The International Sustainability Standards Board commenced a human capital research project in April 2024; as of the most recent update it remains at research stage with no exposure draft, and its next milestone is undated (International Sustainability Standards Board, 2024– present). The predecessor project under the SASB standards, which assessed human capital themes across seventy-seven industries, transitioned without ever issuing a finalised metric set. Nor is the position stable on the voluntary side: the principal human capital reporting standard was withdrawn in August 2025 and replaced by a retitled second edition that moved from guidance to auditable requirements across eleven reporting areas (International Organization for Standardization, 2025). There is, at present, no international standard toward which a brain capital disclosure could converge. Proposition 12 (Measurement asymmetry). The set of human capital indicators currently mandated and assured consists of inputs, costs, demographics, hazards and policy statements. Every validated measure of the workforce's cognitive and psychological state lies outside all disclosure regimes. Enterprise brain capital is therefore systematically unpriceable from public information, and the dimensions of enterprise redefinition that rest on it are correspondingly unobservable to capital markets. Falsified by: a mandated and assured disclosure requirement, in any major jurisdiction, for a validated measure of workforce cognitive or psychological state; or by evidence that investors reliably infer such state from currently disclosed indicators. One structural feature of this proposition should be noticed, because it determines how much of the paper survives if the deskilling argument does not. Proposition 12 is a claim about what disclosure regimes require and what validated instruments exist. It does not depend on the magnitude of AI-driven capability erosion, on the transferability of the endoscopy finding to knowledge work, or on the reallocation conjecture. If Propositions 7 through 9 were all false—if AI turned out to leave capability untouched in every setting— the workforce's cognitive and psychological state would still be unmeasured, still be value-relevant on the evidence in Section 4.3, and still be absent from every reporting regime. The AI argument raises the urgency of the measurement gap and is not among its premises. A reader who rejects Sections 7 and 11's treatment of deskilling can reject it without touching the paper's central empirical claim. This proposition supplies the mechanism for the observational asymmetry documented earlier in this series. Across eighteen enterprises coded on five redefinition dimensions, twenty-seven of ninety cells could not be determined from public information, and eighteen of those twenty-seven—sixty-seven per cent—fell in Organization and Leadership (Kadowaki, 2026d). This is the one place in this paper where that case analysis functions as evidence rather than as illustration, and the reason is worth stating: what it counted Brain Capital Management VURA Working Paper Series 50 was the availability of public information, which is exactly the quantity Proposition 12 concerns. It did not measure capability, and no causal claim is drawn from it. Those two dimensions are unobservable because the asset underlying them has no reporting infrastructure, and Table 7 identifies the missing rows indicator by indicator. The asymmetry is therefore not a limitation of that study's design. It is a property of the disclosure regime, and it is remediable by the bodies that set the regime rather than by better research. 9.2 Design requirements Five requirements follow from the preceding sections, and they eliminate most of what is currently done in practice. State, not spend (Proposition 1). The framework measures conditions, not budgets or programme counts. Programme expenditure, participation rates and policy existence are excluded as indicators. Distributional, not central (Proposition 6). Base indicators are reported at the lower decile, not the mean. A framework reporting means will conceal the deficit precisely where the return is located. Team-level for Belonging (Edmondson, 1999). Psychological safety is a group property. A firm-level figure must be constructed from team-level measurement with the distribution reported, not from an individual-level average, because a firm can hold a satisfactory mean while containing units in which nothing is sayable. Unassisted for Build (Propositions 7 and 9). Build cannot be measured by assisted performance or by self-report. No existing validated instrument satisfies this requirement, and the framework does not pretend otherwise: Section 9.3 excludes the available proxies as inconsistent with the requirement and offers a protocol instead of an instrument. Anti-gaming by construction (Proposition 14, below). Any self-reported indicator that becomes a disclosure target will degrade. The framework must anchor on measures that are costly to falsify, and Section 9.4 concedes that the anchors, not the instruments, carry that weight. Validity conditions travel with figures. Where one construct's measurement depends on another's level—which is the case for Build, whose assessment requires conditions Belonging supplies—the dependency is disclosed rather than assumed away. A figure is reported together with the state of the construct its validity rests on. 9.3 A three-tier architecture The framework separates indicators by how they are generated and how difficult they are to corrupt. The tiers are ordered by increasing informativeness and increasing vulnerability, which is why Tier 1 anchors Tier 2 and Tier 3 exists at all. Brain Capital Management VURA Working Paper Series 51 Figure 4. The three-tier measurement architecture. Tier 1 anchors and validates Tier 2; Tier 3 substitutes verifiable structure for unverifiable intention. Tier 1: Anchors. Hard, administratively generated indicators already inside an assurance perimeter, used as anchors against which Tier 2 self-reports are validated. These are not measures of brain capital; they are measures whose movement is difficult to fake and which should co-move with brain capital if the construct is real. Certified sickness absence days; work-related ill-health cases including recorded mental ill health (GRI 403-10; Global Reporting Initiative, 2018); short-term and long-term disability leave incidence; regretted voluntary turnover; internal fill rate against external hire rate; training hours (ESRS S1-13); year-on-year change in average salary (Financial Services Agency, 2026). The validation logic follows Lee-Baggley et al. (2026), whose belonging scale predicted leave incidence—an outcome already in firm records. Tier 2: States. Validated instruments measuring the constructs directly, collected under census-style conditions and by a third party. Belonging: Edmondson's seven-item team psychological safety scale, reported as a distribution across teams, together with the Belonging at Work Scale. Base: the four-item workplace wellbeing module (De Neve & Ward, 2023) reported at the lower decile; the WHO-5 index, reported as the percentage below the cut-off of 50, which is disease-agnostic and therefore benchmarkable against general population norms (Topp et al., 2015); health-related productivity loss via the WHO Health and Work Performance Questionnaire, whose output is expressible in hours and currency and is therefore the natural bridge to a financial statement (Kessler et al., 2003). Work engagement via UWES-9 (Schaufeli et al., 2006) is included as a state indicator spanning Belonging and Base. Three exclusions from Tier 2 are deliberate, and the third is the framework's most serious concession. Clinical case-finding instruments—the PHQ-9 and GAD-7—are excluded despite their prevalence in workplace research. They create a legally sensitive health-data record, they were never validated for organizational-unit aggregation, and they are gameable in the direction of under-reporting precisely when conditions deteriorate. The WHO-5 is the defensible aggregate choice. Tier 3 — Mechanisms Structural arrangements: payment timing, hardship facility drawdown, dissent route and its usage, AI instructional design, time-securing mechanism Most informative · costly to fabricate · not in any standard Tier 2 — States Validated instruments, third-party collected, census response threshold: WHO-5 · 4-item wellbeing module (lower decile) · PS-7 (share of teams below threshold) · HPQ Directly measures the construct · vulnerable to distortion Tier 1 — Anchors Administratively generated, already assured: certified sickness absence · work-related ill health · disability leave · regretted turnover Not a measure of brain capital — the check that keeps Tier 2 honest validates Internal inconsistency between Tier 1 and Tier 2 is detectable without access to underlying responses (Section 9.4). Brain Capital Management VURA Working Paper Series 52 Second, work engagement is not offered as a measure of Build. An earlier formulation of this framework used it as a partial proxy, and that was inconsistent: Proposition 9 holds that self-report cannot measure Build, and UWES-9 is self-report. A framework cannot rule out an instrument class in one section and adopt a member of it in another. Engagement is a state of the workforce, not a property of its capability, and it is classified here accordingly. Third, and consequently, Tier 2 contains no measure of Build. This is stated rather than concealed, because the alternative—substituting any available self-report or assistedperformance indicator—would violate Propositions 7 and 9 and would produce a number that moves in the wrong direction precisely when the underlying capability is eroding. A framework whose least measurable construct may be its most consequential one has a defect; announcing the defect is preferable to hiding it behind a proxy. What the framework offers instead is an interim protocol, specified so that a firm can begin now and a standard-setter can evaluate it. Unassisted work-sample protocol (interim Build measure). At a fixed interval, a random sample of role-holders completes a held-out task drawn from the same task family as their ordinary work, without tool assistance, under observed conditions, within a fixed time box. Output is scored blind by assessors who do not know the period, against a rubric fixed in advance. The reported quantity is the trend in mean score for the role family, together with the sampling fraction and the completion rate. The protocol is modelled on the only design that has detected the effect in the field: the endoscopy study compared outcomes on procedures performed without the tool, in the ordinary course of work, before and after exposure (Budzyń et al., 2025). Two structural conditions and one construct limitation must be stated before it is used. Two conditions on validity. The protocol is exposed to practice effects on the held-out family, which is why the task must be genuinely held out and rotated. And participation is itself interpersonally risky, so in a low-Belonging organization the result will be resisted or gamed. The framework does not argue its way out of that circularity; it converts it into a disclosure requirement. Any Build figure produced by this protocol is reported alongside the Tier 2 Belonging distribution for the same population and period, and the proportion of teams below the psychological-safety threshold is disclosed as a validity condition on it. A Build figure taken in a population where a third of teams are below threshold is not a measure of capability; it is a measure of what people were willing to display. Disclosing the Belonging distribution lets a reader make that judgement instead of leaving it to the firm. The construct limitation, which the protocol does not overcome. Build as defined in Definition 4 covers two things that behave differently, and they should be separated: Domain-task capability—the ability to do the substantive work of a role without assistance. This is what the protocol measures, and it is a real quantity that AI deployment demonstrably moves (Bastani et al., 2024; Budzyń et al., 2025). • Brain Capital Management VURA Working Paper Series 53 Redefinition-relevant capability—the capacity to notice that the enterprise's model has stopped working, to formulate the problem, and to sustain the argument. This is what Definition 1 indexes brain capital to, and the protocol does not measure it. It is not a scorable discrete task, it has no rubric that can be fixed in advance without prejudging the answer, and its exercise is episodic rather than periodic. The honest position is therefore narrower than a single Build measure implies. The protocol is an early-warning indicator for the first layer and, at best, a lower bound on the second: a firm whose domain-task capability is degrading has almost certainly not preserved its redefinition capability, whereas a firm whose domain-task capability is intact has established nothing about the second layer. The framework offers no measure of the second layer and does not claim one. What it can offer is a structured qualitative complement, and it is included in Tier 3 rather than Tier 2 because it is not a measurement: a periodic review of the firm's actual redefinition episodes—what was noticed, by whom, how it reached a decision, how long it took, and whether it would have been noticed had the same conditions recurred—attested under the procedure in Section 9.3 below. That is a process record, not a score, and Section 11 records that the paper's most important construct remains unmeasured. The protocol is, then, a proxy for one layer and a defective one at that. It is offered because it measures the right kind of thing—performance without the tool—rather than a correlate of it, and because a firm that measures the first layer at least learns whether it is consuming capital. Section 9.6 states the study that would replace it. Tier 3: Mechanisms. Disclosure of structural arrangements rather than of policies or intentions. This tier is the framework's principal departure from existing human capital reporting, and its rationale is the finding, established earlier in this series, that every high-maturity case of enterprise redefinition possessed a structural mechanism for securing time while the contrast set did not, and that such mechanisms are disclosable yet almost never disclosed (Kadowaki, 2026d). The obvious objection is that this collapses into the boilerplate the paper criticizes. A firm asked whether it has installed educational guardrails on its AI deployment can answer yes, having installed a product with a tutoring feature, and the disclosure is then worth no more than row 17 of Table 7, which records only that a firm offers voluntary healthpromotion programmes. The objection is correct against any narrative version of this tier, and the paper accepts it. Tier 3 is therefore specified with a rule that excludes the narrative version: Tier 3 admission rule. A mechanism qualifies for disclosure only if it yields at least one quantity drawn from a system of record the firm maintains for another purpose— treasury, payroll, personnel, case management or access logs. The quantity, its source system and its period are disclosed with it. An arrangement that cannot be reduced to such a quantity is excluded from the framework and is treated as a policy statement, not a mechanism. The rule is what makes the tier auditable, because it moves verification from judging whether a design is genuinely pedagogical—which no auditor can do—to reconciling a • Brain Capital Management VURA Working Paper Series 54 reported figure against a system whose integrity is already within an assurance perimeter. Whether a guardrail is well designed remains contestable; how many employees completed protected unassisted practice in the period, and on what evidence, does not. One theoretically important mechanism does satisfy the rule, and it is worth naming because it shows the rule is not merely restrictive. The basis on which roles are defined— whether a role specifies the tasks allocated to it or the outcomes and judgements it is answerable for—is recorded in the personnel system, because role definitions are documents an employer maintains, and the proportion of roles operated on each basis is a count. This variable does not depend on any particular design proposal; a firm can disclose the proportion, the population it covers and the date of any transition, all reconcilable against the system of record, whatever vocabulary it uses. The companion paper in this series proposes one such design (Kadowaki, 2026c), but the disclosable quantity is the basis of definition rather than the adoption of that proposal. A firm asserting that it “operates with a strong sense of purpose”, by contrast, discloses nothing the rule admits. That is the distinction the tier exists to draw, applied to a construct that Section 5.3 makes the complement of Belonging and Section 7.4 makes the pivot of the reallocation conjecture. Applied alone, however, the rule still cuts too deep, and the objection is the one this paper makes against existing regimes. It excludes the accountability structure that Proposition 3 identifies as Belonging's necessary complement, the pedagogical quality of guardrail design that Proposition 8 makes decisive, and the redefinition-episode review just described—because no system of record counts any of them. A framework that discarded its three most theoretically important qualitative mechanisms in order to be auditable would have reproduced exactly the failure it diagnoses in Regulation S-K and the revised European standards: what can be counted is reported, and what matters is not. The rule therefore admits a second class, distinguished from the first and from narrative. Attested mechanisms (Tier 3b). A mechanism with no system-of-record quantity may be disclosed only if all five of the following hold. (i) Publication: the firm publishes the procedure itself, in full, not a description of its purpose. (ii) Attestation: an independent attestor examines whether that published procedure was followed in the period, and the attestor, the scope of examination and any exceptions are disclosed. (iii) Negative instances: the disclosure includes the cases in which the procedure produced nothing— periods in which the dissent route was not used, episodes reviewed in which nothing had been noticed—reported alongside the positive cases. (iv) Series integrity: a change to the procedure resets the series and is disclosed as a reset, so that improvement cannot be manufactured by redefinition. (v) Employee corroboration: the attestor puts to a random sample of the employees the mechanism covers one question—whether the published procedure describes their experience in the period—and the response distribution, the sampling fraction and the response rate are disclosed with the attestation, subject to the same cell-size suppression as Tier 2. The attestation is to process, never to outcome or quality. Tier 3a and Tier 3b are reported separately and are never aggregated into a single score. The objection to this tier is that it reproduces the policy disclosure the paper criticizes: a firm with a hollow procedure can follow it faithfully and be attested. That objection is Brain Capital Management VURA Working Paper Series 55 correct about attestation and incorrect about the tier, and the difference is condition (i). Attestation is not what disciplines a Tier 3b disclosure; publication is. An attestor cannot certify that a guardrail design is pedagogically sound, but a published design can be read by anyone who wants to judge it, including a competitor, a journalist and an employee who knows whether it describes their working life. Policy disclosure asserts a commitment and reveals no procedure; Tier 3b reveals the procedure and asserts nothing about its quality. A firm with a weak design must therefore either publish the weak design or not disclose, and the first is a materially worse position than a statement of intent. Condition (iii) adds a second asymmetry, because a firm inclined to greenwash does not want to publish the periods in which its mechanism produced nothing. Condition (v) converts an earlier observation—that an employee knows whether a published design describes their working life—from a rhetorical point into a check: a procedure that exists on paper and not in practice will show it in the corroboration distribution, and a firm must either report that distribution or not disclose. The check is not proof against pressure, and the cell-size and response-rate rules of Section 9.4 apply to it; but it moves the cheapest form of hollowness, the procedure no employee has encountered, from undetectable to visible. The precedent is internal-control and service-organization reporting, where an attestor certifies that a described control operated without certifying that it was well designed. The residual exposure should nonetheless be stated exactly, because it is real and the four conditions do not remove it: a firm that publishes a weak procedure, follows it, discloses its negative instances, survives employee corroboration and is attested has made an honest and uninformative disclosure. That is better than a dishonest and uninformative one, and it is worse than a Tier 3a quantity. The paper claims the improvement and not more, and Section 11 records the exposure alongside the gaming problem rather than treating attestation as a solution to it. Construct Mechanism Reportable quantity System of record Belonging Route for raising a problem that does not pass through the person it concerns Number of matters raised through it in the period, and number resulting in a reversed or amended decision Case management Base Wage payment frequency and timing Payment frequency; number of employees using any within-year timing flexibility offered Payroll Base Employee hardship liquidity facility Drawdown rate; mean amount; proportion structured as discharge rather than loan Treasury Base Caregiving support that alters circumstance Paid caregiving leave days taken and subsidised places used, reported separately from referral counts. Channel empirically untested (Section 6.7) Benefits administration Build Protected unassisted practice within AI deployment Participation rate and hours per participant; scored unassisted work samples completed Learning and access logs Brain Capital Management VURA Working Paper Series 56 Construct Mechanism Reportable quantity System of record Build Promotion criteria assessing managerial potential separately from current-role performance Proportion of promotions in the period assessed against the separate criterion Personnel All Time-securing mechanism (one of the five identified types) Which type; and the contractual, ownership or financing instrument that constitutes it Constitutional and financing documents Belonging / Build Basis of role definition: allocated tasks or allocated purpose and judgement principles (Kadowaki, 2026c) Proportion of roles operated on each basis; population covered; date of transition Personnel Belonging (3b) Accountability structure paired with psychological safety Attested to process: the specified structure and whether it operated in the period Third-party process attestation Build (3b) Instructional design of AI deployment, including guardrail architecture Attested to process: the published design and whether deployment conformed to it Third-party process attestation All (3b) Redefinition-episode review (Section 9.3) Attested to process: episodes reviewed, and whether the review procedure was followed Third-party process attestation Table 8. Tier 3 mechanism disclosures. Tier 3a items (unmarked) carry a quantity reconcilable against a system of record the firm maintains for another purpose. Tier 3b items (marked) have no such quantity and are subject to third-party process attestation instead; the two classes are reported separately and never aggregated. Proposition 13 (Verification standard orders informativeness). Brain capital disclosures are informative for enterprise value in descending order of their verification standard: Tier 3a mechanisms, carrying a quantity reconcilable against a system of record, are most informative and will exhibit the greatest cross-sectional variance and the lowest correlation with disclosure tone; Tier 3b mechanisms, attested to process only, are intermediate; and unattested policy or intention disclosure is least informative and will converge to boilerplate. The ordering is a prediction about verification standards, not about the theoretical importance of the constructs, which runs in roughly the opposite direction. Falsified by: Tier 3a disclosures showing no greater predictive content for subsequent outcomes than policy disclosures; Tier 3b disclosures indistinguishable from unattested narrative in variance or predictive content; or Tier 3a and 3b converging to boilerplate at the same rate as policy disclosure. 9.4 The gaming problem Any framework of this kind faces an objection that cannot be argued away and must be designed against. When a measure becomes a target it ceases to be a good measure (Strathern, 1997), and the best-evidenced demonstration is that mandated performance targets in a large public organization produce systematic gaming (Bevan & Hood, 2006). Self-reported wellbeing under managerial observation is close to a worst case: cheap to distort, unverifiable, and with distortion incentives strongest exactly when conditions are worst. Brain Capital Management VURA Working Paper Series 57 This is not hypothetical. Demers et al. (2026) show that it already happens in human capital disclosure: underperforming firms produce more boilerplate in more positive language. And the Illinois trial identified the deeper problem for any voluntary-response instrument: healthier and better-behaved employees select into wellbeing programmes and, by extension, into responding to wellbeing surveys, generating a positive correlation with performance for reasons unrelated to the construct (Jones et al., 2019). The framework's answers are structural rather than exhortative, and the paper concedes that none of them is currently required by any standard. Third-party collection with no managerial access to unit-level results below a minimum cell size. This addresses distortion at source by removing the observer. Census-style response-rate thresholds, with the response rate itself disclosed and a nonresponse penalty. A wellbeing figure derived from a forty per cent voluntary response is not a measure of the workforce; it is a measure of the forty per cent who chose to respond. Disclosing the response rate makes the selection problem visible; requiring a threshold constrains it. Tier 1 anchoring. Tier 2 self-reports are validated against administratively generated Tier 1 anchors that move in the opposite direction under distortion. A firm reporting rising wellbeing alongside rising certified sickness absence and rising regretted turnover has produced an internally inconsistent disclosure, which is detectable without access to the underlying responses. This is the framework's principal anti-gaming device and it is why Tier 1 exists. Trend and distribution rather than level. Cross-firm comparison of absolute levels invites arms races; within-firm trends and within-firm distributions are harder to manipulate and more decision-relevant. Implausibility flags on the response side. The obvious evasion is not distortion of the score but management of who answers. A firm can raise the reported figure by pressing for participation among the satisfied, or by pressing for participation generally while the dissatisfied abstain. Two cheap checks constrain this: a response rate simultaneously very high and rising alongside a rising score is flagged for explanation rather than credited, and unit-level results below a minimum cell size are suppressed rather than reported, so that pressure applied to a small team cannot move a disclosed number. Change limits. A year-on-year movement beyond a specified band requires narrative explanation reconciled to a structural change the firm can name. Improvement without an identifiable cause is treated as a measurement event, not a performance event. Before the concession, one distinction avoids a conclusion that would otherwise follow. If the anchors are what an external reader can trust, it might seem that Tier 2 adds nothing and the framework reduces to existing administrative data. It does not, because the two tiers serve different users. The Tier 1 anchors are lagging, low-specificity outcome proxies: certified sickness absence rises for many reasons and tells a manager nothing about which unit, which constraint or which cause. The Tier 2 instruments are leading and diagnostic: a distribution of team psychological safety identifies where nothing is sayable, and a lower-decile wellbeing value identifies who is strained, which is what an Brain Capital Management VURA Working Paper Series 58 intervention requires and an absence rate never supplies. Tier 2 is therefore first a management instrument and only derivatively an investor instrument, and it becomes the second only where the anchors corroborate it. That is a narrower claim than a disclosure framework usually makes for its own indicators, and it is the claim the gaming evidence permits. None of this solves the problem, and the paper should not pretend that the combination is sufficient. Third-party collection removes the immediate observer but not the employee's correct inference that aggregate results will reach management and may affect the unit; the referee's version of this objection is right. Suppression and flags raise the cost of manipulation without making it detectable in every case. Attestation under Tier 3b certifies procedure, not sincerity. The framework's actual defence is narrower than its list of devices suggests: the Tier 1 anchors are load-bearing and the Tier 2 instruments are not. Certified sickness absence, work-related ill health, disability leave incidence and regretted turnover are generated by systems that do not ask employees how they feel, and they are what makes an inconsistent disclosure detectable. A reader who distrusts a firm's Tier 2 figures loses little by reading the anchors instead. That is a weaker claim than the framework would like to make, and it is the one the evidence on gaming supports. Two things follow that stop this concession collapsing the framework into existing practice. First, divergence between the tiers is itself information: a firm reporting rising Tier 2 wellbeing against rising certified absence and rising regretted turnover has disclosed, involuntarily, that its measurement is corrupted, and an external reader can act on that without trusting either tier. Under existing regimes, where the state measures are simply absent, that signal cannot exist. Second, what is proposed here beyond ISO 30414 and conventional people analytics is not the instruments, which are decades old, but the discipline around them: the anchor set reported as a set, response rates disclosed with every figure, validity conditions travelling with dependent measures, tier separation with no aggregation, and the admission rules of Tier 3. The novelty claim rests on the architecture, and the architecture is exactly what the gaming problem makes necessary. Proposition 14 (Gaming degradation). A brain capital indicator that is self-reported under managerial observation and becomes a disclosure target will degrade in validity, and the degradation will be greatest precisely when the underlying state deteriorates. Anchoring to administratively generated measures that move oppositely under distortion is a necessary condition for the indicator to remain informative. Falsified by: a self-reported wellbeing or safety indicator that retains its correlation with administrative anchors after becoming a disclosure target, across firms and over time. 9.5 Operational definitions Appendix A gives the full indicator specification, and Appendix B a self-assessment protocol for firms that have not yet built measurement infrastructure. Three definitional choices warrant statement here because they differ from common practice. Belonging is operationalized as the proportion of teams falling below a threshold on the seven-item scale, not as the firm mean. The rationale is that the construct is a team Brain Capital Management VURA Working Paper Series 59 property and that the managerial question is where nothing is sayable, not what the average is. Base is operationalized as the lower-decile value of the workplace wellbeing module's stress and satisfaction items, together with the proportion below the WHO-5 cut-off, and paired with a mechanism disclosure of payment timing and hardship facility drawdown. The rationale is Proposition 6. Build has no Tier 2 instrument, for the reason given in Section 9.3, and it is two constructs rather than one. Its domain-task layer is operationalized as a triple, none of whose elements is a validated scale: the Tier 3a protected-practice quantity, the internal fill rate against external hire rate, and the unassisted work-sample trend where the role family admits one, reported with the Belonging distribution as its validity condition. Where a firm already has a hard continuously measured task outcome, that outcome measured under unassisted conditions dominates the work sample and should replace it. Its redefinition-relevant layer is not operationalized at all; the Tier 3b episode review is a process record standing in for a measure that does not exist. Build remains the least measurable of the three constructs, and Proposition 8 makes it potentially the most consequential; the framework does not resolve that tension and Section 11 records it as an acknowledged defect. 9.6 What would falsify the framework Table 9 collects the propositions and the observation that would falsify each. Beyond the individual propositions, three results would falsify the framework as a whole. If Belonging and Base indicators showed the multi-year lag predicted for stock changes, or Build indicators moved within a quarter, the stock–utilization decomposition of Definition 1 would be wrong and the framework's assignment of constructs to tiers would have to be rebuilt. If the Tier 2 state measures showed no incremental association with subsequent redefinition or value outcomes after conditioning on the Tier 1 anchors, the framework would be redundant: firms could simply report the anchors. If the Tier 3 mechanism disclosures converged to boilerplate at the same rate as policy disclosures have (Demers et al., 2026), the tier's rationale would fail. And if a measured brain capital state proved uninformative about redefinition capability—if firms with strong measured states showed no greater propensity to redefine themselves—the construct's link to the axiom in Section 3.5 would be broken and the framework would be measuring something real but strategically irrelevant. The single most valuable study that does not exist is a longitudinal measurement of unassisted cognitive capability at firm scale, across multiple years, in a population deploying AI at varying instructional-design intensities. Its minimum specification is a knowledge-work population of sufficient size to stratify by instructional design; a heldout unassisted task family with a rubric fixed before the first wave; blind scoring; at least three annual waves, since Card et al. (2018) put capability effects at a two- to three-year horizon; and pre-registration of the direction of the predicted effect, because Proposition 9 establishes that participants' own expectations are biased. That study would test Propositions 7 and 8 directly and would replace the interim protocol in Section 9.3 with Brain Capital Management VURA Working Paper Series 60 an instrument. No study in the literature reviewed here approaches it, and the paper's confidence in those two propositions is correspondingly bounded. P Claim Falsifying observation 1 State, not spend Programme expenditure predicts outcomes after conditioning on measured state 2 Transmission, not generation Belonging–creativity association exceeds Belonging–silence association 3 Belonging–accountability complementarity Monotonic return to Belonging invariant to accountability 4 Affordance dependence Belonging intervention succeeds where structural penalty for dissent is high 5 Material–informational asymmetry, Regime A only; Regime B stated as a hypothesis (§6.6) Informational intervention matches material one at equal cost in a liquidity-constrained workforce 6 Distributional concentration Base returns uniform across the strain distribution 7 Assisted–unassisted divergence (in domains with an observable unassisted outcome) Assisted gains reliably predict unassisted capability change 8 Design determines the sign of ΔK, the net capability change; no design is claimed to raise capability Capability change tracks adoption intensity irrespective of design; or a design is shown to drive ΔK positive in a workplace setting 9 Self-assessment bias Self-assessed capability change tracks objective measurement without directional bias 10 Appropriability is conditional on specificity General-skills investment yielding simultaneous productivity, wage and retention gains; or firm-specific capability returns competed away by external wage adjustment 11 Multiplicative level in two utilization fractions; Build in the stock's law of motion Gains from Belonging while Base is deficient or the converse; additive effects; utilization gains at the ceiling; or a contemporaneous Build effect within one period 12 Measurement asymmetry Mandated, assured disclosure of a validated cognitive-state measure in a major jurisdiction 13 Verification standard orders informativeness: Tier 3a > Tier 3b > policy Tier 3a no more predictive than policy; or Tier 3b indistinguishable from unattested narrative 14 Gaming degradation Self-reported indicator retains anchor correlation after becoming a target Table 9. Propositions and their falsifying observations. 9.7 An objection answered: is a framework that cannot measure its own core construct worth having? Section 9.3 established that the redefinition-relevant layer of Build has no measure, and Definition 1 indexes brain capital to redefinition. A reader is entitled to draw the obvious conclusion: that a measurement framework which cannot measure the quantity its own definition makes central has failed, and that the honest repair is to redefine brain capital as whatever the framework can measure—domain-task capability and the two utilization terms. This paper declines that repair, and the reason is the argument of the whole section. Redefining a construct to match the available instruments is precisely the failure the paper diagnoses in existing disclosure. Regulation S-K names headcount because Brain Capital Management VURA Working Paper Series 61 headcount is countable; the revised European standards retained inputs, costs, demographics and hazards and shed the rest; and the result documented in Table 7 is a reporting system in which what can be counted is reported and what determines enterprise value is not. A theory that repeated that move at the level of its own definitions would be committing the same error in miniature, and would purchase measurability by making the construct strategically uninteresting: brain capital reduced to domain-task capability and utilization is a productivity measure, and the paper would then be proposing a new name for something already approximated by existing human capital metrics. Naming a value-relevant asset that is not yet measured is a recognized form of contribution, and the intangibles literature is the precedent—but the analogy has a limit that should be stated at the outset. Corrado et al. (2009) could estimate the unmeasured stock because its investment flows were observable in dollars: research and development, software and organizational spending could be aggregated even though the capital they created appeared in no account. Lev (2001) worked from the same position. This paper’s position is weaker, because for the redefinition-relevant layer neither the stock nor the investment flow is observable in any unit. What survives the disanalogy is the form of the contribution, not its strength: both cases name an omitted asset and demonstrate the omission structurally, but Corrado et al. could bound their asset and this paper cannot. That difference is one reason the paper now describes itself as diagnostic and agendasetting rather than as a measurement contribution, and the title has been changed to say so. What the objection does establish, and what this paper concedes, is that the contribution must be described accurately. This section is not a measurement framework for brain capital. It is three things: a measurement architecture for the two constructs that can be measured; a specification of what an instrument for the third would have to do, together with the study that would build it; and a documented account of why the third resists measurement, which is itself the explanation for why capital markets cannot price the asset. An earlier version of this paper described Section 9 as a measurement framework without qualification, and the section heading and the contribution claims in Section 1.3 have been changed accordingly. 10 Discussion and Implications 10.1 For theory The paper's contribution to the brain capital literature is a change of level, and the change is not merely a matter of scale. Moving from the population to the firm makes brain capital a management variable: something a decision-maker can alter, that differs between organizations employing similar people, and that is bounded by identifiable structural conditions rather than by aggregate social determinants. The population literature's own capital-stock formulation invites this move (Ayadi et al., 2023) and its measurement instruments do not make it. Brain Capital Management VURA Working Paper Series 62 Three theoretical distinctions are offered as generalizable beyond this construct. The state–programme distinction resolves a contradiction that recurs across the organizational wellbeing literature and is not confined to brain capital: the coexistence of value-relevant levels and null-effect interventions is a general pattern, and the resolution —that measuring a state and purchasing an intervention are different acts—applies wherever it appears. The material–informational asymmetry is a specific and testable claim about which class of employer action alters cognitive availability, and it predicts the failure of a large category of current practice. The assisted–unassisted distinction is the most immediately consequential, because the entire corporate discourse on AI and capability currently conflates two quantities that the evidence shows to move independently. The series-level contribution is a completed chain. Future Value Theory supplies the object—the future value a firm exists to create. Enterprise Redefinition supplies the capability that creates it. From Job Description to Purpose Description supplies the role design through which that capability becomes someone's actual work. This paper supplies the substrate all three assume, and finds it unmeasured. The order matters for how the series should be read: the earlier volumes are prescriptive and this one is diagnostic, because a prescription addressed to a capacity no firm reports is a prescription no firm can be held to. The paper also contributes to the human capital literature a specific claim about appropriability under AI. Section 7.5 argued that a firm captures the return to capability investment in proportion to the frictions preventing wage and mobility adjustment. If AI compresses the wage structure by raising the productivity of the less experienced more than the more experienced—which is the consistent finding across Brynjolfsson et al. (2025), Noy and Zhang (2023) and Cui et al. (2026)—then by the Acemoglu–Pischke mechanism it should increase firm-financed general training. That is a prediction about the direction of corporate training investment under AI diffusion, and it is checkable. Finally, the paper closes a loop in this series. The case analysis that preceded it identified an observational asymmetry between the dimensions of enterprise redefinition capital markets can price and the dimensions that constrain redefinition (Kadowaki, 2026d). Proposition 12 supplies the cause, documented against the current regulatory perimeter. This is worth noting because it changes the status of that earlier finding from a limitation of research design to a property of the reporting system—which makes it actionable by standard-setters rather than merely regrettable. 10.2 For management practice Four implications follow, and they are more restrictive than what the practitioner literature on this subject usually recommends. Stop buying programmes; start measuring states. The randomized evidence on individual-level wellbeing interventions is close to null across three independent designs (Song & Baicker, 2019, 2021; Jones et al., 2019; Fleming, 2024). A firm that reallocates programme budget to third-party census measurement of the states in Section 9.3, and to Brain Capital Management VURA Working Paper Series 63 the structural changes those measurements identify, is following the evidence rather than the market. Look in the lower decile, then determine which Base regime it is in. Proposition 6 implies that a firm reading its workforce averages is looking in the wrong place; the recoverable return on Base is concentrated in the tail. But the lever depends on what is binding in that tail, and Section 6.6 separates two regimes. If the lower decile cannot meet an unexpected expense without borrowing, the interventions with randomized support are payment timing and frequency, liquidity provision and obligation discharge—cheap relative to programme spending, almost never used, and in the one randomized test worth a 6.9% output effect (Kaur et al., 2025). Wage payment frequency is a treasury decision most firms have never examined as a cognitive-capability variable. If the lower decile is not liquidity-constrained, none of that applies: the binding claims are caregiving, schedule fragmentation and work design, and the levers are job design and genuine absence coverage rather than timing. These are hypothesis-based guidance and are labelled as such—Section 6.6 states the hypothesis and its test—and a firm adopting them is acting on the paper's mechanism, not on randomized evidence, and should instrument the change so that its own data can bear on the hypothesis. Pair psychological safety with accountability, and measure silence rather than satisfaction. Eldor et al. (2023) falsify the monotonic formulation, and Walton et al. (2023) show that intervention without structural affordance produces nothing. The practical question is not whether employees report feeling safe but whether dissent has a route that does not pass through the person it concerns, and whether that route is used. That is a countable quantity. Treat AI instructional design as a capability decision, not a procurement decision— and treat role definition as part of that design. Proposition 8 is the paper's most actionable claim. The variable that determines whether AI deployment preserves or consumes the firm's cognitive stock is the design of the deployment—guardrails, tutoring structure, protected unassisted practice—and not the licence count. Firms currently measure the latter and not the former. The reallocation conjecture in Section 7.4 extends this one level up, and a firm should weigh it as a conjecture rather than a finding: if a role is defined as a set of tasks, a tool that performs those tasks leaves nothing that exercises capability, whereas a role defined by the question it exists to pose, the principles by which it judges, the exceptions it must catch and the authority to stop retains the demanding portion by construction (Kadowaki, 2026c). The practical question is not only how the tool is configured but what the person is now for. Fund the floor before the development. Section 6.6 gives the ordering, and it is the reverse of the usual one. Capability investment offered to a workforce whose lower decile is materially strained is being offered to people who cannot take it up, because deliberate practice imposes a present cognitive cost for a deferred return. A firm that has a training budget and an unmet wage floor in the same population has sequenced its spending wrongly on the argument given here. Brain Capital Management VURA Working Paper Series 64 A firm should also be aware that its own people's reports of capability gain are not evidence: in the one study measuring both, experienced developers were nineteen per cent slower and believed they were twenty per cent faster (Becker et al., 2025). 10.3 For investors Three points follow for capital allocators, and the first is a warning rather than an opportunity. The asset is currently unpriceable from public information, and that is a stable condition. Proposition 12 documents that no mandated indicator measures workforce cognitive state, that the European direction of travel is toward fewer datapoints, and that the international standard-setter remains at research stage with an undated next milestone. An investor cannot at present distinguish a firm with high brain capital from one with a well-written human capital section, and Demers et al. (2026) show that the correlation between disclosure tone and financial performance runs the wrong way. Mechanism disclosures are the available signal. Proposition 13 identifies the class of disclosure that is costly to fabricate: payment timing policy, hardship facility drawdown, the existence and usage of a dissent route, AI instructional design, internal fill rate against external hire rate, and the structural arrangement that protects long-horizon cognitive work. Most of these are already known to the firm and disclosed by almost none. An investor asking for them is asking for something cheap to supply and difficult to fake, which is the definition of a useful disclosure request. Disclosing a time-securing mechanism is a two-sided signal, and the paper has no model of which side dominates. Section 7.4 argues that a firm cannot sustain redefinitionrelevant capability without structurally defending the time it requires, and Section 9.3 asks firms to disclose the arrangement that does so. The obvious objection is a capitalmarket one and it has not been answered anywhere in this paper: a firm announcing that it holds and protects slack invites the judgement that its capital is inefficiently deployed, and a marginal investor optimizing quarterly earnings or return on equity should discount it. The objection is serious enough to state as an open problem rather than to argue away, but three considerations bear on it. The first is that the mechanisms the case evidence identifies are not discretionary slack. A dual-class structure, a legal form, a contractual commitment from investors, a synchronised commitment from a counterparty and prepayment with release are all commitment devices: they are costly to establish and costly to reverse, which is what makes them credible and what distinguishes them from an unspent budget line. Disclosure of a commitment device is therefore closer to disclosure of a covenant than to disclosure of idle capital. The second is that the asset the device protects is already priced with a premium for exactly the reason it is hard to hold: organization capital earns higher average returns because it is embodied in people who can leave (Eisfeldt & Papanikolaou, 2013).