--- title: "Workshop Audit — July 2026" sidebarTitle: "July 2026" description: "The second cycle, and the first on the Workshop. A full-inventory review of 22 categories and 111 tool profiles under the repaired instrument, with a declared coverage deviation, a falsification pass on every category, and dispositions on all six carry-forwards from April." aiSummary: "The second Workshop Audit cycle and the first run under the repaired instrument. Reviews all 22 designed categories (20 published, 2 provisional) and all 111 tool profiles across Foundation, Knowledge, and Bond. Declares a coverage deviation: category definitions and the full tool inventory were read directly, tool profiles at structured-summary depth with targeted full reads, not full text on all 111. Disposes of all six April carry-forwards: three were closed incidentally by the May restructure without any instrument noticing (institutional-design gap, emotional-regulation gap, cross-cultural Bond gap), two redundancy questions resolve in favor of keeping both tools with the distinction now visible, and the trust-propagation question narrows to one tool and stays open. The category rubric produces the cycle's central finding: Reading What's Operating holds 25 of 111 profiles under a claim broad enough to absorb anything, and functions as an unsegmented remainder rather than a category. Further findings: the Bond carries ten designed categories against six each for Foundation and Knowledge, which sharpens rather than resolves April's largest structural question; three categories share the display name Catching Your Own Drift; and Speaking Honestly When It Costs carries two tools doing different jobs. The falsification pass ran on all 22 categories and published its cases. Proposed enactments are marked and unenacted pending caretaker adjudication." --- Workshop Audit — Cycle 2026-07
Context
## The Second Cycle This is the second cycle of the [Workshop Audit](/audit/the-workshop-audit) and the first run on the Workshop. The inaugural April 2026 cycle reviewed 73 instruments in a flat list. The May restructure replaced that list with a three-level tree, and the inventory has since grown to 111 published profiles across 20 published of 22 designed categories. No cycle has reviewed it. The cycle is one quarter late. Its July mark passed with the month closing, and the July Range Audit carries the gap as its open question 2. It also runs under an instrument repaired immediately before it, in the same session: the six-question rubric had no unit of analysis for categories, no coverage rule, and no mechanism that would have produced a different result in April than one reclassification and seventy-two confirmations. Repairing before running was the caretaker's call. The repairs are recorded in the [Amendment Log](/governance/amendment-log) as Tier 3, Flagged. Two consequences follow for how this record should be read. The method is one cycle old and has no track record. And the falsification pass, which is the repair carrying the most weight, is adopted provisionally: the next cycle is required to judge whether the cases below are arguments or ritual.
01 // Coverage
## Coverage and Depth — Declared The instrument requires each cycle to state what it read. This one carries a deviation, and it is stated first rather than buried. **Read directly and in full:** the canonical category registry, including the designed order, discipline assignment, and one-line claim for all 22 categories and both provisional placeholders; the category entry pages; the complete tool inventory of all 111 profiles by title and description, enumerated against the filesystem rather than against a register, and independently counted to confirm 111 across 20 published categories. **Read at structured-summary depth:** the tool profiles. Each profile carries an authored summary stating its mechanism, its practice, its lineage with named sources, and its cross-references. These are condensations written to represent the page, not extracts skimmed from it, and they are sufficient to run disciplinary fit, redundancy, and scope honesty. They are **not** sufficient to run field movement or instrument reliability on a contested tool, which is why the tools receiving those questions below were read in full. **This is a bounded reading and therefore a declared protocol deviation** under the instrument's coverage clause. The deviation is: full-text reading was performed on the profiles receiving prose findings and on the profiles named in carry-forwards, not on all 111. What the cycle can claim is bounded accordingly. Structural findings about categories, redundancy, placement, and naming rest on evidence adequate to them. A claim that any individual tool's field has or has not moved rests on a full read where one was performed, and is not asserted where one was not. **Remedy scheduled:** the next cycle carries a full-text obligation on the Reading What's Operating and Watching Your Own Reasoning inventories, which together hold 42 of 111 profiles and received the least per-tool attention here. **Depth model applied.** Full rubric prose for all 22 categories. Tool-level prose where a rubric question produced a finding, where a carry-forward named the tool, or where the falsification pass built a case. Every remaining tool appears by name in a no-finding list under its category. The lists and the Workshop should match exactly; a reader finding a tool in neither is looking at a gap in this record.
02 // Dispositions
## The Six Carry-Forwards from April The instrument now requires an explicit disposition on every open question carried in. April generated six and nothing carried into it, so this is the first cycle in which update hygiene has ever been exercised. Three dispositions are the same kind, and that kind is itself the cycle's second finding. ### 1. Bayesian Reasoning's Foundation fit — **resolved, confirmed** April moved Bayesian Reasoning from Knowledge to Foundation and asked that the move be stress-tested. Under the restructure it landed in Foundation → Calibrating Confidence to Evidence, alongside Calibration Training and Interiority Calibration. The category's job is proportioning belief strength to evidence strength, which is exactly the work the April reclassification argued Bayesian Reasoning does. The placement is not merely retained; it is now positively supported by a category boundary that did not exist when the move was made. Confirmed. Closed. ### 2. Base Rate Neglect's placement — **resolved by relocation** April placed this at the Foundation/Knowledge boundary and left it in Knowledge, noting a future cycle should decide. It now sits in Foundation → Watching Your Own Reasoning, with the other cognitive biases. The question April deferred has been answered in the direction April leaned. Closed. ### 3. Nash Equilibrium and Inadequate Equilibria — **resolved, both retained** April found the distinction real but asked that the pages make the relationship explicit. Both tools now sit in the same category, which sharpens rather than blurs the comparison, and both now cross-reference each other by name. The distinction holds on inspection and is sharper than April could state it. Nash Equilibrium diagnoses strategic stability: mutual best responses, no unilateral improvement available. Inadequate Equilibria diagnoses *exploitability*: whether an actor exists who could fix the system and capture enough benefit to bother, where authority is bottlenecked, and whether local wedges remain when global repair is blocked. One asks whether a state is stable. The other asks why nobody is fixing it and where the seam is. Different mechanisms, different practices, different outputs. April's requested remedy was enacted. Closed. ### 4. Information Degradation and Signal vs Noise — **resolved, both retained** April framed this as an information-theory / network-dynamics overlap. The framing was slightly off: network dynamics is not part of the actual adjacency. The real pair is the two information-theory-rooted tools, and both now sit in Knowledge → Checking Your Map Against Reality, cross-referencing each other. The distinction is clean. Signal vs Noise governs the input: what should be allowed to change the map, given a defined question, a detection threshold, and the asymmetry between false positives and false negatives. Information Degradation governs the path: what happens to information as it travels, through resolution loss, source loss, context loss, compression, retelling, and summary drift. A filter at the gate and a decay function along the channel. Co-location made the distinction more legible, not less. Closed, with the note that April's original framing named the wrong second field. ### 5. The trust-propagation evidence base — **narrowed, remains open** This is the only carry-forward that survives, and it survives in a more specific form than April stated it. Graduated Reciprocity carries a solid research base: Axelrod and Rapoport on direct reciprocity, Osgood's GRIT, and Ostrom's graduated sanctions as an institutional neighbor. No concern. **Trust Thermocline** is a different case. Its lineage is John Bull's popularization of the term and Bruce F. Webster's earlier Thermocline of Truth metaphor, drawn from product, publishing, and organizational-trust discourse, with psychological contract breach and exit/voice/loyalty as adjacent research neighbors rather than as its source. The profile is honest about this — it says so on the page — and the underlying phenomenon of accumulating unvoiced breach followed by sudden collapse is well attested in the adjacent literature it names. But the tool's own frame is industry discourse, and it is the thinnest-sourced instrument in Calibrating Trust to Behavior. The tool earns its place on usefulness and on the honesty of its own lineage statement. The carry-forward stays open, narrowed from "trust propagation" to this one tool, with the next cycle asked whether the adjacent research can be promoted into the tool's primary lineage or whether the tool should be presented explicitly as a practitioner heuristic rather than a research-grounded instrument. ### 6. The Bond discipline's structural clarity — **remains open, and sharpened against the framework's interest** April called this the largest structural question carried forward: whether the Bond is as clearly defined as the Foundation and the Knowledge. The restructure did not answer it. It produced evidence bearing on it, and the evidence does not point the comfortable way. The Foundation carries six designed categories. The Knowledge carries six. **The Bond carries ten.** Trust calibration, honest speech under cost, receiving disagreement, repair, cooperative diagnosis, bad-faith operation, cost-bearing continuation, drift, belonging, and stewardship across time. The tool counts are close across disciplines — 35, 37, 39 — so this is not a size effect. The Bond required nearly twice as many cuts to organize a comparable body of tools. Two readings are available and the audit does not get to pick the flattering one by default. The first: relational work is genuinely more differentiated than thinking work or mapping work, and ten categories reflect real structure. The second: a discipline whose territory needs ten cuts where its siblings need six has a less settled center, and the extra cuts are the discipline's boundary problem expressed as taxonomy rather than solved by it. The audit cannot separate these from the category counts alone. What it can say is that the restructure did not resolve April's question, and that anyone treating the ten categories as evidence the Bond is now well-defined has the inference backwards. Carried forward with a specific test proposed for the next cycle in the open questions below. ### The disposition that needed naming Three of the six — the institutional-design gap, the emotional-regulation gap in Foundation, and the cross-cultural gap in Bond — are handled below under their own finding, because they share a disposition the instrument now requires to be named as such. All three are closed. None was closed by an audit.
03 // Finding One
## The Framework Answered Its Own Audit Without Noticing April identified three gaps. All three are closed. The audit did not close any of them, no record connected the work to the questions, and a full cycle passed with the questions still listed as open. **The institutional-design gap.** April's words: the Knowledge tier "has nothing specific to political economy, institutional design, or governance structures," Ostrom appears only in passing under Tragedy of the Commons, and "this is the largest gap the audit identifies." The Workshop now carries **Rules-in-Use**, inherited directly from Elinor Ostrom's Institutional Analysis and Development framework and citing McGinnis and Cole; **Action Situation Mapping**, the IAD action-situation itself; **Polycentric Governance**, carrying Vincent Ostrom, Tiebout and Warren's metropolitan governance work, Elinor Ostrom's Nobel lecture, and the Bloomington School by name; and **Mechanism Design** for incentive engineering. The largest gap April named is not narrowed. It is filled, by four tools, with the discipline's foundational framework named as lineage. **The emotional-regulation gap.** April: the Foundation "is heavy on cognitive biases and light on emotional regulation," with no dedicated tool for the emotional dimension of range-holding. There is now a dedicated *category* — Staying Steady Under Pressure — holding **Affect Labeling**, with the Lieberman et al. 2007 fMRI work and the 2011 labeling-reappraisal-distraction study as lineage, and **The Discipline of Assent** from Stoic practice. **Psychological Flexibility**, ACT-derived, sits adjacent in Holding Beliefs Without Identity. Closed. **The cross-cultural gap in Bond.** April: the tier "has nothing on cross-cultural communication beyond the High/Low Decoupling tool." The Workshop now carries Ubuntu, Li, Sangha, Indaba, Talanoa, Hoʻoponopono, Peacemaking Circles, Kaitiakitanga, Seven-Generation Thinking, Two-Eyed Seeing, Anekāntavāda / Syādvāda, Satyagraha, and Wu Wei. Closed, and not narrowly. **The finding is not that three gaps closed.** Three gaps closing is good news and would be an unremarkable line in a healthy record. The finding is the mechanism. A build-out pass answered the audit's three most substantive content questions, and the audit had no way to know. The questions sat in the April record. The tools went into the Workshop. Nothing connected them, and the process register kept listing the questions as open for a full cycle. This is a failure of the audit, not of the build-out. An instrument whose open questions can be answered without the instrument noticing is not tracking its own state, and its "open questions" section is a list of things nobody has checked rather than a list of things that remain true. The repaired instrument's mandatory-disposition rule exists because of exactly this, and the incidentally-closed disposition was written into it during the repair on a first-read suspicion that this section now confirms. **Range position:** the framework's practice is healthier than its instrument. The gaps were real, they were closed by good work, and the record of that work was invisible to the mechanism built to read it.
04 // The Category Layer
## The Twenty-Two Categories The category rubric runs here for the first time. Four questions: boundary integrity, naming honesty, payload, and coverage read across the discipline. ### Finding Two: Reading What's Operating is not a category **Knowledge → Reading What's Operating. 25 tools.** Its claim: "Seeing what is operative in any system at any scale — the same Control–Decay patterns from the individual mind through the civilization, across short and long horizons." The category holds 25 of the Workshop's 111 profiles. Twenty-two percent of the entire practice apparatus sits under one heading, in a discipline whose other four published categories hold 2, 2, 6, and 2. Its contents span game theory (Prisoner's Dilemma, Nash Equilibrium, Schelling Points, Mechanism Design, Positive-Sum vs Zero-Sum), coordination failure at scale (Moloch, Tragedy of the Commons, Inadequate Equilibria), institutional analysis (Rules-in-Use, Action Situation Mapping, Polycentric Governance, Legibility), systems dynamics (Entropy, Feedback Loops, Emergence, Leverage Points), adaptive dynamics (Antifragility, Red Queen Effect, Evolutionary Mismatch, Lindy Effect), measurement and suppression (Goodhart's Law, Chilling Effects, Network Effects), Buddhist causation (Dependent Origination), and AI safety (Instrumental Power Tripwires). **Boundary integrity fails.** The rubric's test is whether a reader holding a tool and the category list can say which category it belongs to and why. For this category the answer is residual: a tool belongs here if it is about systems and is not map-checking, triangulating, acting, or drift. That is a definition by exclusion, which is what a remainder looks like. **Naming honesty fails in the specific way that hides the first failure.** "Seeing what is operative in any system at any scale" is not inaccurate. It is unfalsifiable as a boundary. A claim broad enough that no systems tool could fail it cannot do the work a category name is supposed to do, which is to tell a reader what is inside and what is not. **Payload passes, and that is why the problem is easy to miss.** The 25 tools are individually strong, well sourced, and doing real work. Nothing here is filler. A category can be well-stocked and still be the wrong container. The steelman, which the audit takes seriously: the category may be making a genuine unifying claim rather than collecting a remainder. "The same Control–Decay patterns from the individual mind through the civilization" asserts that cross-scale pattern recognition is *one* skill, and splitting it into domain buckets would fragment a unified practice into a filing system, teaching the reader that game theory and thermodynamics are separate lenses when the Codex's claim is that they are the same lens at different scales. That is a real argument and it is the reason this finding proposes a study rather than a split. **Proposed enactment: none this cycle.** The audit does not have enough to redraw a boundary this load-bearing inside a review, and a 25-tool split proposed by the same pass that found the problem would be the audit legislating from a single reading. Proposed instead: a dedicated category-architecture study before the next cycle, with the four visible seams — strategic interaction, coordination failure at scale, institutional reading, and system dynamics over time — as its starting hypotheses, and the unifying-claim steelman as the position it has to defeat. ### Finding Three: three categories share one name Each discipline carries a category called **Catching Your Own Drift**. All three have the same display name. The Foundation's route is `catching-your-own-drift`; Knowledge and Bond carry discipline-qualified routes. The parallel design is right — each discipline should name its own failure modes — and the Range-facing symmetry is one of the Workshop's better structural choices. Two problems sit on top of it. **Naming honesty.** A reader encountering "Catching Your Own Drift" in navigation, search, a citation, or an AI summary cannot tell which of three categories is meant. The disambiguation lives in the route, which is not what a reader reads. **An unearned default.** The Foundation's instance holds the unqualified route while its siblings carry qualifiers. Nothing in the architecture makes the Foundation's drift category more canonical; the asymmetry is an artifact of which one was built first. Payloads also diverge sharply: Foundation 4 tools, Knowledge 2, Bond 7. Parallel names across categories doing visibly different amounts of work. **Proposed enactment:** give all three discipline-qualified display names — *Catching Your Own Drift — Foundation*, *— Knowledge*, *— Bond* — and either qualify the Foundation route or state on the page why it keeps the short one. Low cost, and it removes a genuine reader-facing ambiguity. ### Finding Four: Speaking Honestly When It Costs carries two tools doing different jobs **Bond → Speaking Honestly When It Costs. 2 tools.** Claim: producing honest signal inside cooperation under social pressure, especially when the easier move is to soften, delay, or hide. **Preference Falsification** fits precisely: the gap between private belief and public statement when social pressure makes honesty costly. **Stewardship of the Epistemic Commons** is about information hygiene — source-before-share, the amplification test, public correction — treating the information environment as a commons that careless amplification pollutes. That is a real practice and a well-sourced tool. It is not speaking honestly when it costs. Its own cross-references point almost entirely at Knowledge tools: Signal vs Noise, Information Degradation, Report Fidelity, Chilling Effects. **Payload fails.** Half the category's contents do not do the category's stated job. With only two tools, that is the whole margin. April flagged the disciplinary-fit question on this tool and resolved it in Bond's favor: Knowledge tools describe the commons dynamics, the Bond tool describes the practitioner's responsibility within it. That reasoning holds, and the audit does not reopen the discipline assignment. The category assignment is a different question, and it was made after April. **Proposed enactment: none, pending.** Two options, both with costs. Move the tool to a better-fitting Bond category — but no current Bond category is about information conduct. Or hold it and accept that the category name under-describes its contents. Recorded as a finding, routed to the same category-architecture study as Finding Two. ### The remaining eighteen categories Run through all four questions. No findings. **Foundation.** *Watching Your Own Reasoning* (17) — the largest Foundation category and the one that most invites the Finding Two treatment, but its claim is a genuine single job (making your own thinking visible to you while it happens) and its tools all serve it; boundary holds. *Holding Beliefs Without Identity* (4), *Calibrating Confidence to Evidence* (3), *Revising Beliefs Under Evidence* (5), *Staying Steady Under Pressure* (2) — clean cuts, honest names, payload matches claim. **Knowledge.** *Checking Your Map Against Reality* (6), *Triangulating Across Disciplines* (2), *Acting on What You See* (2) — all three carry tight claims and matching contents. Acting on What You See holds only two tools, both source-inherited, which is thin but not a failure: the category's job is distinct and its coverage question is noted below. **Bond.** *Calibrating Trust to Behavior* (6), *Receiving Disagreement Well* (4), *Repairing After Rupture* (6), *Diagnosing Cooperation* (3), *Cooperating Under Bad Faith* (5), *Belonging Through Practice* (3), *Stewardship Across Time* (3) — boundaries clean, names honest, payload matched. Cooperating Under Bad Faith deserves specific mention for doing the hardest naming work in the Workshop well: its claim explicitly bounds exclusion to "a particular setting" rather than in general, which is the safeguard the tool inside it needs. ### The two provisional categories **Continuing to See Under Cost** (Knowledge, designed order 4) and **Sustaining Cooperation Through Cost** (Bond, designed order 7) hold reserved places with no tools. The architectural claim behind them is legible and good: each discipline should carry a category for continuation under sustained cost. The Foundation's is built and published — Staying Steady Under Pressure. The other two are reserved against the same pattern. **Coverage says the Knowledge placeholder is warranted.** Its claim — continuing to see despite the isolation that structural clarity costs — names work no published Knowledge category does. Acting on What You See is about moving from perception; Checking Your Map is about correcting the model. Neither covers the over-time cost of continuing to see. **Coverage says the Bond placeholder needs a harder look.** Its claim is "continuing to extend oneself in cooperation when the cost has come due, without Fusion or Severance." Three published categories already carry pieces of that: Speaking Honestly When It Costs holds cost-bearing honest signal, Cooperating Under Bad Faith holds The Cooperative Vulnerability, and Fusion and Severance — the two failure modes the claim names as its boundaries — are tools inside Catching Your Own Drift — Bond. A reserved place whose stated payload is distributed across three neighbors is a weaker claim than a genuinely empty territory. **Proposed enactment: none.** Both retained this cycle. But the Bond placeholder is put on notice: if it takes no tool by the next cycle, the audit should ask whether it is a real gap or a symmetry artifact, and retirement by amendment should be on the table alongside publication.
05 // The Falsification Pass
## Cases Built Against the Equipment The instrument requires the strongest available case for retiring or merging each category's weakest-standing tool, published whether or not it succeeds. Twenty-two cases were built. Six produced findings and are given in full. The remaining sixteen are summarized with the tool targeted and the ground the case failed on, so a reader can see what was attempted rather than take the attempt on trust. ### Cases that produced findings **Moloch** — *Reading What's Operating.* The case: Moloch's direct modern source is a 2014 blog post, and the pattern it names is already covered inside the same category by four better-sourced instruments — Prisoner's Dilemma, Nash Equilibrium, Inadequate Equilibria, and Tragedy of the Commons — all four of which Moloch's own page cross-references. A tool whose cross-references are a list of tools that do its job more rigorously is a candidate for merge. **Outcome: survives, with a note.** The case fails on a distinction the formal tools do not carry. Prisoner's Dilemma describes a two-party payoff structure. Tragedy of the Commons describes extraction from a shared resource. Inadequate Equilibria describes why nobody fixes a bad state. Moloch describes the multi-agent competitive race in which every participant sees the destruction, none can unilaterally stop, and the optimizing pressure itself is the antagonist. That is a distinct shape, it operates at civilizational scale, and it is load-bearing in the AI-safety discourse the Codex engages directly. The lineage is thin, and the profile should say plainly that its source is an essay rather than a research literature. **Lindy Effect** — *Reading What's Operating.* The case: the practical yield is close to a restatement of base rates, and the actionable discipline it implies — do not remove inherited structure you do not understand — is carried more usably by Chesterton's Fence, which the Lindy page itself cross-references as "an adjacent caution about inherited structures." Merge into Chesterton's Fence or Antifragility. **Outcome: survives, narrowly, and this is the closest call in the cycle.** Lindy makes a claim Chesterton's Fence does not: survival duration is *evidence* about expected remaining duration for nonperishable things. Chesterton's Fence is about epistemic humility before removing something; Lindy is about reading endurance as a signal. Different claims. But the audit records that this was the weakest survival in the pass, and that a future cycle finding the two collapsing in practice would have grounds. **Trust Thermocline** — *Calibrating Trust to Behavior.* Case and outcome given under carry-forward 5 above. Survives on usefulness; scope-honesty question stays open. **High/Low Decoupling Communication** — *Diagnosing Cooperation.* The case: its primary lineage is online decoupling-and-contextualizing-norms discourse, with Edward T. Hall's high-context/low-context work named on the page as "adjacent but distinct." A published Bond tool whose primary source is internet discourse, holding a third of a three-tool category, is the weakest-standing instrument in its category. **Outcome: survives, with a scope-honesty note.** The phenomenon is real and cross-culturally attested, the practice is usable, and intercultural pragmatics gives it a genuine research neighborhood. But the profile should present the discourse lineage as its origin and the research as adjacent, which is what it currently does — so the finding is that the tool is correctly labeled and thinly sourced, not that it is mislabeled. **Murphyjitsu** — *Revising Beliefs Under Evidence.* The case: this is CFAR's branded name for a technique whose primary research source, Gary Klein's premortem, is named on the page and is both peer-reviewed and widely adopted in practice. The Workshop names the derivative and cites the source. **Outcome: survives as an instrument, but the naming finding stands.** The technique's Codex-relevant content — run the failed-future simulation, find the most plausible break, patch, repeat — is Klein's premortem plus the inner-simulator framing. Retitling to "Premortem" or "Premortem / Murphyjitsu" would name the source and keep the handle. Recorded as a proposed enactment for adjudication. **Stewardship of the Epistemic Commons** — *Speaking Honestly When It Costs.* Case and outcome given under Finding Four. ### Cases that failed cleanly Sixteen further cases were built and did not survive contact with the tool. Named here with the ground each failed on: **Dunning-Kruger Effect** (*Watching Your Own Reasoning*) — attacked on the regression-to-the-mean critique, which argues a substantial share of the original effect is a statistical artifact. Failed: the profile already carries Nuhfer's critique and the qualified contemporary picture, which is precisely the remedy April requested and is here confirmed as enacted. The practical claim survives the statistical critique. **Charitable Interpretation** (*Holding Beliefs Without Identity*) — attacked on disciplinary fit, as relational work sitting in Foundation. Failed: the primary work is how the practitioner reads ambiguity before self-defense engages, which is thinking discipline. April's reasoning holds. **Interiority Calibration** (*Calibrating Confidence to Evidence*) — attacked as speculative for a published tool. Failed: it governs confidence calibration about an open question, which is the category's exact job, and refusing both denial theater and inflated claims is the calibration discipline applied where calibration is hardest. **The Discipline of Assent** (*Staying Steady Under Pressure*) — attacked as redundant with Noticing. Failed: Noticing observes; assent-testing governs whether an impression becomes judgment. Sequential, not duplicative. **Epistemic Cowardice / Epistemic Arrogance** (*Catching Your Own Drift — Foundation*) — attacked as Codex-native vocabulary rather than instruments. Failed: naming the two directions of Foundation drift is the category's stated job, and failure-mode vocabulary is a legitimate instrument class. **Rectification of Names** (*Checking Your Map Against Reality*) — attacked as a philosophical doctrine flattened into a management tool. Failed: the profile carries the Analects citations and the modern cautions against exactly that flattening. **Wu Wei** (*Acting on What You See*) — attacked as too diffuse to operationalize. Failed: the profile bounds it to acting from contact with conditions rather than from domination or self-display, which is specific and testable. **Ideology / Paralysis** (*Catching Your Own Drift — Knowledge*) — attacked as a two-tool category too thin to publish. Failed: the pair is complete by design, naming both drift directions; thinness is not incompleteness. **Anekāntavāda / Syādvāda** (*Triangulating Across Disciplines*) — attacked as inaccessible. Failed: accessibility is a presentation question, not an instrument-reliability question, and the tool does work no Western-sourced instrument in the Workshop does. **Skin in the Game** (*Calibrating Trust to Behavior*) — attacked as Taleb-branded restatement of principal-agent theory. Failed: the practice reads exposure-to-consequence as a trust signal, which principal-agent framing describes but does not operationalize for a practitioner. **Loyal Opposition** (*Receiving Disagreement Well*) — attacked as a political-theory import without practitioner application. Failed: giving dissent standing inside a shared order is directly usable and distinct from Productive Conflict, which governs the exchange rather than the standing. **Common Knowledge Generation** (*Repairing After Rupture*) — attacked as game-theoretic machinery in a repair category. Failed: making breach and new terms mutually visible, and visibly visible, is the mechanism repair actually turns on. **Sabotage Diagnostics** (*Cooperating Under Bad Faith*) — attacked as a Codex-native essay rather than an inherited instrument. Failed: five named behavioral signatures read at pattern scale with an explicit high threshold and a Charitable-Interpretation check is an instrument, and its self-imposed guard against becoming a Control move is the strongest anti-abuse design in the category. **Fusion / Severance** (*Catching Your Own Drift — Bond*) — attacked as duplicating Cult Dynamics and Coordination Collapse. Failed: those name group-level failures; these name the individual's drift into or out of the Bond. **Li** (*Belonging Through Practice*) — attacked as ritual formalism with no modern application. Failed: patterned conduct forming belonging is the category's stated mechanism, and the profile handles the formalism critique directly. **Seven-Generation Thinking** (*Stewardship Across Time*) — attacked as rhetorical rather than operational. Failed: admitting descendants as stakeholders in a present decision is a specific and runnable move. **On this pass, honestly.** Six of twenty-two cases produced findings and none produced a retirement. That ratio is what a first run looks like and it is not self-evidently healthy: a pass that never retires anything is one cycle away from being a ritual that certifies the inventory. The instrument anticipated this and requires the next cycle to judge these cases as arguments or theater. The audit's own view, offered for that judgment: the Moloch, Lindy, and Murphyjitsu cases are real arguments the audit expected to win and lost; the sixteen summarized cases are thinner, and several were built against tools the audit did not seriously expect to fall.
06 // The Tool Layer
## Reviewed, No Finding Every tool not given prose above, by category. The lists below and the Workshop should match exactly. **Foundation — Watching Your Own Reasoning (17):** Scout Mindset, Noticing, Confirmation Bias, Motivated Reasoning, Availability Heuristic, Affect Heuristic, Base Rate Neglect, Fundamental Attribution Error, Sunk Cost Fallacy, Tribal Cognition, Attention as Resource, Wise Attention, Observer-Conditional Integrity, Optimization-Trace Awareness, Reasoning-Fidelity Boundary, Self-Model Grounding. *(Dunning-Kruger Effect given above.)* Attention as Resource carries April's Foundation/Knowledge boundary note; the placement holds, and the systemic dimension is carried by Knowledge tools as April intended. **Foundation — Holding Beliefs Without Identity (4):** Identity Decoupling, Intellectual Humility, Psychological Flexibility. *(Charitable Interpretation given above.)* **Foundation — Calibrating Confidence to Evidence (3):** Bayesian Reasoning, Calibration Training. *(Interiority Calibration given above.)* **Foundation — Revising Beliefs Under Evidence (5):** Chesterton's Fence, Double Crux, Steelmanning, The Update Protocol. *(Murphyjitsu given above.)* **Foundation — Staying Steady Under Pressure (2):** Affect Labeling. *(The Discipline of Assent given above.)* **Foundation — Catching Your Own Drift (4):** The Controlled Mind, The Decaying Mind. *(Epistemic Cowardice and Epistemic Arrogance given above.)* **Knowledge — Reading What's Operating (25):** Action Situation Mapping, Antifragility, Chilling Effects, Dependent Origination, Emergence, Entropy, Evolutionary Mismatch, Feedback Loops, Goodhart's Law, Inadequate Equilibria, Instrumental Power Tripwires, Legibility, Leverage Points, Mechanism Design, Nash Equilibrium, Network Effects, Polycentric Governance, Positive-Sum vs Zero-Sum Framing, Prisoner's Dilemma, Red Queen Effect, Rules-in-Use, Schelling Points, Tragedy of the Commons. *(Moloch and Lindy Effect given above.)* **Knowledge — Checking Your Map Against Reality (6):** Information Degradation, Signal vs Noise, Report Fidelity, Memory Provenance, Objective-Trace Audit. *(Rectification of Names given above.)* **Knowledge — Triangulating Across Disciplines (2):** Two-Eyed Seeing. *(Anekāntavāda / Syādvāda given above.)* **Knowledge — Acting on What You See (2):** Satyagraha. *(Wu Wei given above.)* **Knowledge — Catching Your Own Drift — Knowledge (2):** *(Ideology and Paralysis given above.)* **Bond — Calibrating Trust to Behavior (6):** Good Faith as Default, Graduated Reciprocity, Trust Diagnostics, Consentful Modification. *(Skin in the Game and Trust Thermocline given above.)* **Bond — Speaking Honestly When It Costs (2):** Preference Falsification. *(Stewardship of the Epistemic Commons given above.)* **Bond — Receiving Disagreement Well (4):** Connection Before Correction, Productive Conflict, Corrigibility-Autonomy Range. *(Loyal Opposition given above.)* **Bond — Repairing After Rupture (6):** Trust Repair, Hoʻoponopono, Indaba, Peacemaking Circles, Talanoa. *(Common Knowledge Generation given above.)* **Bond — Diagnosing Cooperation (3):** Psychological Safety, Schelling Fences. *(High/Low Decoupling given above.)* **Bond — Cooperating Under Bad Faith (5):** Adversarial Dynamics, The Cooperative Vulnerability, The Exclusion Problem, Trust Mining. *(Sabotage Diagnostics given above.)* **Bond — Catching Your Own Drift — Bond (7):** Coordination Collapse, Cult Dynamics, Defection Cascades, Echo Chambers, Groupthink. *(Fusion and Severance given above.)* **Bond — Belonging Through Practice (3):** Sangha, Ubuntu. *(Li given above.)* **Bond — Stewardship Across Time (3):** Inter-Instance Integrity, Kaitiakitanga. *(Seven-Generation Thinking given above.)* ### The bias-variance tradeoff disposition Carried in from the June 2026 prep note, which asked whether the concept earns a standalone Knowledge profile, merges into an existing Checking Your Map Against Reality tool, or stays a core-Codex teaching analogy. **Disposition: teaching analogy only. No Workshop profile.** The prep note's own mission-weight question is the right one — does this add a practice distinct from Signal vs Noise and Checking Your Map Against Reality — and the answer is no. The practice it proposes ("what real variation is this model refusing to see, and what noise is this model starting to obey?") is Signal vs Noise run in both directions. The underfit reading is the map hardened against the territory, which Checking Your Map Against Reality holds as its Control pull. The overfit reading is the map chasing noise, which is Signal vs Noise. The concept's genuine contribution is pedagogical: it makes the one-dial intuition click for readers who think in models, and it distinguishes the Range from mere moderation. That is teaching work, and it already lives in The Problem where the prep note recommends elevating it. A standalone profile would add a name without adding a practice, in the discipline that most needs its tool count to mean something. Recorded as closed, with the core-Codex elevation proceeding on its own track under the writing rules.
07 // Changes
## Changes Enacted **None.** No tool is added, retired, merged, or reclassified in this cycle. No category is split, renamed, or retired. The proposals below are proposals; nothing in the Workshop changes on the strength of this record alone. That is a defensible outcome for a review that found the inventory sound at the tool level and found its problems one level up, where the audit's own judgment is least tested. It would not be a defensible outcome twice.
08 // Watchpoints
## Watchpoints Observations below the finding threshold, recorded and routed. **Moloch's lineage presentation.** The profile should state that its direct source is an essay rather than a research literature. Routed to the profile's next revision, not elevated. **Lindy Effect's survival margin.** Recorded as the narrowest survival in the falsification pass, so a future cycle reaching the opposite conclusion has this cycle's reasoning to argue against. **Acting on What You See holds two tools**, both source-inherited, in a discipline where the neighboring category holds 25. Not a failure — the category's job is distinct — but the thinnest published category in the Workshop and worth watching for whether it is a real category or an early one. **The AI-specific profiles are distributed, not gathered.** Eleven AI-specialized tools now sit inside human-discipline categories rather than in a separate tier, which is the correct architecture and a real improvement on April's segregated AI tier. No finding. Noted because the change was significant, deliberate, and has not previously been reviewed by this instrument.
09 // What Carries Forward
## Open Questions **1. Reading What's Operating's boundary.** The category-architecture study proposed in Finding Two, with the four candidate seams as hypotheses and the unifying-claim steelman as the position to defeat. The next cycle checks whether the study ran and what it concluded. **2. The Bond's structural clarity.** Carried from April, sharpened, with a specific test proposed: for each of the Bond's ten categories, can the caretaker state in one sentence what work it does that no other Bond category does? Six of ten answered cleanly in this cycle's category review. If a future cycle finds the same six answering cleanly and the same four requiring paragraphs, that is evidence for the second reading — that the extra cuts express a boundary problem rather than solve one. **3. Trust Thermocline's lineage.** Whether the adjacent research (psychological contract breach, exit/voice/loyalty) can be promoted to primary lineage, or whether the tool should be presented explicitly as a practitioner heuristic. **4. The Bond provisional category.** Whether Sustaining Cooperation Through Cost is a real gap or a symmetry artifact, given that its stated payload is distributed across three published neighbors. Retirement by amendment on the table if it takes no tool. **5. Whether the falsification pass was argument or ritual.** Required by the instrument. The next cycle reads the twenty-two cases above and records its judgment. This cycle's own view is on the record to be argued with. **6. The coverage deviation.** Full-text obligation on Reading What's Operating and Watching Your Own Reasoning — 42 profiles — carried to the next cycle.
10 // Reasoning Log
## Where Judgment Was Load-Bearing **On not splitting Reading What's Operating.** The evidence for a boundary failure is strong and the audit believes the finding. It did not propose the split because a 25-tool restructure proposed by the same pass that discovered the problem, on a single reading, at summary depth, would be the audit legislating rather than reviewing. The declared coverage deviation applies most directly to this category. Proposing a study rather than a cut is the honest move given what this cycle actually read. **On the Bond's ten categories.** The audit had two readings available and no way to choose between them from the data it had. It recorded both and named which one the framework would prefer, because an audit that presents the flattering reading as the finding has failed at the thing it exists to do. **On the incidentally-closed disposition.** This finding was suspected during the instrument repair, before the review ran, and the disposition rule was written into the instrument partly in anticipation of it. That sequence is worth stating: the rule was not derived neutrally and then found to apply. A reader may reasonably discount the finding's independence on those grounds. The evidence itself — the four Ostrom-lineage tools, the dedicated emotional-regulation category, the thirteen cross-cultural instruments — does not depend on the order in which it was noticed. **On the falsification pass returning no retirements.** The audit considered proposing the Moloch merge to avoid a zero-retirement first run, and did not, because retiring a tool to demonstrate that the pass has teeth is the mirror image of confirming everything to avoid friction. Both substitute a posture for an evaluation. **Post-run note, 2026-07-31.** Caretaker review after this cycle closed found the paragraph above misdiagnoses its own discomfort. The hesitancy was not a calibration problem. It was the absence of a receiving process: the instrument as run gave a successful case nowhere to go except a line in an adjudication table, which left the burden of proof to be improvised each time and made every case carry more weight than a single pass should carry. The instrument is amended in the same deploy as this record — a successful case now opens a retirement review, the leaning is a trigger rather than a verdict, and the tool stays in the Workshop while the review runs. Open question 5 is therefore partly answered before the next cycle reaches it: the pass's problem was structural, not a matter of nerve. The findings, cases, and dispositions in this record are unchanged and were reached under the method as it ran. Nothing above has been reconstructed.
*Workshop Audit conducted July 31, 2026. Workshop architecture v0.2 — 111 published profiles across 20 published of 22 designed categories. Instrument renamed and repaired the same day, recorded in the Amendment Log as Tier 3, Flagged; this is the first cycle run under the repaired method. Auditors: Carsten Geiser (Founding Caretaker) and Claude (AI Partner, Cowork mode, claude-opus-5 — environment-exposed model designation). The coverage and depth declaration, including this cycle's protocol deviation, appears in section 01. Next cycle: October 2026, or sooner if a trigger forces one.*