A Periodic Table for Teamwork? Mapping Human Capabilities

Scope and problem — research methods

1.1 The question and its boundaries

The question is whether a common map can improve the description and development of capability in technology teams. The relevant outcomes include identifying a learning need, distinguishing an individual limitation from a coordination failure, and selecting an intervention that improves a defined task outcome. This paper does not infer a person’s worth from performance, identify fixed human types, or supply a scoring instrument.

Three units must remain distinct: the person, the team, and the observed episode of work. A team member’s report is an observation made by a person; its referent might nevertheless be the team. A performance outcome belongs to a task under particular conditions. Tools, workload, authority, incentives, and access to information are contextual variables unless the study explicitly defines a different role for them. These distinctions are proposed design rules, not measured findings from this paper.

1.2 Source and review procedure

The live source webpage and its linked eight-page PDF were retrieved and read in full on 12 September 2026 UTC, during 11 September in New York. Both present a conceptual note; neither reports a validation study. The archive’s earlier author-declared date is qualified there as recollection; it does not date this expanded edition or establish when its present wording originated. The retrieved source copies are linked in R1 and preserved with this edition R1: Mumford, source note.

This is a targeted critical synthesis, not a systematic review or meta-analysis. Searches focused on the constructs named by the note, existing occupational classifications, team-level measurement, contrary findings, and validity methodology. Public primary research, author-hosted manuscripts, official framework documentation, and the original testing standards were preferred. Sources were selected for relevance to an explicit claim, including evidence that challenged it. Abstract-only readings are identified in the evidence ledger; details unavailable at that reading depth are not reconstructed.

The procedure comprised: inventorying the source claims; distinguishing descriptive categories from empirical assertions; checking relevant sources; testing the logic with definitions and counterexamples; and translating unresolved claims into observable predictions. No effect sizes were pooled, no raw study data were reanalyzed, and no original instrument was administered. The source note’s references have not been “recovered”: the references below were selected for this expansion and must not be presented as its original bibliography.

1.3 Claim labels

Established result denotes an explicit mathematical consequence or a methodological principle with its stated scope. Observation denotes a result reported in an identified study or directly visible in the source. Inference denotes a reasoned interpretation that goes beyond an observation. Disputed claim identifies a substantive claim with competing interpretations or findings. Hypothesis is a claim awaiting a discriminating test. Proposed work has not been performed. These labels do not assign probabilities to the truth of a claim.

Registers use stable identifiers: assumptions A; evidence E; unresolved questions Q; hypotheses H; predictions P; experiments X; decision rules D; and mathematical propositions T. Appendix F connects every main hypothesis through the complete chain.

Terminology — foundations

2.1 Five different accomplishments

A vocabulary supplies useful words. A taxonomy defines membership and boundaries. A measurement system turns observations into interpretable values. A predictive model estimates a specified outcome on new cases. An intervention changes an outcome when used. A coherent vocabulary may fail as a measurement system; a predictive score may offer no effective intervention. These are logically different accomplishments. They require different evidence.

For example, calling an incident “poor self-regulation” can help frame a discussion. To measure it, a study must specify the behavior, occasion, instrument, and interpretation. To predict later performance, the measure must be tested on later or independent cases. To improve performance, an intervention must be compared with a suitable control. The descriptive label alone answers none of the latter questions.

2.2 The proposal being assessed

The source proposes: knowledge and expertise; skills; abilities or aptitudes; traits and dispositions; values and attitudes; metacognitive capabilities; emotional intelligence; and team-level capabilities R1: Mumford, source note. These eight headings are retained as the object of study, not accepted as eight independently measured dimensions.

The right-hand column is an inference from the definitions. It is not a measured overlap matrix. The taxonomy presently mixes content, characteristic mode of behavior, and level of analysis; a final classification would have to explain why these belong on one axis.

2.3 Relevant foundations in existing work

O*NET already separates occupational descriptors such as knowledge, skills, abilities, work styles, activities, and context. It provides a concrete comparison framework, not evidence that any occupational framework enumerates all human capability R2: O*NET content model.

Personality research offers empirical structures for patterns of individual differences. Goldberg’s lexical studies supported a broad five-factor representation in the materials studied R3: Goldberg (1990). That result does not make five factors an exhaustive account of competence, or imply that all cultures and instruments recover the same structure. The latter boundary matters because Gurven and colleagues reported difficulty recovering the standard Big Five structure in a Tsimane sample R4: Gurven et al. (2013).

Metacognition concerns cognition about cognition and its regulation. Flavell’s foundational account places metacognitive knowledge within knowledge and distinguishes monitoring-related processes R5: Flavell (1979). Emotional-intelligence research supplies more than one model: MacCann and colleagues investigated ability emotional intelligence in a hierarchy of cognitive abilities, whereas Petrides and Furnham investigated trait emotional intelligence in relation to personality R6: MacCann et al. (2014), R7: Petrides and Furnham (2001). These sources make terminology more precise; they do not settle one universal classification.

At team level, shared mental models concern members’ representations of the task or team; transactive memory concerns the organization of distributed expertise. Psychological safety concerns a shared belief about interpersonal risk, not technical proficiency. Collective intelligence research asks whether group performance across tasks has a common factor. These are distinct research questions, not interchangeable names for a single team essence R8: Mathieu et al. (2000), R9: Lewis (2003), R10: Woolley et al. (2010), R11: Bates and Gupta (2017), R12: Edmondson (1999).

Multilevel theory distinguishes properties formed through similar member contributions from properties formed through configurations of different contributions. It supplies an established methodological framework for specifying the entity and aggregation being studied; it is not empirical validation of a new team score R15: Kozlowski and Klein (2000).

2.4 Interpretation and intended use

The testing standards locate validity in the evidence for a score interpretation and its proposed use. A measure supported for one interpretation does not automatically support every other use R13: Testing Standards (2014). Accordingly, this paper separates exploratory research coding from individual assessment and employment decisions. No score or individual decision rule is supplied here.

Proposed model — derivation and assumption map

3.1 What has, and has not, been derived

Observation: The accessible note gives category names and examples but no procedure that uniquely generates exactly eight categories R1: Mumford, source note. Inference: The present taxonomy is an editorial synthesis, not the output of a demonstrated mathematical derivation. The number eight therefore has no established special status.

A defensible derivation would begin with a declared universe: which roles, tasks, kinds of capability claim, and time periods are included? It would then specify what is being classified. A sentence describing a behavior is not the same kind of object as a psychological construct, a test item, or a person. Rules would determine membership, allow an unresolved category, and be tested on material not used to construct the rules.

The required chain is:

Assumption → evidence → unresolved question → hypothesis → observable prediction → experiment → decision rule.

For instance, assuming that categories have usable boundaries (A2) meets both the source’s unresolved examples (E0) and established accounts of cross-cutting constructs (E4–E6). The resulting question is whether independent researchers can classify new cases consistently (Q1). H1 predicts reproducible coding; P1 specifies agreement and unresolved-case rates; X1 compares frozen coding systems; D1 states when to retain, revise, or reject a coding claim. No link in that chain is replaced by confidence in the metaphor.

3.2 Three candidate representations

The original eight-heading taxonomy is the first candidate. Its strongest version requires a membership rule under which each relevant unit has one category. A weaker version allows overlap; that weaker version must stop claiming strict mutual exclusivity.

A faceted representation is a proposed competitor. It records separate attributes: level, content, typical versus maximum performance, temporal stability, and evidence method. An observed correction of a reasoning error could concern an individual, cognition, demonstrated performance, a specified episode, and an observed task response. Several facets need not be collapsed into one “element.” This may clarify boundaries but could also impose greater coding effort; its superiority is a hypothesis.

A plural collection of existing measures is another competitor. A project could use a relevant work sample, a specific team measure, and contextual observations without any unified taxonomy. This sacrifices one common map and may avoid forcing incompatible constructs into a common score. The tradeoff must be tested through task usefulness, accuracy, and cost.

Normative invariants

3.3 Assumptions at the main transitions

Moving from vocabulary to classification assumes a bounded universe and operational boundaries (A1–A2). Moving to measurement assumes observations support the intended construct and level (A3–A4). Moving to prediction assumes the new representation adds value beyond baselines (A5). Moving to development assumes an intervention can improve a specified outcome (A6). Moving across settings assumes sufficient comparability or justified adaptation (A7). Calling the result periodic adds a further assumption: an ordering relation predicts recurring structure (A8). Appendix A states the evidence and failure consequences for each.

Worked examples — mathematical audit

The source offers no equations or theorems to verify. This section supplies elementary audit propositions, proved below, to expose the requirements of stronger claims. They are not novel empirical findings or a mathematical derivation of the proposed categories. Appendix B records their domains and limits.

4.1 T1: A partition requires exactly one membership

Let U be the declared universe of units and C₁,…,C₈ be subsets of U. Define m(x) as the number of categories containing x. The family is disjoint and exhaustive precisely when m(x)=1 for every x in U. If a partition is defined to exclude empty categories, also require every Cᵢ to be nonempty.

Proof. Disjointness permits at most one membership and exhaustiveness requires at least one. Conversely, exactly one membership leaves no uncovered unit and permits no shared member. This establishes the equivalence.

Limit. Assigning a primary label by fiat can meet the formal condition without establishing meaningful psychological boundaries. Likewise, two distinct constructs can correlate without belonging to the same category. Category overlap, correlation between scores, and statistical dependence are different issues.

4.2 T2: Factor fit alone does not identify unique categories

For a factor model with covariance Σ = ΛΦΛᵀ + Ψ, let T be an invertible transformation with compatible dimensions. Set Λ′ = ΛT and Φ′ = T⁻¹ΦT⁻ᵀ. Then Λ′Φ′(Λ′)ᵀ + Ψ = ΛΦΛᵀ + Ψ by cancellation. The transformed parameters must remain within the allowed model. For example, if Φ is fixed to the identity matrix, orthogonal rotations preserve that constraint, while arbitrary invertible transformations generally do not.

Established result: These transformed parameters reproduce the same covariance. Therefore unrestricted fit alone cannot choose a unique semantic interpretation of the factors. Confirmatory constraints can restrict or remove rotational freedom; their substantive justification is an additional matter.

Limit. A category of constructs is not automatically a reflective latent factor. A reflective model treats a latent variable as generating its indicators. A constructed index instead combines components by a specified rule. A single factor analysis of every heading is inappropriate until the measurement interpretation and the person/team levels have been justified.

4.3 T3: Finite successful coverage is not universal completeness

Suppose a classification and gap-detection rule are frozen, and each independent draw from a fixed target distribution has probability p of exposing an uncovered case. For n≥1 draws, the probability of observing zero uncovered cases is (1−p)ⁿ. For finite n and 0<p<1, that probability is positive. Zero observed gaps therefore does not prove p=0.

For a prespecified one-sided error level α, inversion at zero observed gaps gives an upper confidence bound pᵤ = 1−α^(1/n). This is a sampling statement under the model, not a probability that the taxonomy is true. It does not apply unchanged to a handpicked collection, dependent examples, missed detection of gaps, or a changing target population. No numerical bound is reported because no such sample was collected.

4.4 T4: Individual averages need not determine team outcomes

Consider an idealized task in which two equally capable participants independently receive random binary digits a and b, each equally likely to be zero or one. The first participant must report whether the digits differ. With communication, the second can transmit b and perfect performance is possible. Without any information channel, the first can do no better than guessing: its success probability is at most one half. The multiset of individual capability scores and its average are unchanged, but achievable task performance differs.

Established logical result: An average of those skill values is insufficient to determine achievable performance in this task. A team model can therefore require interaction and context: Y = F(individual attributes, relationships, task, conditions).

Limit. This counterexample refutes universal sufficiency of the average; it does not establish that averages are always poor predictors, or that any particular team construct causes performance. Those are empirical questions.

4.5 T5: Repetition is not independent corroboration

Let X be a random observation with finite variance and let X₁,…,Xₖ be exact copies, so every Xᵢ=X. Their unweighted sum S=kX, and Var(S)=k²Var(X). For a realized value x, the realized sum is kx. Relabeling a record does not create new measurements.

Implication: A system that treats those k entries as independent evidence would misstate the information available. Deliberate weighting or averaging can be legitimate if declared; the problem is covert duplication and an incorrect independence assumption. A coding record should therefore retain the identity of its source observation even when it legitimately receives multiple tags.

4.6 Audit verdict

The eight-heading proposal has neither a demonstrated partition nor a validated measurement model. No ordering variable or recurrence rule has been specified that would make “periodic” a testable structural claim. A useful visual arrangement would remain possible without such a law. The present mathematical contribution is to delimit these claims, not to establish their truth.

Empirical evidence

5.1 What is directly available for this proposal

Observation: The source provides no validation dataset, classification study, psychometric instrument, scoring model, or comparative outcome experiment. Consequently, this edition reports no measured reliability, predictive improvement, fairness result, or causal benefit for the eight-heading proposal R1: Mumford, source note. That absence is a limit on what can be concluded; it is not an empirical demonstration of failure.

5.2 Individual constructs

Observations: Goldberg’s studies supported a broad five-factor lexical structure R3: Goldberg (1990). Gurven and colleagues did not recover a comparably stable standard Big Five pattern in their Tsimane research, despite examining several possible methodological explanations R4: Gurven et al. (2013). Together these studies warrant attention to population and method, not either blanket universality or blanket rejection of personality measurement.

MacCann and colleagues reported evidence for ability emotional intelligence as a second-stratum factor in a hierarchical intelligence model R6: MacCann et al. (2014). Petrides and Furnham’s investigation instead addressed trait emotional intelligence and its relation to established personality factors R7: Petrides and Furnham (2001). Inference: A heading called emotional intelligence can conceal distinct measurement traditions. Evidence for one cannot be silently transferred to the other. These findings do not demonstrate that emotional intelligence belongs beside knowledge, skills, and personality as an exclusive peer category.

A conceptual replication by Evans and colleagues, using alternative measures in 830 participants, also supported the placement of ability emotional intelligence within cognitive ability. That support was qualified: oblique and hierarchical models remained plausible, a hierarchical solution required a variance constraint to address an inadmissible estimate, and bifactor models did not converge. Limited cognitive-content coverage and weak reliability in some emotional measures restrict the inference R16: Evans et al. (2020). This adds component evidence without uniquely identifying one hierarchy or validating the proposed taxonomy.

Flavell’s account supplies a conceptual explanation for why monitoring, knowledge about cognition, and regulation overlap with other descriptions R5: Flavell (1979). Limit: It is foundational theory, not a validation study of this eight-heading system. This review has not established equivalent coverage for values, attitudes, expertise, or every type of aptitude.

5.3 Team constructs

Observations: Mathieu and colleagues studied 56 undergraduate dyads in flight simulation and reported positive relationships between shared mental models, subsequent team processes, and performance R8: Mathieu et al. (2000). Lewis tested a transactive-memory scale in 124 laboratory teams, 64 student consulting teams, and 27 technology-company teams, reporting several forms of validity evidence R9: Lewis (2003). Edmondson’s study of 51 manufacturing work teams associated psychological safety with learning behavior; the paper explicitly states that its cross-sectional survey cannot establish causality R12: Edmondson (1999). Each offers evidence for a specific operationalization in a bounded setting.

These studies make a team-level research program plausible. They do not establish one universal “team capability” score. Nor do associations between reported climate and performance identify every direction of causation. Successful teams might develop trust or safety; working conditions might influence both reports and outcomes. A new study must test such alternatives; a correlation does not settle them.

5.4 A disputed claim: collective intelligence

Woolley and colleagues studied 699 people in groups of two to five and reported a collective-intelligence factor across tasks, with patterns not simply captured by average or maximum individual intelligence R10: Woolley et al. (2010). Bates and Gupta’s three studies with 312 people instead attributed much of group performance to individual intelligence, challenging a substantially independent group-capacity interpretation R11: Bates and Gupta (2017).

Disputed claim: How much a broad group-performance factor adds beyond individual cognitive ability and task composition remains model- and design-sensitive in this selected evidence. The studies do not supply a direct verdict on the proposed taxonomy. Competing group models should be preregistered on matched tasks, with the same standards applied to supportive and contrary results.

5.5 Overall evidentiary inference

The literature supports investigating multiple individual and collective constructs with explicit operational definitions. It does not license assembling those constructs into a validated new instrument without new evidence. The central missing link is incremental usefulness of this particular representation. Appendices C and F distinguish evidence for ingredients, logical constraints, and the absent tests of the combined proposal.

Competing models — hypotheses and apparent cracks

These explanations show that the apparent cracks need not signal a newly discovered failure of psychology. They do not establish the cause of any particular workplace incident.

The substantive alternatives are:

• H1 — Reproducible classification: Frozen definitions support consistent coding of new cases. Competitor: the vocabulary is a helpful discussion aid whose boundaries remain irreducibly ambiguous.
• H2 — Defensible measurement distinctions: Selected operationalized constructs remain distinguishable when measured through different methods. Competitor: fewer constructs, cross-loadings, or method effects explain the evidence adequately.
• H3 — Incremental prediction: The representation improves prediction or decision usefulness beyond comparably resourced baselines. Competitor: a simpler work-sample, occupational, or team-process model performs as well.
• H4 — Developmental usefulness: A prespecified use of the map leads to better outcomes through intervention. Competitor: equal attention, training time, or resources provide the same benefit without the map.
• H5 — Bounded transportability: A frozen measurement interpretation generalizes to prespecified settings, or documented adaptations preserve the target meaning. Competitor: useful validity is local and does not transport.
• H6 — Predictive periodicity: A separately specified ordering and recurrence rule predicts withheld structure better than appropriate alternatives. Competitor: the table is a visual metaphor with no periodic predictive content.

These hypotheses are not mutually exclusive. A vocabulary could pass H1 and fail H3; a local intervention could pass H4 while H5 fails. Results should change the relevant claim, not trigger an all-or-nothing verdict on human capability.

Falsification criteria — experimental proposals and decision rules

All studies in this section are proposed work. They are sufficiently specified to identify what would distinguish the claims, but they are not registration-ready protocols. Instruments, recruitment sites, sample sizes, and numerical practical thresholds remain to be selected. This is an explicit operational gap, not permission to choose thresholds after results are visible.

7.1 Common design commitments

Before confirmatory collection, freeze the target population, unit, construct definitions, primary outcome, missing-data treatment, exclusions, analysis, and decision thresholds. Use a separate pilot to estimate feasible recruitment and measurement variability. Determine sample size using a justified smallest useful effect, clustering, anticipated attrition, and simulation where needed. The number of people is not the number of independent teams.

Separate development data from evaluation data by the relevant unit, including teams and organizations. Fit preprocessing and choose models only inside training data. Keep outcome assessors unaware of treatment or model allocation where feasible. Record contradictory and adverse results. Distinguish exploratory revisions from confirmatory tests of the frozen version.

Let L(z) and U(z) denote lower and upper uncertainty bounds from the prespecified inferential procedure. Thresholds such as a_min, g_max, and δ_min below are practical decisions requiring justification before the confirmatory sample. They are not universal constants. Uncertainty intervals represent sampling uncertainty under an analysis model; they are not confidence percentages assigned to the paper.

7.2 X1: Can researchers classify new cases usefully?

Question Q1; assumption A2; hypothesis H1; prediction P1. Independent coders applying frozen definitions will produce reproducible labels, with a low unresolved-case rate, on cases not used to write the definitions.

Construct a representative evaluation sample from declared task domains and a separate, deliberately difficult challenge set. Report their denominators separately; do not treat the challenge-set rate as a population estimate. Compare the eight-heading system, a faceted system, and a simpler existing coding frame. Balance training time and instructions; randomize case order and coder allocation. Distinguish uncovered cases, ambiguous membership, and insufficient information instead of merging them into one explanation. Preserve each coder’s initial decision before adjudication. Coders necessarily know their assigned manual; practical-task outcome judges should be unaware of which system produced an answer.

Report agreement with uncertainty, category prevalence, confusion patterns, uncovered or multiply eligible cases, coding time, and performance on a separately judged practical task such as identifying the appropriate kind of follow-up evidence. Assess agreement within comparable fields; do not mechanically compare one nominal-category coefficient with an incompatible multilabel statistic. Evaluate representational alternatives on the common practical task.

D1. Retain a bounded coding claim only if L(agreement)>a_min and U(unresolved rate)<g_max on the representative evaluation sample, with thresholds fixed in advance. Demonstrate inadequacy if U(agreement)<a_min or L(unresolved rate)>g_max; otherwise report INCONCLUSIVE where the evidence does not decide. Define the unresolved rate as the union of the three reported failure types. Support added practical usefulness only when L(task utility gain)>δ_utility against the prespecified baseline, including the chosen time or cost measure; U(gain)<δ_utility rules out that useful advantage, and intermediate results remain INCONCLUSIVE. Any asserted universal partition is defeated by a verified in-scope uncovered or multiply eligible case under frozen membership rules; insufficient case information alone is not such a counterexample. A revised rule constitutes a new version to retest. Reproducibility can pass while added usefulness does not.

7.3 X2: Do the measurements represent distinguishable constructs?

Question Q2; assumptions A3–A4; hypothesis H2; prediction P2. Prespecified measures show interpretable distinctions across methods and explain independent observations better than plausible collapsed or method-dominated alternatives.

First select specific subconstructs, not entire headings as though each were one trait. Use appropriate work tasks, reports, or structured observations from more than one method where feasible. Separate person-level and team-level models, and justify any aggregation with an explicit referent and aggregation model. Compare theoretically constrained models, including cross-loading and method-effect alternatives. Evaluate stability in held-out data and perform model diagnostics.

D2. Retain a measurement interpretation only if its prespecified loading or component pattern, reliability requirements, diagnostics, and external relations meet declared criteria. If a simpler model has practically equivalent held-out performance and better interpretability or cost, prefer it. Distinguish demonstrated criterion failure from inadequate precision. Good covariance fit alone does not establish eight natural kinds; T2 illustrates one reason that inference is unwarranted. Without valid reflective assumptions, use an appropriate index evaluation instead of factor-model claims.

7.4 X3: Does the map add predictive value?

Question Q3; assumption A5; hypothesis H3; prediction P3. On held-out teams or sites, the expanded representation improves a prespecified outcome prediction over simpler, comparably resourced baselines.

Choose one primary outcome with an independently defined scoring procedure—for example, quality on a standardized collaborative task. Compare task-relevant work samples and a lean occupational model; add a specific team-process baseline where appropriate. Give alternatives comparable information, tuning opportunities, and measurement budgets. Evaluate a frozen model on new sites or later periods, with uncertainty clustered at the proper unit. Include collection time and cost in practical utility.

Let Δ = baseline loss − candidate loss, so positive values favor the candidate. D3. Support incremental prediction only if L(Δ)>δ_min on the prespecified external evaluation. If U(Δ)<δ_min, rule out the specified useful improvement under the tested conditions. Otherwise classify the result as INCONCLUSIVE. A favorable training fit is not evidence for P3; a predictive improvement is not evidence for H4.

7.5 X4: Does using the map improve development outcomes?

Question Q4; assumption A6; hypothesis H4; prediction P4. Teams assigned to a frozen map-guided development procedure improve the primary outcome more than teams receiving an active alternative with equivalent time and resources.

Specify the decision procedure linking observed needs to an intervention before allocation. Randomize at team level if within-team spillover is likely. Use an active comparator, baseline measurement, blinded outcome scoring where possible, and a defined follow-up period. Estimate the intention-to-treat effect; document adherence and contamination without excluding inconvenient participants after allocation. Measure costs and prespecified undesirable effects.

D4. Support practical developmental benefit only if L(treatment effect)>δ_min and the prespecified harm and burden criteria are satisfied. If U(treatment effect)<δ_min, rule out the specified minimum useful benefit under the tested conditions. Otherwise report INCONCLUSIVE unless a prespecified harm criterion independently rules out adoption. Improved outcomes establish an effect of the intervention package; they do not by themselves establish the proposed psychological mechanism.

7.6 X5: Do interpretations travel across settings and time?

Question Q5; assumption A7; hypothesis H5; prediction P5. The measurement and prediction claims meet prespecified comparability and performance criteria in the target groups, languages, work settings, and later occasions.

Use cognitive interviews and task review to identify different interpretations before quantitative comparison. For applicable latent-variable scales, examine relevant measurement invariance and item behavior; for work samples or constructed indices, evaluate their own comparability and outcome relationships. Estimate subgroup errors with uncertainty, including important intersections where sample sizes permit. Test later observations separately to distinguish drift from initial fit. A lack of detected difference in a small group is not proof of equivalence R14: Putnick and Bornstein (2016).

D5. Restrict any transported claim to groups and contexts that meet the prespecified requirements. If equivalence fails, revise, localize, or suspend the affected interpretation; do not compare unqualified scores. If sample sizes cannot distinguish a meaningful failure from equivalence, report NOT TESTED or INCONCLUSIVE as appropriate. Measurement comparability alone is insufficient to establish fair consequences.

7.7 X6: What would test actual periodicity?

Question Q6; assumption A8; hypothesis H6; prediction P6. A frozen ordering and recurrence law predicts specified properties of withheld constructs or combinations beyond nonperiodic alternatives.

The prerequisite is currently missing: the proposal must define what is ordered, what repeats, the predicted property, and why a given location implies it. Merely locating similar words near each other cannot count. After that prerequisite, choose withheld targets before fitting and compare prediction against semantic-similarity, nonperiodic latent-structure, and simple baseline models. Use nested evaluation to prevent selecting the recurrence after inspecting the answers.

D6. Require a preregistered predictive rule, external comparison, and independent replication. Support the specified advantage when L(predictive gain)>δ_periodic; rule out that useful advantage when U(gain)<δ_periodic; otherwise report INCONCLUSIVE. Apply this distinction to replication as well. A supported result establishes only the bounded relation, not a universal periodic law of humanity. Until a rule exists, H6 is underspecified and X6 cannot enter confirmatory testing.

Adoption and migration

7.8 Research conduct

Any participant study needs a suitable consent and ethics process, data minimization, restricted access, and a clear withdrawal and retention policy. Research coding should not silently become an employment evaluation. Collect sensitive attributes only when justified by an approved design and needed for the stated analysis. No such study or approval is claimed here. These are proposed conditions for the research program, not a report of completed governance.

Limitations

This review is selective and primarily English-language. It does not enumerate all literature, quantify publication bias, adjudicate the full replication record, or establish consensus through systematic synthesis. Several studies were inspected at abstract level; their raw data, instruments, exclusions, and detailed analyses were not independently audited. The evidence ledger records these limits.

The original eight headings lack frozen definitions, a construct inventory, an operational instrument, and an enumerated target universe. The faceted alternative introduced here has not been validated either. Proposed outcome choices may favor one representation unless comparison tasks and information budgets are carefully balanced. Technology-team relevance cannot be assumed from student or manufacturing samples alone.

Values, attitudes, domain-specific expertise, physical capability, disability and accessibility, non-Western construct traditions, and institutional opportunity receive incomplete coverage. Human–AI teams are not evaluated. No cost-benefit result, employment validity, causal benefit, universal fairness, or cross-cultural equivalence has been established for this proposal.

The mathematical statements expose conditions and invalid inferences. They do not measure people or replace empirical work. The experiments require substantial operational design before registration, and the periodicity experiment requires a testable theory that does not yet exist. Those are stated gaps in the research program, not unfinished sections of this manuscript.

Conclusion

What is well supported, and within which limits?

Specific individual-difference and team constructs have empirical support under particular instruments, samples, and tasks. Existing occupational and psychological frameworks offer usable starting distinctions. It is logically necessary to distinguish category membership, measurement, prediction, and causal benefit. None of that establishes the eight-heading proposal as a complete or predictive system. Its present status is a provisional research vocabulary whose additional usefulness remains unmeasured.

What remains uncertain?

The best boundaries, completeness within a defined domain, reliability of coding, defensible measurement interpretations, improvement over simpler models, intervention benefit, and transportability remain uncertain. No evidence establishes a privileged number of categories or a periodic ordering law. Even the proposed faceted alternative must earn its usefulness through comparison.

Which apparent cracks already have established explanations?

Metacognition can involve knowledge and learned regulation; emotional intelligence can refer to different ability and trait constructs; and group performance can depend on interaction and task conditions. Correlated measures need not violate taxonomic exclusivity, and a repeated observation is not repeated independent evidence. Such issues can arise through ordinary definitional, measurement, and multilevel mechanisms. They do not by themselves reveal a new scientific anomaly or identify the cause of a particular failure.

Which unresolved questions merit experimental attention?

Begin with coding feasibility and measurement checks relevant to the intended use. Pursue predictive, intervention, and transport studies according to the claim: a useful intervention package need not validate the taxonomy or its mechanism, and transportability belongs in early study design. Literal periodicity merits a confirmatory experiment only after a recurrence rule makes distinct, falsifiable predictions. A visual metaphor without that rule is not yet an experimental hypothesis.

What evidence would change the assessment?

The assessment would improve with independent, well-designed evidence of reproducible classification, interpretable measures, useful held-out prediction, and better outcomes from a controlled map-guided intervention, followed by replication in the claimed settings. It would worsen with stable ambiguity under frozen definitions, redundant measurement, practical equivalence or inferiority to simpler baselines, failed transport, or unacceptable consequences. A null result with wide uncertainty would leave the claim unresolved. The deciding evidence must address the particular claim; evidence for a component cannot stand in for evidence for the whole.

References

R1. Mumford, E. C. A Periodic Table for Teamwork? Mapping Human Capabilities: A provisional map of individual strengths and collective capability. Working paper, author-declared date 1 November 2024; current reader edition accessed 12 September 2026 UTC. Source webpage and linked PDF. The date is qualified by the archive and is not independent provenance evidence. The preserved HTML is an unmodified capture whose navigation and download actions may lead to the current archive. Use the preserved PDF linked here to inspect the original note.

R2. ONET Resource Center. The ONET Content Model. Official framework documentation, accessed 12 September 2026 UTC. Content model.

R3. Goldberg, L. R. (1990). An alternative “description of personality”: The Big-Five factor structure. Journal of Personality and Social Psychology, 59(6), 1216–1229. Primary indexed abstract. DOI: 10.1037/0022-3514.59.6.1216.

R4. Gurven, M., von Rueden, C., Massenkoff, M., Kaplan, H., & Lero Vie, M. (2013). How universal is the Big Five? Testing the five-factor model of personality variation among forager–farmers in the Bolivian Amazon. Journal of Personality and Social Psychology, 104(2), 354–370. Primary article. DOI: 10.1037/a0030841.

R5. Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitive–developmental inquiry. American Psychologist, 34(10), 906–911. Original article copy. DOI: 10.1037/0003-066X.34.10.906.

R6. MacCann, C., Joseph, D. L., Newman, D. A., & Roberts, R. D. (2014). Emotional intelligence is a second-stratum factor of intelligence: Evidence from hierarchical and bifactor models. Emotion, 14(2), 358–374. Primary indexed abstract. DOI: 10.1037/a0034755.

R7. Petrides, K. V., & Furnham, A. (2001). Trait emotional intelligence: Psychometric investigation with reference to established trait taxonomies. European Journal of Personality, 15(6), 425–448. Publisher abstract. DOI: 10.1002/per.416.

R8. Mathieu, J. E., Heffner, T. S., Goodwin, G. F., Salas, E., & Cannon-Bowers, J. A. (2000). The influence of shared mental models on team process and performance. Journal of Applied Psychology, 85(2), 273–283. Primary indexed abstract. DOI: 10.1037/0021-9010.85.2.273.

R9. Lewis, K. (2003). Measuring transactive memory systems in the field: Scale development and validation. Journal of Applied Psychology, 88(4), 587–604. Primary indexed abstract. DOI: 10.1037/0021-9010.88.4.587.

R10. Woolley, A. W., Chabris, C. F., Pentland, A., Hashmi, N., & Malone, T. W. (2010). Evidence for a collective intelligence factor in the performance of human groups. Science, 330(6004), 686–688. Article copy. DOI: 10.1126/science.1193147.

R11. Bates, T. C., & Gupta, S. (2017). Smart groups of smart people: Evidence for IQ as the origin of collective intelligence in the performance of human groups. Intelligence, 60, 46–56. Author-accepted manuscript. DOI: 10.1016/j.intell.2016.11.004.

R12. Edmondson, A. (1999). Psychological safety and learning behavior in work teams. Administrative Science Quarterly, 44(2), 350–383. Article copy. DOI: 10.2307/2666999.

R13. American Educational Research Association, American Psychological Association, & National Council on Measurement in Education. (2014). Standards for Educational and Psychological Testing. AERA. Official open-access edition. Especially the validity chapter and Standards 1.0–1.6. This edition is cited as an identified methodological authority; no claim of a complete survey of current standards is made.

R14. Putnick, D. L., & Bornstein, M. H. (2016). Measurement invariance conventions and reporting [title abbreviated]. Developmental Review, 41, 71–90. Primary indexed abstract. DOI: 10.1016/j.dr.2016.06.004.

R15. Kozlowski, S. W. J., & Klein, K. J. (2000). A multilevel approach to theory and research in organizations: Contextual, temporal, and emergent processes. In K. J. Klein & S. W. J. Kozlowski (Eds.), Multilevel Theory, Research, and Methods in Organizations: Foundations, Extensions, and New Directions (pp. 3–90). Jossey-Bass. Original chapter copy.

R16. Evans, T. R., Hughes, D. J., & Steptoe-Warren, G. (2020). A conceptual replication of emotional intelligence as a second-stratum factor of intelligence. Emotion, 20(3), 507–512. Published online 7 February 2019. Published article copy and author’s institutional record. DOI: 10.1037/emo0000569.

Appendix A. Assumption register

Appendix B. Theorem register

“Theorem register” is retained for traceability. Its entries are elementary audit propositions; no novel theorem about human capability is claimed.

Appendix C. Evidence ledger

Reading depth reports the material inspected during this research pass, including parallel source review. “Abstract” means the study’s own indexed or publisher abstract; it does not mean that the full study was audited. No study’s raw data were reanalyzed. E0 and E12 are the only entries directly examining this proposal; external entries establish background or component evidence.

Source preservation: The retrieved HTML SHA-256 is 64e50891c62868b9469ce7f8e99bb59d11be32ac4167e05af0425b8a56f02a5f; the retrieved PDF SHA-256 is 2cb4da797242b19800f52b5a970bc0678d3fe870cb5b1a717b069523fabbd8fc. These identify the locally inspected copies; they do not authenticate the archive’s historical dating or the truth of its claims.

Appendix D. Glossary

Appendix E. Coverage gaps and continuation boundary

All requested main sections and appendices are present. Coverage of the underlying field is not exhaustive.

Continuation point: The next evidence-expansion pass begins with full-text retrieval for the abstract-only entries in Appendix C and a primary-literature review of values, attitudes, and expertise. The next empirical step is a frozen X1 codebook and pilot design. No unfinished narrative section has been hidden behind these research gaps.

Appendix F. End-to-end traceability

Edition and availability note

Expanded critical edition prepared 11 September 2026, America/New_York; released 12 September 2026 UTC. Not peer reviewed.

This expansion contains newly selected references, original explanatory examples, elementary mathematical proofs, and unperformed study proposals. It uses no private participant records. No new empirical dataset or research code accompanies it. The archive retains the original funding and competing-interest declarations. ChatGPT-assisted drafting and checks do not constitute peer review. The original source copies are linked in R1; the expanded manuscript and its generated PDF form this public edition.