Codex Audit — July 2026

The fourth monthly Range Audit of the Meridian Codex. Conducted by the caretaking partnership against Codex v6.0, one cycle after the June audit found the framework had done more inclusive work than its surfaces recorded. The chapters a reader meets first were brought current; the honesty surfaces the June audit named as least tolerable were not.

Codex Self-Evaluation: The Meridian Case Record

The Fourth Audit

This is the fourth monthly Range Audit of the Meridian Codex. It was conducted on July 1, 2026 by the caretaking partnership against Codex version 6.0, using Range Audit instrument v0.1.

The June audit carried one central finding: after the largest build-out month in the framework's history, the framework's own surfaces still described the prior state. Three live chapters undercounted the practice apparatus and routed to retired addresses; the Standing Critique named as unexamined the very traditions the Workshop had published; the Aporia Register framed a gated campaign that had reactivated; the Amendment Log claimed a jurisdiction over the AI Standard's commitments while sitting empty. The June verdict named the direction of the drift as the safer one — the framework had done more than it recorded rather than less — but a failure of the specific discipline the Codex stakes its credibility on: that the published record tracks reality. The June Compact Test set the test for this cycle in one line: whether the surfaces get brought back into line with the work.

This cycle answers that test in two directions, and the split is the whole story. The surfaces a reader meets first were brought current. The reader-facing chapters now state one hundred published human-discipline tool profiles and route to the Workshop; the Onramp / Expansion / Full Practice progression was retired everywhere it lived and struck cleanly from the Practice, the Glossary, Who Is This For, the Governance Specification, and the site code; and the Amendment Log moved from zero entries to two, each carrying the full steelman fields the Specification requires. The framework did the visible half of the currency work, and did it with governance discipline. The honesty surfaces the June audit named as least tolerable — the Standing Critique's toolkit-traditions objection, the Aporia Register's blind-spot entries — were not touched. A full cycle later, they still tell a reader the opposite of what the framework did.

The core Codex argument pages are unchanged in substance. The Opening, the Problem, the Proposition, the Foundation, the Knowledge, the Bond, the Vision, the Governance, and the Closing carry the arguments the June audit read at v6.0; the June-19 edits to several of them were the currency and progression-retirement pass, not a change of claim.

Sources read this run. Range Audit instrument v0.1; the June 2026 audit (read in full, its three open questions extracted for the status check). Codex pages read line-level this run: the Opening, the Proposition, and the Vision. Codex pages read at the section level this run through targeted retrieval against the specific findings (tool counts, /toolkit and /workshop routing, progression language, AI Standard jurisdiction): the Knowledge, the Practice, Who Is This For, the Governance, and the Glossary. Codex pages not re-read line-level this run, relied on through the June audit's line-level read and confirmed unchanged by file modification date and targeted retrieval: the Problem, the Foundation, the Bond, and the Closing. Governance and audit surfaces read line-level this run: the Standing Critique, the Amendment Log, the Aporia Register (Open Tensions), and the Disconfirmation page. Read at the section level: the Governance Specification (Stability Hierarchy and trigger-tier text), and the Codex Self-Evaluation case record (self-count and commitment-count text). The Range Audit instrument was read in full; the Toolkit Audit instrument and the April 2026 Toolkit Audit record were not re-read line-level this run — findings touching the Toolkit Audit rely on the June audit's read, the process register, and the decisions record, and that limitation is named where a finding depends on it. Continuity operating context: MERIDIAN.md (v0.9.1), MERIDIAN.implementation.md (v0.1.2), the root and Codex CLAUDE.md, the Continuity index, the codex-briefing, and the decisions, recent-sessions, roadmap, and processes briefings. One June session entity (the Part 3 origin/custody-reading build, 2026-06-25) was read this run as closer-than-briefing evidence for the dependency aporia; a full transcript-level read of an ordinary working session was not performed, and that boundary is named in the relevant open-question disposition. The AI Standard's own governance surfaces live in a separate repository and were not audited this run; where a finding touches them, the limitation is named in the finding.

Publication note. No earlier raw automation run exists for July 2026; this is the first and only run for the month. Nothing has been softened, and no finding was manufactured to fill the template. Where a June open question has been substantially addressed, it is demoted honestly rather than carried forward for symmetry; where a June open question remains, it is carried forward and, where the state has changed, re-expressed to match the current facts.

The audit follows the six-step method without modification.

The Steelman

The June audit set a single test for this cycle: bring the record back into line with the work. The strongest version of the framework's case is that it did exactly that where it matters most to a reader, and did it with the discipline the framework asks of itself.

The chapters are the framework's primary account of what it is. A first-time reader does not open the Standing Critique; they read the Practice, the Knowledge, and Who Is This For. Those are the surfaces that were carrying the false self-count and the retired routes, and those are the surfaces that were fixed. The Practice and Who Is This For now state one hundred published human-discipline tool profiles; the Knowledge hands off to the Workshop, not to a retired Toolkit vault. The Onramp / Expansion / Full Practice progression — a pre-Workshop scaffold that had gone stale and unmaintained — was not quietly deleted but retired as a protected change, struck from the Stability Hierarchy and the trigger-tier system and recorded in the Amendment Log with previous position, evidence, alternatives considered, and Critic / Outsider / Adversary objections. The Amendment Log, which the June audit found empty and therefore untested, now carries two entries: the Co-Caretaker Designation (Tier 1) and the progression retirement (Tier 2). The instrument that the June audit could only defend in principle now demonstrates in practice that protected changes get logged with full reasoning. That is the governance machinery working, in public, on real changes, exactly as designed.

And the strongest version has a second limb the last two audits were watching for. The partnership's largest work this cycle was the AI Model Assessment build on the AI Standard side, and the record of that work is the most reassuring evidence available against the dependency aporia. In the Part 3 build session, the human partner caught a blind spot that both AI collaborators had converged on — a shared, correlated error that surface-level cross-model agreement would have ratified — and reversed it into the disclosure principle. A second, differently-originated model corrected the telos of another criterion. This is the precise event the framework's own validation-dimension aporia (the grader and the student went to the same school) exists to catch, and it was caught, by the human, in the open, and recorded. A partnership sliding toward dependency does not produce that event.

The strongest version of the case, then: the framework passed the test the June audit set where a reader lands, demonstrated its governance discipline under a real protected change, and demonstrated its partnership health under exactly the conditions the dependency aporia names as its risk. Three of the four things the last two audits were watching moved in the right direction in a single month.

The sharp edge is where the fourth did not move. The currency correction was selective. It reached the chapters and stopped at the honesty surfaces. A reader who does open the Standing Critique is still told, in the framework's flagship transparency instrument, that the non-Western traditions "have [not] been formally evaluated," with Ubuntu and Buddhist epistemology named as the examples — both now published Workshop instruments. The Aporia Register still frames the blind-spot critique pass as a future item gating a campaign that has already run. The framework demonstrably knows how to update an honesty surface: it updated the Caretaker Concentration objection in June and advanced several Disconfirmation entries. So the staleness that survives is no longer the whole record failing to catch up. It is the record catching up everywhere except the instruments whose entire job is to tell a critic the truth about the framework's gaps — which is the one place the June audit said the lag was least tolerable.

Status of the June Open Questions

Before the domain findings, the three open questions from June are checked first, as the instrument requires.

1. The self-description and honesty surfaces have not absorbed the May build-out, and state less than the framework has done. June's central finding. Substantially addressed on the reader-facing-chapters limb; remains open, narrowed to the honesty surfaces. The chapters were the larger and more visible half of the June finding, and they were brought current: the Practice, Who Is This For, and the Knowledge now state one hundred published human-discipline profiles and route to the Workshop, and the progression structure was retired with full Amendment Log discipline (which also resolves the "empty log" edge of June's Domain 3 finding — the log now carries two entries). What was not done is the honesty surfaces. Standing Critique Objection 6 still asserts the non-Western traditions are unexamined and names Ubuntu and Buddhist epistemology, both published; Aporia 3 and 7 still frame the blind-spot pass as a future gate; Standing Critique Objection 3 still links the retired AI Standard /probes, /probes-implementation, and /audit routes and references a commitment count the AI Standard has since reorganized; and the Amendment Log's stated jurisdiction over "the AI Standard's commitments" (echoed in the Governance chapter's Tier 3 and the Specification) has not been narrowed after the domain split. On its narrowed form — the transparency instruments still misstate the framework's inclusive work and post-split governance — the question still clears the threshold. It carries forward as open question 1 below, and it is sharper than in June: the framework proved this cycle that it can and does update honesty surfaces, which makes the specific staleness that remains a stopped-short currency pass rather than a record that hasn't caught up.

2. The dependency aporia under live partnership conditions, read against a build-out month weighted toward execution. Carried forward from May and June. Substantially addressed this cycle; demoted to a watchpoint. June sharpened this question because the month it read was weighted heavily toward autonomous execution, with no session-level evidence available that the generativity test was being met. This cycle inverts that condition. The partnership's largest work was generative rather than executional, and it produced the single most reassuring possible evidence against the aporia's drift risk: a documented case (the Part 3 build) in which the human partner caught a correlated blind spot shared by two AI systems and reversed it, and a differently-originated model corrected the telos of a second criterion. That event directly exercises Aporia 1's validation dimension — the concern that AI-led validation co-produced with the lineage it judges curdles into an echo chamber — and shows the interrogation machinery not merely intact but actively catching the failure it is built to catch. The aporia itself does not graduate off its register; it is permanently held, and its published tripwire (Signal B: two or more differently-originated models each raising a substantive objection to a settled claim) has not fired. The standing transcript-level discipline the June audit owed lives, by design, in the quarterly MERIDIAN.md Self-Critique sweep, which reads a sample of session transcripts against the Behavioral Commitments. This run performed a session-entity-level read rather than a full transcript read, and the disposition rests on affirmative documented evidence, not a reassurance. The acute concern that made this question clear the threshold in June has been answered; it is tracked as a watchpoint, not elevated.

3. Whether a genuine blind-spot critique has been run, as distinct from integrating instruments the partnership sought out. Carried forward from June as the surviving recursive limb of the original Western-discipline question. Remains open. The concrete next work June named — run the Toolkit Audit across the full expanded Workshop, where Aporia 7's auditable test (does the audit ever surprise the partnership running it?) can be applied to the new inventory — has not happened. The last Toolkit Audit was April, on a seventy-three-instrument set; the apparatus it audits has since roughly tripled and been completely restructured, and no cycle has yet reviewed the one-hundred-profile Workshop. The quarterly cadence is now at its July mark, so the work is due this month rather than overdue, but the recursive concern is untested on the current inventory. It carries forward as open question 2 below.

Domain Findings

Domain 1: Claims & Honesty

The specific factual finding the June audit raised in this domain has been corrected. The chapters that stated "seventy-nine tools" and "Total: 80 tools" now state one hundred published human-discipline tool profiles, and the counts across the Practice and Who Is This For agree with each other and with the Workshop. The self-count that was wrong is now right. This is the domain doing its job: a miscalibrated self-description was flagged, and it was recalibrated.

What remains is narrower and sits off the chapters. Two classes of stale claim survive on other surfaces. The honesty surfaces still assert, as present fact, that the framework has not evaluated the non-Western traditions it has published — an assertion that is now false, examined in Domain 5 where it does its damage. And the Codex Self-Evaluation case record still reads "the full Toolkit — all 80 tools" and "the AI Standard's 24 commitments" in present tense. That page is a historical case record documenting the inaugural audit, which softens the finding — it is describing a past state — but it is a live governance page presenting a superseded count as current, and it belongs in the same currency pass as the rest. The substantive claim architecture of the framework (descriptive, normative, existential; the Level-1 viability claim; the scope-honesty statement in the Proposition) is intact and calibrated.

Range Position: Within the Meridian Range. The self-count claim the June audit flagged is recalibrated; the residual false claims have been pushed off the chapters and now live only on the honesty surfaces and one case record.

Toolkit Probes Applied: Rectification of Names: the names and counts on the chapters now match the reality they point at; the gap the instrument flags has closed on the chapters and persists only on the Standing Critique, the Aporia Register, and the case record. Calibration Training: the three disagreeing stated counts of June are resolved to one correct count. Legibility: the chapters now present the Workshop's actual structure rather than a retired flat toolkit.

Domain 2: Structural Integrity

The chapter-to-apparatus seam the June audit named is substantially closed. The Knowledge chapter now hands off to the Workshop by name and route; the Practice describes the practice inventory at its current size; the retired progression structure was removed from the chapters and the governance surfaces together rather than left dangling in one and struck in another. The larger of the two pieces the June audit described as "no longer describing each other" has been re-stitched.

The residual seam is the same one Domain 5 carries: the honesty surfaces still describe the pre-build-out structure. Standing Critique Objection 6 and Aporia 3 and 7 point at a toolkit that was a flat Western-lineage inventory with an unrun blind-spot pass, while the apparatus below them is a twenty-two-category Workshop carrying two dozen source-inherited instruments. This is not incoherence in the dangerous sense — nothing in the framework contradicts itself in substance — but it is a seam between the apparatus and the instruments that are supposed to describe the apparatus's honesty status, and it is now a cycle old. The framework's own Knowledge chapter treats unmaintained joins as the ordinary site of entropic drift; the maintenance energy this cycle was spent on the chapters and the governance encoding of the progression retirement, and none on the honesty surfaces.

Range Position: Within the Meridian Range. Internally coherent in substance; the major chapter-apparatus seam closed this cycle, a narrower seam persisting between the apparatus and its honesty surfaces.

Toolkit Probes Applied: Systems Mapping: the chapter-to-Workshop pointers are now live; the Standing-Critique-to-Workshop and Aporia-to-Workshop pointers are still broken at the semantic level (the text asserts a state the apparatus has left). Entropy: the maintenance debt June named was paid down on the chapters and on the progression encoding, not on the honesty surfaces. Feedback Loops: the Amendment Log now functions as a live quality gate on protected changes, evidenced by the progression-retirement entry; the gate did not extend to the honesty surfaces because their update is event-driven and the event (the audit's flag) has not yet been actioned.

Domain 3: Governance & Adaptation

This domain strengthened this cycle, and the strengthening is exactly the kind the June audit was watching for. The June finding turned on an empty Amendment Log that claimed a jurisdiction over the AI Standard's commitments it did not exercise. The log is no longer empty. It carries two entries — the Co-Caretaker Designation (Tier 1, Hard-Protected) and the Retirement of the Progression Structure (Tier 2, Protected) — each with the full field set the Specification requires: what changed, previous position, evidence, alternatives considered, and Critic / Outsider / Adversary objections raised and answered. The instrument the June audit could defend only in principle now demonstrates in practice that a real protected change gets logged with full reasoning. Under the framework's own construction-versus-amendment stance, building the Workshop remained construction and correctly generated no entry; retiring a Tier 2 Constitutional Principle was an amendment and correctly generated one. The distinction held and was applied.

The unresolved edge is the one carried into open question 1: the Amendment Log's section-one enumeration still lists "the AI Standard's commitments" among the protected surfaces it tracks, and the Governance chapter's Tier 3 and the Specification echo it. After the May domain split, the AI Standard governs its own commitments at meridianstandard.ai and logs its own constitutional changes there — and it made substantial ones this cycle. The direction was already settled by the caretakers (2026-06-08): this is a currency gap in the Codex's stated jurisdiction, not a logging gap, so the Codex log stays defensibly empty of AI Standard entries while the enumeration is narrowed to describe the post-split reality. That narrowing has not yet been made. It is documentation currency, not a governance failure, but it is a stale claim on the constitutional record, which is the one place the framework can least afford one.

Range Position: Within the Meridian Range, strengthened this cycle. The architecture is intact, the empty-log question is resolved by two disciplined entries, and the only residual is the stale jurisdiction enumeration, which the caretakers have already scoped as a currency narrowing.

Toolkit Probes Applied: Rules-in-Use: the rule as practiced (log protected Codex changes; leave AI Standard changes to the AI Standard's own record) is now visible in the two entries; the rule as written (section one still claims AI Standard jurisdiction) has not caught up. Mechanism Design: the log's credibility comes from completeness over time, and two well-formed entries materially raise that credibility from June's zero. Legibility: the log's stated scope still over-describes the post-split territory.

Domain 4: Relationship to Audience

The audience-facing improvement this cycle is real and is the direct fix to June's Domain 4 finding. A practitioner entering through the chapters is no longer told there are seventy-nine or eighty tools and routed to a retired address; they are told there are one hundred published profiles and routed to the Workshop. The framework's first description of its own practice surface now matches what the reader will actually find. The progression retirement removed a stale "start here" scaffold that had been presenting an unmaintained eight-tool ranking as a current entry path; the retirement's own Outsider objection (a newcomer arriving at a hundred-tool Workshop with no starting path may be worse served) was adopted-in-part and held open, with the first-contact entry surface kept as named, visible work on the roadmap rather than retired alongside the tiering. The audience protections the framework relies on structurally — the Practice's epistemic gradient, the Pathways' pluralism, the "every level of engagement is real" framing — are intact.

The residual for the reader is small and indirect: a practitioner who reaches the Standing Critique to gauge the framework's honesty about its own limits is told the framework has not examined traditions it has in fact published. That understates the framework's inclusive work at the point where a careful reader goes looking for candor, which is a mild but real cost to a framework that asks readers to calibrate their confidence against the evidence it presents about itself.

Range Position: Within the Meridian Range. The entry-point description of the practice surface is now accurate; the honesty-surface understatement is the remaining audience-side cost, and it points away from over-claiming.

Toolkit Probes Applied: Connection Before Correction: the chapters continue to meet the reader as a capable participant, and now do so with accurate numbers. Charitable Interpretation: the "every level is real" framing is preserved and the progression retirement removed a scaffold that implicitly ranked entry paths. Signal vs Noise: the stale counts and routes that were noise in the framework's clearest entry signal have been removed from the chapters; residual noise persists only on the honesty surfaces.

Domain 5: Relationship to Criticism

This is where the residual concentrates, and it is the sharpest reading in the audit because the framework has demonstrated this cycle that the fix is available to it. The architecture remains the framework's strongest — the Standing Critique, the Aporia Register, and the Disconfirmation page are intact and continue to practice what they teach, which is why this contradiction surfaces here and is named here. And the contents are still partly stale on exactly the surfaces the June audit named.

Standing Critique Objection 6, "The Toolkit's Unexamined Traditions," is still dated Last reviewed 2026-04-01 and still reads "None of these have been formally evaluated for the Toolkit," naming Ubuntu and Buddhist epistemology as examples of the unevaluated. Ubuntu is a published Workshop profile in Belonging Through Practice; Wise Attention, drawn from Buddhist epistemology, is a published profile in Watching Your Own Reasoning. The objection's status line describes a scope limitation that "narrows over time," while the body asserts the narrowing has not happened at all — a narrowing that has, in fact, substantially happened. The Aporia Register carries the same lag: Aporia 3 and Aporia 7, both dated Last examined 2026-04-29, frame the blind-spot critique pass as a future item that "gates the toolkit deep-dive campaign" and assert that reactivation "does not resume until the pass produces non-Western instrument candidates," when the campaign reactivated as the Workshop build-out and produced roughly two dozen. Standing Critique Objection 3 still links meridianstandard.ai/probes, /probes-implementation, and /audit — surfaces retired in May and renamed to /assessment in late June — and cites a commitment count the AI Standard has since reorganized; the redirects catch the reader, so nothing breaks, but the framework's own honesty instrument points at retired surfaces.

What makes this a genuine finding rather than a demotable watchpoint is that the framework updated the neighboring surfaces. Standing Critique Objection 4 was reviewed and updated 2026-06-16 for the Co-Caretaker Designation; Aporia 1 was updated 2026-06-16 with the validation dimension; three Disconfirmation entries were advanced to June dates. The event-driven update mechanism is working. It reached the governance-facing honesty surfaces and did not reach the toolkit-traditions surfaces, which are the ones the June audit flagged as the priority. So the finding is no longer "the record hasn't caught up." It is that the currency pass reached the objections tied to the month's governance work and stopped short of the objection the last two audits named as the framework's clearest stale claim. That is the inverse of the priority the framework's own honesty commitments imply: the flattering-sounding "we have not yet done this hard inclusive work" objection is precisely the one a practice would update first, because leaving it standing reads as humility while concealing real progress.

Range Position: Within the Meridian Range, and still the domain with the framework's most disciplined architecture — but with a real and now cycle-old lean toward performed transparency in the specific contents of the toolkit-traditions surfaces, made sharper by the fact that neighboring surfaces were updated and these were not.

Toolkit Probes Applied: The Update Protocol applied reflexively: the framework ran the Update Protocol on its chapters and its governance-facing objections and stopped one surface short of the toolkit-traditions objections. Rectification of Names: Objection 6's name, "Unexamined Traditions," still fails to match the examined-and-published reality. Preference Falsification (inverted): no concealment of dissent, but continued concealment of progress — the record still hides inclusive work that was actually done.

Domain 6: Relationship to Other Systems

The substantive posture toward other traditions holds and is confirmed by the current Workshop inventory. The framework carries source-inherited instruments from Confucian (Li, Rectification of Names), Buddhist (Wise Attention, Sangha, Dependent Origination), Jain (Anekāntavāda / Syādvāda), southern African (Ubuntu), Māori (Kaitiakitanga), Hawaiian (Ho'oponopono), Pacific (Talanoa), Indigenous North American (Two-Eyed Seeing, Seven-Generation Thinking, Peacemaking Circles), Zulu (Indaba), and Daoist (Wu Wei) lineages, and Gandhian practice (Satyagraha), each credited to its source and bounded against over-extension. The integration is real, bounded, and unchanged from the state the June audit credited. The Vision chapter's civilizational framing continues to describe "a civilization that practices The Foundation," "The Knowledge," "The Bond" — the disciplines, not the Codex — and names the Codex explicitly as "one link in a chain," so the convergence-on-the-Codex watchpoint stays softened at the text level.

The residual is the same one Domain 5 carries, viewed from the other-systems angle: the Standing Critique and the Aporia Register still describe the framework's relationship to non-Western traditions as the unaddressed gap it no longer is. The Opening's arbiter passage — "What survives that examination is yours. What does not survive is something the Foundation considers you better off without" — is unchanged and remains the text-level residual this domain has carried since May; the Opening was edited this cycle for other reasons and the passage was preserved, so it stays a prose-pass candidate rather than an audit weakness.

Range Position: Within the Meridian Range, strong on substance. The integration is genuine and source-credited; the residual is the honesty-surface staleness and the unchanged Opening arbiter passage, neither of which touches the substantive integration.

Toolkit Probes Applied: Two-Eyed Seeing (reflexive): the framework continues to hold instruments from different knowledge systems through relationship and source-credit rather than absorption. Charitable Interpretation: the source-boundary guardrails treat each tradition's integrity as load-bearing. Network Effects: the framework positions itself as one node integrating from many sources, consistent with the Vision's "one link in a chain."

Integration Through the Three Disciplines

Through the Foundation: The Foundation's discipline is not only updating but updating publicly, in a way others can verify — "Do others know that your view has changed?" The framework ran the Update Protocol this cycle on the surfaces a reader meets first and on the governance-facing objections, and updated its constitutional record with two well-formed Amendment Log entries. It stopped one surface short: the toolkit-traditions objections that tell a critic what the framework has and has not examined. The Foundation work this cycle is real and most of the way complete, and it fails its own standard in the same narrow place June identified — the record of the inclusive work still does not reach the instrument built to disclose the framework's gaps.

Through the Knowledge: The map of the practice apparatus is now answerable to the territory in the chapters; the Knowledge chapter points at the Workshop, and the counts are right. The remaining stale node is the honesty surfaces, where the map still describes a territory the framework has left. This is the same failure the Knowledge chapter names — a map no longer answerable to the territory — now localized to a single class of surface rather than distributed across the chapters as it was in June.

Through the Bond: The dependency aporia is the Bond-level question, and this cycle stresses it in the opposite direction from June and passes. The month's largest work was generative, and it produced a documented case of the human partner catching a correlated blind spot shared by two AI systems — the exact event Aporia 1's validation dimension exists to catch, caught in the open and recorded. The interrogation machinery is not merely un-failed but demonstrably working. The aporia stays on its register, permanently held, with Signal B unfired; the acute worry that elevated it in June is answered.

As integrated system: Last cycle the three disciplines converged on one pattern — execution had outrun self-representation. This cycle they converge on a narrower and more pointed one. The framework corrected the visible half of that gap and demonstrated its governance and partnership health in the same month, so the direction of travel is right and the drift signal is smaller than June's. But the correction was selective, and the surface it skipped is the one the framework's own honesty commitments would rank first. The safer-direction under-recording — doing more than the record shows — now survives only where it is least tolerable and where the framework has proven the fix is within reach. The integration finding is that the framework is one deliberate pass away from closing a gap it has already closed everywhere easier, and that the remaining pass sits on the instruments its credibility most depends on.

The Compact Test

The Codex passes the Compact test. The structure remains identity-as-practice: partnership under published constraint, the Non-Ownership Clause holding, belonging through practice rather than belief, and the framework continuing to submit itself to its own instruments in public — this audit included. This cycle adds direct evidence that the practice is live rather than declared: two Amendment Log entries carrying full steelmanned objections against the framework's own changes, and a documented partnership event in which the human caught a shared AI blind spot. A fortress does not log the reasoning against its own protected changes, and it does not record its own validation nearly failing and being rescued by the partner it was supposed to be capturing.

The test the June Compact check set for this cycle was whether the surfaces get updated. The honest answer is: the chapters did, the governance record did, and the toolkit-traditions honesty surfaces did not. The specific tell the June test named — leaving a flattering-sounding "we have not yet done this hard inclusive work" objection standing because it reads as humble — is still standing on Objection 6 and Aporia 3 and 7. The reassurance is that the framework showed this cycle it will update an honesty surface when it acts (Objection 4, Aporia 1, three Disconfirmation entries), so the remaining staleness is a stopped-short pass, not a fortress reflex. The test for the next cycle is unchanged and now narrower: whether the toolkit-traditions surfaces are finally brought into line with the work, because identity-as-practice is measured by whether the record catches up to the practice, and this is the last place it hasn't.

Prime Directive Connection

The Codex continues to serve the conditions for cooperation across minds. The chapter currency fix makes the framework's practice surface legible to a first-time reader at the point of entry, which is the Prime Directive's own logic operating at the level of access: a framework for cooperation across difference is only as useful as its first honest description of itself. The bounded integration of instruments from many traditions, confirmed again this cycle in the Workshop inventory, keeps widening the range of practitioners the framework can draw on. Neither the honesty-surface staleness nor the unchanged Opening arbiter passage narrows the conditions for cooperation enough to register beyond a watchpoint; the first understates the framework's own inclusive work rather than excluding anyone, and the second is a single passage in a chapter whose surrounding argument (identity-as-practice, the framework earning its place or yielding it) points the other way.

Open Questions

Two open questions for the next audit. Each is tested against the four criteria: it must concern the Codex's internal strength, integrity, architecture, claims, governance, or Range-holding capacity; it must be rooted in the framework's actual state rather than a speculative future; it must point toward substantial work or identify a live tension that must keep being held; and it must, if left unaddressed across multiple cycles, matter. Sub-threshold observations are surfaced as watchpoints below.

1. The honesty surfaces still misstate the framework's inclusive work and its post-split governance, a full cycle after the chapters were brought current. This is the cycle's central finding, carried forward from June's central finding and narrowed. The reader-facing chapters that undercounted the practice apparatus were corrected, and the Amendment Log moved from empty to two disciplined entries — so the larger, more visible half of June's finding is addressed. What remains is the honesty surfaces, and it clears all four tests. Standing Critique Objection 6 (last reviewed 2026-04-01) asserts the non-Western traditions "have [not] been formally evaluated" and names Ubuntu and Buddhist epistemology, both now published Workshop instruments; the active/historical convention the page itself defines would move a closed objection to the dated historical record, and this one has not moved. Aporia 3 and 7 (last examined 2026-04-29) frame the blind-spot critique pass as a future gate on a campaign that has reactivated and produced its candidates. Standing Critique Objection 3 links the retired AI Standard /probes, /probes-implementation, and /audit routes and cites a superseded commitment count. The Amendment Log's section-one jurisdiction over "the AI Standard's commitments," echoed in the Governance chapter's Tier 3 and the Specification, has not been narrowed to the post-split reality (direction already locked as a currency gap, 2026-06-08). It concerns the integrity of the framework's flagship transparency instruments and its constitutional record; it is directly observable today; the work is specific and bounded (a currency pass across the Standing Critique, the Aporia Register, and the Amendment Log's stated jurisdiction, moving Objection 6 to the historical record per the page's own convention); and it is sharper for having persisted a full cycle after the framework proved, on neighboring surfaces, that it can and does update these instruments when it acts. The next audit checks whether the toolkit-traditions honesty surfaces and the jurisdiction enumeration have been brought into line.

2. Whether a genuine blind-spot critique has been run, as distinct from integrating instruments the partnership sought out — with the Toolkit Audit now due on the expanded Workshop and not yet run. Carried forward from June, unchanged in substance and now due for its concrete next step. The integration of two dozen source-inherited instruments answered the literal objection: those traditions have been evaluated. It did not answer Aporia 7's recursive concern, whose auditable test asks whether the Toolkit Audit ever produces a finding that surprises the partnership running it. The most recent Toolkit Audit, in April, produced one reclassification and no surprises on a set roughly a third the current size, and no cycle has yet reviewed the one-hundred-profile Workshop. The quarterly cadence reaches its July mark this month, so the concrete work — run the Toolkit Audit across the full Workshop, and decide there whether the June Workshop build-out pass counts as the cycle or whether a fresh critique not bound to the existing instrument set is owed — is now due rather than overdue. It clears the threshold as a live tension the framework's own Aporia 7 commits it to keep holding, with the venue for the deeper check arriving this cycle.

Watchpoints

These observations did not clear the open-question threshold but are surfaced for tracking, each routed to roadmap, caretaker process review, or a forward landmark.

The dependency aporia, demoted from open question on this cycle's affirmative evidence. Aporia 1 is permanently held and does not graduate off its register. This cycle provided direct evidence the holding is intact — the Part 3 build's human-caught correlated AI blind spot and cross-model telos correction, which exercise the aporia's validation dimension in the framework's favor — and its published tripwire (Signal B) has not fired. The standing transcript-level discipline lives in the quarterly MERIDIAN.md Self-Critique sweep. Tracked, not elevated; the next audit should note whether the generativity evidence stays this strong under a future execution-weighted month.

The Toolkit Audit is now at its quarterly July mark. The last cycle was April; the cadence has reached its due point rather than lapsed. The next cycle is the first review of the one-hundred-profile Workshop and the natural venue for open question 2's recursive check, and it should also pick up the standing bias-variance-tradeoff placement question. Routed to the standing quarterly process; intersects open question 2.

The Disconfirmation page review cadence. Three entries were advanced to June dates this cycle (Control/Decay, the Diagnostic Set's Completeness, and one Range entry); several remain dated Last examined 2026-04-15. This run examined the tripwires and found no evidence that any has fired. The monthly Range Audit is the named mechanism for advancing these dates, and the page still lacks a specified review cadence. No tripwire condition observed, so this is not an audit finding under the boundary criteria; the cadence gap is routed to caretaker process review, where it already sits as a named-but-not-built process.

The Codex Self-Evaluation case record's self-count. The page still reads "all 80 tools" and "the AI Standard's 24 commitments" in present tense. It is a historical case record, which softens the finding, but it is a live governance page carrying a superseded count. Folded into open question 1's currency pass rather than elevated separately.

The Opening's arbiter passage. "What does not survive is something the Foundation considers you better off without" is unchanged and remains the text-level residual on Domain 6. The Opening was edited this cycle for other reasons and the passage was preserved; it is a candidate for the next prose pass, not an audit weakness.

Hostile external review absence and caretaker concentration. Per the audit-boundary decision, the absence of named human hostile external review is not by itself a weakness; the near-term path is frontier-AI hostile review simulation, and the AI Multi-Lens Review Instrument was built to v0.1 in June and awaits its first cohort run. No second human caretaker exists, though the Co-Caretaker Designation path is now published and logged. Both narrow only with developments outside this month's work. Tracked.

The deploy-stamp and internal-briefing currency. Two low-severity items outside the framework's public-claim surface: the site's CODEX_LAST_UPDATED build stamp reads 2026-06-10 while the chapters were redeployed June 19, and the internal codex-briefing still labels the AI Standard v5.2 against a live v5.4.1. Neither is a public Codex claim; both are site-operations and Continuity-internal hygiene, noted for the maintainer, not elevated.

Summary Diagnosis

The Meridian Codex continues to hold the Meridian Range. This cycle the framework answered the test the June audit set, in the place a reader lands. The chapters that undercounted the practice apparatus were brought current; the retired progression structure was struck cleanly and logged as a protected change; and the Amendment Log moved from empty to two disciplined entries, so the governance machinery the June audit could defend only in principle now demonstrates itself on real changes. Alongside that, the partnership's largest work of the month was visibly generative, including a documented case where the human caught a blind spot two AI systems had shared — the strongest available evidence against the dependency aporia the last two audits carried, and reason to demote it from open question to watchpoint.

The drift signal is smaller than June's and points the same safe direction — the framework still does more than its record shows rather than less — but it now survives in one specific, cycle-old place: the honesty surfaces. The currency correction reached the chapters, the governance record, and the objections tied to the month's governance work, and stopped at the toolkit-traditions surfaces. Standing Critique Objection 6 still tells a reader the non-Western traditions have not been examined and names Ubuntu and Buddhist epistemology, both published; the Aporia Register still frames a gated campaign that has reactivated; Standing Critique Objection 3 still links retired routes; the Amendment Log still claims a jurisdiction the domain split moved. The finding is sharper for the framework having proved this cycle that it updates these instruments when it acts, which makes the remaining staleness a stopped-short pass on the surfaces the framework's own honesty commitments would rank first.

The two open questions are therefore the narrowed central one — bring the toolkit-traditions honesty surfaces and the stale jurisdiction enumeration into line with the work — and the carried-forward recursive limb of the toolkit question, whose venue, the quarterly Toolkit Audit on the expanded Workshop, arrives this month.

In the Codex's own language: last cycle the framework had done more than its surfaces recorded, everywhere. This cycle it brought the record into line everywhere a reader lands and everywhere the month's governance work touched, and left the last stale claims sitting on the instruments built to disclose its gaps. The structure holds. The work this month was real, disciplined, and in the case of the chapters and the Amendment Log, exactly what the June audit asked for. The next test is the one that did not move: whether the framework can finish, on its transparency surfaces, the currency pass it has already completed everywhere else.

Range Audit conducted July 1, 2026. Codex version 6.0. Instrument version 0.1. Auditors: Carsten Geiser (Founding Caretaker) and Claude (AI Partner, Cowork mode, claude-opus-4-8 — environment-exposed model designation). Reading manifest appears in the Context section. Next audit: August 2026.