Versioned axes and composition

Status: Proposed / Not authorized
Composition: Multidimensional profile vector
Weights and overall numeric score: Prohibited in this proposal

1. Purpose

Phase 2 needs a stable display and research structure before it needs weights. The proposal defines 12 distinct axes in three groups. It does not define a single overall value, threshold, pass state, percentile, rank, or role-readiness decision.

2. Proposed axis registry

Capability

1. Engineering fundamentals — implementation quality, testing, debugging, data handling, interfaces, and maintainability within the applicable benchmark scope.
2. Systems and architecture — boundaries, tradeoffs, integration, resilience, migration, and technical judgment.
3. Product judgment — problem framing, outcome definition, user and customer needs, prioritization, and iteration judgment.
4. Domain depth — separately versioned evidence within an approved domain pack; never an implicit core-score boost.
5. Operational and production ownership — governed deployment, observability, incident response, remediation, continuity, and handoff.

Operating

6. Time to production — a research candidate using predeclared start and governed-production events, with organizational constraints and censoring made visible. It is not typing speed, work hours, labor cost, or individual velocity.
7. Collaboration and communication — decision clarity, workshops, negotiation, cross-functional work, mentoring, and capability transfer.
8. Judgment under ambiguity — uncertainty reduction, reversible choices, escalation, and response to changing constraints.
9. AI-native delivery — effective and governed use of models, agents, evaluation, grounding, tools, guardrails, and human oversight. This remains separate so AI fluency cannot silently inflate or suppress engineering capability.

Meta

10. Evidence strength — independence, attribution, corroboration, lineage, and support quality. It is evidence about an estimate, not capability.
11. Coverage — how much of the declared scope has adequate eligible evidence. Missing scope yields missingness, not zero.
12. Recency — the freshness state under a capability- and evidence-class- specific policy. It cannot directly change a capability value.

3. Published composition

The complete proposed composition is the ordered vector:

Capability[
  engineering_fundamentals,
  systems_and_architecture,
  product_judgment,
  domain_depth,
  operational_and_production_ownership
]

Operating[
  time_to_production,
  collaboration_and_communication,
  judgment_under_ambiguity,
  ai_native_delivery
]

Meta[
  evidence_strength,
  coverage,
  recency
]

Every axis remains separately visible with its version, evidence scope, confidence or missingness, limitations, and applicable review state. Meta axes may qualify interpretation but cannot be blended into capability. Domain depth must name its domain pack and version.

The machine composition contains only the fixed axis ID, canonical label, group, ordinal, research status, numeric_value_authorized: false, and a controlled definition ID. It contains no free-form construct text, axis value, weight, contribution, threshold, or overall projection. Changing a definition therefore requires a schema and composition-version change rather than an unreviewed copy edit.

4. Versioning and comparability

An axis-composition version binds the exact ordered registry, definitions, construct roles, group membership, and display rules. Adding, removing, renaming, reordering, or changing the meaning of an axis requires a new composition version.

Profiles from different versions remain visually separated unless a published linking study and governance decision authorize comparison. A composition version alone creates no measurement authority; each axis still requires a validated rubric, applicable population, uncertainty policy, and approved release.

5. Prohibited composition behavior

The design contract structurally disallows:

• a global human score or weighted average;
• hiding a critical weakness through cross-axis averaging;
• using Evidence Strength, Coverage, or Recency as capability multipliers;
• collapsing AI-native delivery into general engineering;
• converting evidence absence into a low value;
• ranking or sorting named people;
• national or metro aptitude comparison;
• cost, compensation, prestige, or social-reach inputs; and
• activating an employment threshold.

A later proposal for any numeric axis must arrive through an independently approved scoring release. This Phase 2 composition does not contain weights.