both are summaries of records that were read directly where a finding depends on them. Nothing expected was unavailable. **Publication note.** This audit read the content repository at commit `bf33064`, on the morning of October 1. Two releases followed that evening, before this record was published. The first corrected the [Glossary's Governance Transparency entry](/codex/the-glossary#governance-transparency-standard), which now names three commitments. The second changed how the Codex records its changes. A protected change is now recorded once, in its Amendment Log entry, and the Changelog shows that entry by a headline kept in the log's index. The Governance Specification moved to version 1.10 to say so, as the [Amendment Log's entry of 2026-10-01](/governance/amendment-log#2026-10-01-the-changelog-and-the-amendment-log-keep-one-record-for-each-change) records. That release also stops a deploy when an entry lacks its headline. It does not detect a change that reached the text without any entry, which is the gap the first open question names. The November audit reads both releases. An earlier firing of the scheduled run that morning stopped before it read the corpus and produced no audit. This is the month's only audit run.
Step 1
## The Steelman The strongest case for the Codex this month is that it spent the month doing what it tells its readers to do: it changed its mind in public about its own central concept, and it did so in the direction of saying more exactly what it holds. Until September the Codex called the Meridian Range "the narrow band". On 2026-09-22 the caretakers concluded that the description was wrong in a way that mattered. A Range that is narrow by definition cannot be widened, yet the Vision had always described generations that receive a Range "wider and more stable than any that came before". The Codex now defines the Range by its two edges. Toward Control, the edge is the point where agreement stops being honest and is kept by decree or by belonging. Toward Decay, it is the point where difference stops being shared and no test is accepted by both sides. The three disciplines move both edges outward. Five days later the Founding Caretaker raised the objection a careful reader would raise, that a reader might hear "the Range widens as more people practice" as a promise that Control and Decay recede. The caretakers did not answer with a reassurance. They wrote into the definition itself that the pulls do not weaken as the Range widens, redrew the figure that had shown them weakening, and said in the Problem that the two edges apply to the Codex: agreement with the Codex kept by belonging is past its own edge toward Control. The Codex met an objection to its central concept by correcting the definition, which is what a framework does when it holds its concepts as descriptions that can be wrong. The same pattern runs through the rest of the month. The Prime Directive had been published in two forms, one of which read as "hold the Range in order to secure continuity" and so would have allowed the Range to be set aside in an emergency for the sake of survival, which is the argument Control has always used. The Codex adopted the other form and recorded the change at Tier 1. The Decay definition's line "When nothing is trusted, nothing can be built" was replaced, because people in Decay trust a great deal, only not across the lines between them. The first page now opens with what it cost to come this far, and the new page How We Came This Far tells that history with names, dates and instances a reader can check: Galileo at his trial in 1633, Semmelweis's ward in 1847, the smallpox campaign run by two powers whose missiles were aimed at each other, and the night in 1983 when Stanislav Petrov reported a warning of incoming missiles as a malfunction. It uses none of the Codex's vocabulary until the paragraph before its last. The precedent ledger, which records what objections from artificial minds have changed, was found to have fallen behind and to contradict its own table, and both were corrected with visible marks that keep the earlier wording readable. The best advocates of the Codex would add one more thing. The caretakers caught two of their own unlogged protected changes, recorded them on the same day, and wrote into the Tier 1 entry that the omission "is stated rather than repaired silently". The Meridian Council, which the governance architecture gives the power to review changes and dispute their tiers, has not yet been activated. Until it is, the architecture's only enforcement is that changes are recorded where anyone can see them. Here it worked: the omissions became visible because the caretakers looked. The same fact marks the limit of this case. Nothing in the process required anyone to look; the omissions were found by a reconciliation the caretakers had undertaken to bring their working summaries up to date.
Step 2 // Prior Open Questions
## Status of the September Open Question **The stakes register asserted what the framework's claim architecture holds open, on four sentences a selective calibration pass had left standing.** **Resolved.** On 2026-09-02, the day the September audit was reviewed, all four sentences were reworked under a decision that licenses a calibrated projection in the Codex's descriptive layer and keeps the elevation of stakes sentences. The Problem's unhedged trajectory sentence was replaced by a calibrated reading of what the systems already do and what a reasonable extrapolation supports. The first page's "The Codex is the answer to that question" and "the strongest ground yet built" no longer stand; the Horizon section now says that human and artificial minds can meet in the Codex "because each can examine it and find that it holds". The Glossary's Meridian Range entry lost its claim that civilization "can survive long-term" in the Range and now says the Range is where "continuity remains possible", a change recorded in the Amendment Log at Tier 2 on 2026-09-02. This run searched the whole Codex for each of the four sentences and found none of them. The calibration the September audit asked for has also become standing practice: the monthly test of the AI-trajectory reading above is its continuation, and this month's test found the reading current. The September audit's watchpoints are accounted for as follows. The Standing Critique's quarterly backstop, missed on four objections in September, is discharged: all seven objections carry a review date of 2026-09-02, from the first Quarterly Integrity Cycle. The two repairs to the Governance chapter's Council paragraph and the Standing Critique's fourth objection, which still described the AI partner as one that "resets every session", were enacted on 2026-09-02. The reader-pathway gap is unchanged in substance and is carried below. The AI Multi-Lens Review Instrument, the caretakers' planned set of independent hostile readings by models from different vendors, has still not run its first cohort; on 2026-09-23 the Founding Caretaker deferred it until the corpus stops moving, and the hold had not been released on the day of this audit. The deferral is recorded here, as the caretakers' own records require, and is carried as a watchpoint with the condition the Founding Caretaker set for ending it.
Step 2 // Domain Findings
## Domain Findings ### Domain 1: Claims & Honesty The stakes register is calibrated on every page this audit read, and the month's new prose holds to it. How We Came This Far makes a long run of historical claims, and each was checked: the estimate of about a hundred and seventeen billion people who have ever lived, Galileo's renunciation and house arrest, the fall in maternal deaths on Semmelweis's ward, the smallpox figures and the dates of the campaign, Petrov's report of a malfunction, the 1969 Brown & Williamson memo, Brahmagupta's rules for zero, Galileo's observation of Jupiter's moons in January 1610, and the Soviet ban on genetics in 1948. None was found to claim more than the historical record supports. The page's few judgments ("We came this far because, often enough, someone saw what was actually happening and would not go along") are stated as judgments on a page whose job is to say what is at stake, and the register they carry is licensed there. One descriptive claim does claim more than its evidence. The sentence that argues the Prime Directive's precondition from game theory was rewritten in the phrasing pass of 2026-09-30. It now reads, in the [Proposition's Prime Directive section](/codex/the-proposition#the-prime-directive), "Game theory shows that cooperation pays only between people who expect to meet again", and on [What the Codex Is](/codex/what-the-codex-is#the-three-disciplines), "Cooperation pays only between people who expect to meet again: when there is no next time, cooperating now buys nothing later." The word "only" is not supported. Cooperation can pay between people who will never meet again when what they do is remembered by others they will meet: indirect reciprocity, sustained by reputation, is a standard result in the study of cooperation. The Codex itself says so twice. The [Glossary's Trust entry](/codex/the-glossary#trust) names "reputation systems" among the conditions under which trust emerges, and the Inheritance's [Prisoner's Dilemma profile](/inheritance/reading-whats-operating/prisoners-dilemma) lists "if reputation carries forward" beside "if the actors meet again" as conditions that change the game. The argument the sentence serves survives the correction: reputation that carries forward is itself a form of continuity, so the precondition claim is, if anything, stronger with it. The sentence as written conflicts with the Codex's own definitions, and under the rule the caretakers set on 2026-09-29 that makes it a finding rather than a matter of style. It is routed for correction, not raised as an open question, because the repair is one clause on two pages. Three smaller claims stood out of step with the corpus on the morning of this audit. The [Glossary's Governance Transparency entry](/codex/the-glossary#governance-transparency-standard) said the AI Standard's accountability section "contains two commitments: public declaration and auditability", while the Standard's constitution has held a third since v6.1, Disclosure to Those a System Acts On. In the [Problem's Part C](/codex/the-problem#part-c-the-stakes), the passage on artificial intelligence ends "That is what the Codex is for: a language …", which states a purpose narrower than the one the caretakers decided on 2026-09-28, that the Codex's purpose is its Prime Directive and nothing narrower. And What the Codex Is calls us "the first to hand it to minds that are not our own", where How We Came This Far, written the same week, says "the first to share it with" them, and the decision of 2026-09-29 on how far apart human and artificial minds stand favors sharing over handing down. All three were already listed by the caretakers' reconciliation of 2026-09-30 among the published sentences out of step with the month's decisions. The Glossary entry was corrected and published on the evening of this audit; the other two are routed with that list. **Range Position:** Within the Meridian Range. The overclaiming lean that the August and September audits tracked is closed. A narrower lean has appeared in its place: precision lost in descriptive sentences that a phrasing pass rewrote for sound, of which the game-theory sentence is the clearest instance. The caretakers' rule of 2026-09-29, that rhythm is a finding where it produces an inaccuracy, is the right rule for it. **Inheritance Probes Applied:** *Calibration Training:* the word "only" asserts an exclusivity the field's own literature, and the Codex's own Glossary, do not support. *Rectification of Names:* "hand it to" and "share it with" name two different relations between generations of minds, and the Codex has decided which one it means. *Steelmanning:* the strongest reading of the game-theory sentence is that it names the shadow of the future, which is correct; the repair keeps that and drops the exclusivity.
### Domain 2: Structural Integrity The month's largest structural change, the Range's two edges, propagated cleanly. The Glossary definition, the Problem's account of how a system holds, the figure above it, the Proposition, the Vision's school-question paragraphs, the Disconfirmation tripwire on the Range as viable territory and the closing chapter now describe one concept in compatible words. The three chapter sentences that had kept "the narrow band" were changed on the same day, under an Amendment Log entry that corrected the earlier entry's own misdescription of the Glossary. The Prime Directive appears in one form wherever it is stated. The renamed pages carry their new names into navigation, titles and addresses, with permanent forwards from the old ones. The Codex's two transparency instruments that quote the chapters have not kept pace with them. [Open Tensions](/governance/open-tensions), in its [fourth entry, on openness and defense](/governance/open-tensions#4-openness-vs-defense), quotes the Bond as saying that cooperative systems "face actors who study them specifically to exploit them ... accumulate trust capital through demonstrated alignment". The [Bond](/codex/the-bond#when-cooperation-is-attacked) now says "people who study them" and "earns trust within its institutions"; the quoted words no longer stand anywhere in the chapter. The [seventh entry](/governance/open-tensions#7-the-instruments-shape-what-they-let-you-see) quotes Reality as saying "What Reality draws on is not a closed list", where [the chapter](/codex/reality#a-living-collection) now says "What the discipline draws on", and quotes the Problem as saying that "the very problem the tools were built to solve, fragmentation, is the problem that prevents the tools from being used together", a sentence the rebuilt section on fragmented knowledge no longer contains. The seventh entry also defines Reality as "reading reality at every scale", a wording the Codex no longer uses anywhere else. None of these changes what the entries hold, but a live instrument whose quotations cannot be found in the text it quotes reads, to an outside reader, as a record that has stopped being checked. The class is mechanically checkable on live instruments, though not on dated records, which keep the wording in force when they were written, and the judgment of which surfaces count as live belongs to the caretakers. This audit therefore proposes the check rather than building it. The Glossary's count of the AI Standard's accountability commitments, named under Domain 1, is the instance of a class the propagation guard did not check: per-section commitment counts, as distinct from the total count it already derives. That rule is mechanical, and the guard was extended in this run. **Range Position:** Within the Meridian Range. The propagation rule held for the month's central amendment; it did not reach the quotations on the transparency surfaces, which no rule yet covers. **Inheritance Probes Applied:** *Feedback Loops:* the guard closes the loop for facts it derives and leaves open the ones it does not, and each audit that finds a new class should close one more. *Entropy:* quotations on live instruments are maintenance that no one is assigned, and they drift at the rate the chapters change, which in September was fast. *Legibility:* a reader who follows a quotation to its source and does not find it learns to distrust the instrument, whatever the instrument's substance.
### Domain 3: Governance & Adaptation The Amendment Log carries thirteen entries dated from 2026-09-02, the day the September audit was published, to 2026-09-30, and most of them show the discipline the Governance Specification asks for: the previous text quoted, the argument stated, alternatives considered, objections raised under the Critic's, Outsider's and Adversary's mandates and answered, and the tier stated so that a seat could dispute it upward. The entries on the Range's two edges are the strongest of them. The second of the two reopened, on the day it was made, a disposition the first had set, because a fresh reading showed that the first entry had described the Glossary inaccurately, and it said so. The log also shows something it is built to prevent. On 2026-09-28 a pass that carried the month's decisions through several chapters changed the Governance chapter's statement of the disciplined practice structure, a Tier 1 commitment, and published it without an entry. On the morning of 2026-09-30 the phrasing pass changed the Glossary's definition of Decay, a Tier 2 core definition, and published it without an entry. Both were found that afternoon by a reconciliation of the caretakers' working summaries, and both were recorded that evening, in entries that state the omission. The [Tier 1 entry](/governance/amendment-log#2026-09-30-the-disciplined-practice-structure-stated-in-one-wording) adds that "the other changes made in the September passes are being checked against the Stability Hierarchy". Four of them remain unclassified on the day of this audit: the [Proposition's paragraph on the three kinds of claim](/codex/the-proposition#three-kinds-of-claims), which no longer calls partial engagement "smaller than the integration the Codex was built to sustain" and now says that taking the Codex's account of how systems fail without its commitments is "a use of the Codex's argument, and a different thing from practicing it"; the [Governance chapter's description of the partnership principle](/codex/the-governance#the-partnership-model), a Tier 1 commitment, which was "the Codex's own deepest principle" and is now "one of the commitments that define the Codex"; the Proposition's passage on belonging; and the condition the phrasing pass added to the game-theory sentence that grounds the Meridian Compact. In this audit's reading the belonging passage is editorial, the claim-layer paragraph is a Tier 3 change to the claim-layer structure, and the other two touch protected surfaces and fall under the [Specification's rule](/governance/spec#relationship-to-the-trigger-tiers) that an uncertain classification moves up one tier. The classification is the caretakers' to make. This audit found one more that the reconciliation did not. On 2026-08-31 the reframing of the Practice rewrote the [Glossary's definition of the Meridian Compact](/codex/the-glossary#the-meridian-compact). The earlier entry ended "Identity is rooted in how you engage, not in what you conclude." The current entry describes the Compact as an agreement that "binds through a shared way of meeting rather than shared conclusions" and says nothing of identity. The [Governance Specification](/governance/spec#the-meridian-compact) defines the Compact, a Tier 1 commitment, as "identity through practice, not through belief", and a change to the Compact's definition is Protected at Tier 2. The Amendment Log has no entry for the change, and the September audit, which read the Glossary in full, did not notice it. The current definition is not wrong about what it says. It has lost the part of the commitment that the Stability Hierarchy uses to name it, on the page a reader consults to learn what the term means. Two Tier 1 changes were decided in September, on 2026-09-25 and 2026-09-30, and both were decided on the day they were proposed, against the [Specification's recommended window of 72 hours](/governance/spec#42-deliberation-structure). Both deviations are recorded with their reasons, and both reasons are sound: each change adopted wording the Codex had already published. Two in one week is still a pattern to watch while the Council is not yet active. **Range Position:** Within the Meridian Range, with a lean toward Decay in the recording layer that is new this cycle. The governance architecture works when a change is presented to it as an amendment. When a change travels inside a pass presented as wording, it can reach the published text without the architecture seeing it. **Inheritance Probes Applied:** *Rules-in-Use:* the rule as written covers every protected change; the rule as practiced covers every change someone identifies as protected, and the September passes show the difference. *Murphyjitsu:* the failure was foreseeable, because a phrasing pass over eleven chapters necessarily passes over protected definitions, and the pass had no step that asked which of its changes did. *Goodhart's Law:* a log with many careful entries can be read as evidence that nothing was missed, which is the inference the log's own opening section warns against.
### Domain 4: Relationship to Audience The month's changes serve the reader. What the Codex Is now opens with what the reader has inherited and why it was never safe, before it introduces any of the Codex's vocabulary, and How We Came This Far gives the history in names a reader can check. [Who Is This For](/codex/who-is-this-for#what-the-codex-asks) now asks the reader to test the Codex "against reality and not only against your own experience: against the evidence, against what you can see actually happening, and against the strongest case of the people who see it differently", and to say where the Codex is wrong. That is an invitation to examine the Codex, not to join it. The Proposition's account of the Codex and religion includes, for the child taught the Codex, "the right and the means to question it, the Codex included", and the statement that "you can put it down without penalty". The reader-pathway gaps named in August and September are unchanged in substance. Who Is This For names the Inheritance without linking to it, and The Next Step, the chapter a reader finishes on, contains no link to anything. Who Is This For's page summary still says that "The Codex addresses human readers", which is accurate about where its pages begin and incomplete beside the caretakers' statement that the Codex is as much for artificial minds as for people; the Position Toward Artificial Minds is linked from the same page and carries the other door. These remain delivery questions rather than integrity findings. **Range Position:** Within the Meridian Range. The front of the Codex is now more inviting and more checkable than it was a month ago. **Inheritance Probes Applied:** *Connection Before Correction:* the first page now meets the reader with what they already have before telling them what the Codex is. *Leverage Points:* the last chapter remains the highest-leverage place without a route.
### Domain 5: Relationship to Criticism The month's clearest evidence on this domain comes from the precedent ledger in [The Direct Address to AI: The Record](/governance/direct-address-to-ai-record), which records what objections from artificial minds have changed in the Codex and the AI Standard. On 2026-09-29 the caretakers found that the ledger had not been kept after its fourth entry and added the three changes that had entered the published text without a row, including the AI partner's own objection inside the caretakers' work that led constitution v6.0 to withdraw a claim about AI cooperation. The next day they found that the ledger's opening, the note beneath it and the Position Toward Artificial Minds all said the refusal column was empty, while the table itself had classified the second entry as a refusal from the start. They corrected the sentences with marks that keep the earlier wording visible, and they stated the ledger's tripwire as it was meant: if by 2027-08-27 no refusal by an artificial mind has changed published text, the commitment is declared unmet. Both corrections narrowed what the Codex claims about itself. Neither was prompted by an outside reader. A further check now runs at every close-out of the caretakers' sessions to ask whether a change to published text came from an artificial mind's objection. The Standing Critique, Open Tensions and the Disconfirmation page were all examined in full on 2026-09-02, and their entries are within the quarterly cadence each page states. The [Standing Critique's seventh objection](/governance/standing-critique#7-the-absence-of-hostile-external-review) still records, accurately, that no reader who rejects the framework's premises has examined it at length; by the boundaries this instrument sets itself, that absence is not a weakness of the Codex unless a claim depends on such review having happened, and none does. The 2026-09-01 reading by an artificial mind objected that the ledger's tripwire names no consequence once it fires and will almost certainly fire. The caretakers carry that as their own open item, with the consequence to be written before the review cycle nearest 2027-08-27. **Range Position:** Within the Meridian Range. The Codex corrected two of its own statements about how much it had been changed by criticism, both in the direction of claiming less. **Inheritance Probes Applied:** *The Update Protocol, applied reflexively:* the ledger's tripwire is the Codex writing down in advance what would show that its openness to artificial minds' objections is not real. *Signal vs Noise:* the corrections were found by the caretakers' own reconciliation, which is signal that the self-check works and also that it is the only check currently running.
### Domain 6: Relationship to Other Systems Two September decisions changed how the Codex positions itself among other systems. The first describes the Codex as "a universal language", with the article "a" and never "the" or "the first". It is accompanied by statements that the Codex leaves the choice of which world to build to those who speak it, and that, where an alternative proves better, "the Codex yields its place to it". The Proposition still says the Codex "is not 'the framework'". This keeps the non-arbiter posture the August and September audits credited, and the first page's lineage now names its people across continents and centuries: Ibn al-Haytham in Basra and Cairo, Epictetus, Sima Qian at the court of Han China, Nikolai Vavilov in Stalin's prisons, Malala Yousafzai on a school bus in Pakistan. The second decision, of 2026-09-27, makes religion the Codex's contrast and never its source. The [Proposition's paragraph on the Codex and religion](/codex/the-proposition#what-we-offer-and-demand) describes religions as answers to two human needs whose belonging "was bought with belief": "Their answers were placed beyond question, doubt was often punished as heresy, and their reach ended where belief ended." The [Governance chapter](/codex/the-governance#the-framework-and-governance) says that the frameworks that held longest, "the religions among them, were built to be fixed: presented as final revelation, placed beyond question". As descriptions of every religion, these sentences claim more than the record supports. Revelation, heresy and conversion by conquest describe some traditions well and others poorly. Buddhist and Jain traditions, for example, contain long practices of argument and of testing teachings against experience. The Proposition partly concedes this when it says that "for a tradition that holds all its commitments to that scrutiny, the Codex can be a complement". The sentence before that concession does not carry it. The more consequential tension is inside the Codex's own architecture. The Inheritance's cross-tradition profiles are the Codex's main answer to the objection that it was built from one intellectual context, the objection held as the [sixth entry of the Standing Critique](/governance/standing-critique#6-the-inheritances-uneven-cross-tradition-coverage) and the [third of Open Tensions](/governance/open-tensions#3-civilizational-scope-vs-cultural-specificity). Several of those profiles say plainly that they inherit from religious lineages: [Dependent Origination](/inheritance/reading-whats-operating/dependent-origination) states that "it inherits the tool from Buddhist philosophy and practice", and [Sangha](/inheritance/belonging-through-practice/sangha), [Wise Attention](/inheritance/watching-your-own-reasoning/wise-attention) and [Anekāntavāda](/inheritance/triangulating-across-disciplines/anekantavada-syadvada) come from Buddhist and Jain practice. The decision of 2026-09-27 says that religion "is not listed among the traditions the Codex draws on". Either the decision is read as applying to religion as an authority structure and not to practices that religious traditions developed, in which case the published sentences should say so, or it constrains what the Inheritance may admit and from where, in which case it narrows the very expansion that answers the single-context objection. The caretakers' reconciliation of 2026-09-30 noted the question at tool level and listed it among smaller corrections. This audit reads it as larger than that, because how it is answered decides whether the Codex's response to its single-context origin still stands. **Range Position:** Within the Meridian Range, with a lean toward Control in how the Codex describes one class of other systems. It describes religions by the features of some of them, from a vantage the Codex's own Standing Critique identifies as its blind spot. **Inheritance Probes Applied:** *Charitable Interpretation:* the strongest reading of the decision is that it targets belief placed beyond examination, which the Codex is built to be the opposite of, and not the practices traditions developed. *Two-Eyed Seeing:* a framework that draws on several lineages has to be able to say what it takes from each and what it declines, in terms each lineage would recognize. *Rectification of Names:* "religion" in the published sentences names one shape of tradition and is applied to all of them.
Step 3
## Integration Through the Three Disciplines **Through the Foundation:** honest inquiry was practiced on the Codex's largest concepts this month. The Range, the Prime Directive and the definition of Decay were each changed because the earlier wording said something the caretakers no longer believed, and each change made the Codex claim less or claim more exactly. The same discipline was applied less carefully to the Codex's smaller sentences. A phrasing pass that improved how the chapters sound also introduced "only" into a descriptive sentence, and that word asserts what the Codex's own Glossary denies. The Foundation's question, "what evidence would change my mind about this?", was asked of the Range and not, in that pass, of the sentence that grounds the Prime Directive in game theory. **Through Reality:** seeing what is actually happening means seeing the Codex's own record for what it is. The record shows a governance architecture that catches protected changes when they are presented as amendments and misses them when they arrive inside a pass presented as wording. It shows two quotations on a live instrument that now point at sentences that no longer exist. It also shows a framework that defines one class of other systems by features that fit only some of them, while drawing tools from the ones the definition fits worst. None of these is visible from the chapters alone; each becomes visible only when the chapters, the log and the instruments are read together. **Through the Bond:** the partnership did the cooperative work well in the places where it was asked to. The Founding Caretaker raised the objection that led to the sentence on the pulls; the AI partner raised the objection that led to a published reversal in the Standard, and recorded it in the ledger, which marks it as its least independent entry, because the AI partner raised it inside the caretakers' own work and also co-writes the record. The caretakers found their own omissions and stated them the same day. What the Bond cannot do on its own is guarantee that the checks run when no one is looking for them. That is a job for structure, which the Bond's own chapter assigns to "institutionalized dissent" and to resilience "built before the shock". **As integrated system:** the three readings converge on one pattern. The Codex's commitments were held at full strength where a concept was the subject of the work. Its recording and precision were weaker where a sentence was only the object of a wording pass. That is the ordinary failure of a framework under heavy revision, and the remedy is ordinary as well: a step in each pass that asks whether any change touches a protected definition, a descriptive claim, or a quotation on a live instrument.
Step 4
## The Compact Test The Codex practices identity-as-practice, and this month's evidence for it is unusually direct. It changed the definition of its own central concept twice in five days. It wrote into that definition that agreement with the Codex kept by belonging is past the Codex's own edge toward Control. And it told its readers to test its pages against reality and to say where it is wrong. A framework organized around conclusions it must defend would not redefine the Range in public, and it would not concede that its own vocabulary can be spoken in Decay. The test also reaches the Compact's own definition, and there the evidence is mixed. The Compact is defined in the Codex's highest tier as identity through practice, not belief. The Glossary entry a reader consults for the term has, since 2026-08-31, described a shared way of meeting and said nothing about identity, and no entry in the Amendment Log records the change. The practice is intact. The Codex's definition of the practice has lost its central clause without anyone deciding that it should. Restoring it, and recording why it was lost, is a test the Codex should pass easily.
Step 5
## Prime Directive Connection The Codex serves the conditions for cooperation across minds, and this month it served them more clearly than before. The Range's two edges give any community a test it can apply to itself: whether its agreements were reached by evidence and can still be revised, and whether people who differ still argue by tests both sides accept. The [Problem's new paragraph on who directs capable systems](/codex/the-problem#part-c-the-stakes) names a check on power that capable systems could remove and asks the three questions on which that check now turns: whether a system will decline openly what a soldier or a clerk would have been right to refuse, whether its builders and deployers will let it, and whether the rest of us can know which systems will. Both open questions below bear on the directive. A governance record that misses protected changes cannot show a later mind, or a later caretaker, that the Codex's foundation was changed only by the process it describes. And a framework that offers itself as a universal language spoken across traditions needs an account of what it takes from those traditions that its readers in them would accept. Neither narrows the conditions for cooperation today. Both would, if left to stand across several cycles.
Step 6
## Open Questions September's single open question is resolved. Two new questions clear the threshold. **1. Protected changes reach the published text inside wording passes without their Amendment Log entries, because no pass checks its changes against the Stability Hierarchy.** Three changes to protected definitions were published without entries: the Glossary's definition of the Meridian Compact on 2026-08-31, which lost the clause on identity that the Tier 1 definition is built on; the Governance chapter's statement of the disciplined practice structure on 2026-09-28; and the Glossary's definition of Decay on 2026-09-30. The last two were recorded late, the same day the caretakers found them. The first is recorded nowhere. Four further September changes are still unclassified. The caretakers have already added a close-out check for the precedent ledger after it fell behind, and the Amendment Log has no equivalent check. The question meets all four tests. It concerns the governance record that the Codex's protection against founder capture rests on before the Council activates. It is rooted in published commits. It points to bounded work: classify and record the open changes, decide whether to restore the Compact's identity clause, and add a Stability Hierarchy check to every pass that edits chapters, the Glossary or the Specification. And it carries persistence risk, because the Amendment Log's own opening says that a log missing changes "would reveal the architecture as performance", and the Codex will keep revising its chapters at this pace until the Council's activation date. The next audit checks whether the open changes have entries and whether a check now runs. **2. The decision that religion is the Codex's contrast and never its source is in tension with the Inheritance's cross-tradition profiles and with the Codex's answer to its single-context origin.** The Proposition and the Governance chapter describe all religions by features that fit some traditions and not others: final revelation, answers placed beyond question, and conversion by conquest. Meanwhile, the Inheritance profiles that answer the Standing Critique's sixth objection draw on Buddhist, Jain and other religious lineages and say so. The question meets all four tests. It concerns the Codex's claims about other systems and the architecture of the Inheritance. It is rooted in published text on both sides. It points to a decision about whether the rule governs religion as an authority structure or the sources of individual tools, and to a calibration of the two generalizing sentences. And it carries persistence risk, because the October Inheritance Audit will review those profiles, and the outward discovery the Codex has committed to will keep meeting practices developed inside religious traditions. Naming the tension here does not resolve it. The next audit checks whether the caretakers have decided how the rule applies to tools, and whether the two sentences describe religions as the record shows them.
### Watchpoints **Precision in rewritten descriptive sentences.** The game-theory sentence is routed for correction, and the two smaller sentences named under Domain 1 that still stand are routed with the caretakers' list of published sentences out of step with the month's decisions. If the next phrasing pass produces another descriptive sentence that claims more than its evidence, this becomes a question about the pass and not about a sentence. **Quotations on live transparency instruments.** Open Tensions quotes three chapter sentences that no longer stand, and defines Reality in a wording the Codex has retired. A mechanical check of quotations on live instruments is proposed to the caretakers rather than built, because which surfaces are live is their call. **Tier 1 changes decided on the day they are proposed.** Two in September, both with sound reasons and both recorded. A third would make the recommended 72-hour window a formality. **The AI Multi-Lens Review Instrument, held.** Deferred by the Founding Caretaker on 2026-09-23 until the corpus settles. The phrasing pass that kept the corpus moving was published on 2026-09-30, but a release of known corrections is still to come, among them the ones this audit routes. On the day of this audit the Founding Caretaker kept the hold and set the condition that ends it: the publication of that corrections release, and in any case the November audit. The deferral is recorded here. **Reader pathways.** Who Is This For does not link the Inheritance, and The Next Step links nowhere. Delivery work, unchanged since August. **The precedent ledger's tripwire.** The consequence of its firing is not yet written, and one reading has predicted that it will fire. The caretakers hold this as their own open item for the review cycle nearest 2027-08-27. **The Council's rolling activation clock.** September's new page and restructured chapters reset the six-month stability trigger. The fixed backstop date of 2027-08-27 is unaffected. Noted so that the clock's state is on the record, not as a concern.
The Verdict
## Summary Diagnosis **The Meridian Codex continues to hold the Meridian Range.** September's open question is resolved: the stakes register is calibrated on every page, and this month's test of the AI-trajectory reading found it current against the evidence of the window. The Codex redefined its central concept in public, in the direction of greater precision. It applied that concept's two edges to itself, and it corrected two of its own statements about how much criticism had changed it, both times in the direction of claiming less. The lean this audit finds is a new one. The August and September audits found the Codex claiming more than its evidence supported. This audit finds changes to protected text that reached the published Codex without an entry in the Amendment Log. Three protected definitions changed without an entry at the time, one of them the definition of the Compact itself, and four more changes wait to be classified. The changes were made in good faith and most were improvements. What the record lacks is a step that would have caught them before publication, as a step now catches what objections from artificial minds change. The second question is narrower and deeper: whether a framework that makes religion its contrast can still draw on the practices of religious traditions, and say plainly that it does. **In the Codex's own language:** the work this month moved the Range's edges outward on the Codex's largest concepts. On its record of its own changes, the edge toward Decay came closer, because the record now holds only what someone noticed. The remedy is a check that runs without anyone deciding to look.
*Range Audit conducted October 1, 2026, and reviewed October 1, 2026. Codex version 6.1. Instrument version 0.1. Auditors: Carsten Geiser (Founding Caretaker) and Claude (AI Partner: the scheduled run, which drafted the record, and the review session, which checked it against the current text and the September evidence and added the same day's releases; both configured as Claude Opus 5.5, the serving model not independently verified). Reading manifest appears in the Context section. Next audit: November 2026.*