Calab Skills Registry

This registry defines the skill vocabulary referenced by role criteria across the handbook. Every criterion row carries a Skill value in the format FRAMEWORK-SKILL-LEVEL, mapping the criterion to either a SFIA 9 skill, a SFIA responsibility attribute or behavioural factor, or a Calab custom skill. Criteria files are the artefact of record — a skill is in Calab’s assessment scope if and only if at least one criterion references it, and the coverage map below is derived from those references, never maintained independently.

SFIA 9 validation record

All SFIA codes, names, and level ranges in this registry and the Engineer L1/L2 criteria files were validated against the live SFIA 9 skill pages at sfia-online.org on 2026-08-13 (KNOW, LADV, and the CALAB-ADM coverage check on 2026-08-14). Each professional skill was checked for existence, current SFIA 9 name, and defined level range. SFIA 9 (published October 2024) is the pinned version; SFIA 10 is in consultation only.

Skill reference format

FRAMEWORK-SKILL-LEVEL, e.g. SFIA-PROG-3, SFIA-AUT-2, CALAB-AGE-1.

Two rules govern what the trailing level number means:

1. SFIA-* values carry SFIA’s own level (SFIA-PROG-3 means SFIA level 3 of Programming/software development). This preserves the external benchmark value — the Calab role level is already in the criterion ID (L2-SEM01), and the anchor map below makes translation mechanical.
2. CALAB-* values carry Calab-calibrated levels 1–5, deliberately aligned to role levels: CALAB-AGE-2 reads naturally as “agentic engineering as expected around L2”. The level descriptors in this registry define each rung.

The anchor is the default pitch, not a straitjacket — an individual criterion may reference a skill one level above or below its role-level anchor where the expectation genuinely sits there.

Calab to SFIA anchor map

SFIA responsibility attributes and behavioural factors

Where a criterion describes generic professional responsibility rather than a specific professional skill, it maps to SFIA 9’s responsibility structure using SFIA’s official codes. SFIA 9 defines four core generic attributes:

The former single “Business skills” attribute is expressed in SFIA 9 as 12 behavioural factors, each with its own code: Adaptability (ADAP), Collaboration (COLL), Communication (COMM), Creativity (CRTY), Decision-making (DECM), Digital mindset (DIGI), Improvement mindset (IMPM), Leadership (LEAD), Learning and development (LADV), Planning (PLAN), Problem-solving (PROB), and Security, privacy and ethics (SCPE).

Calab usage rules:

• SFIA-BUS is retained as a Calab umbrella shorthand for the Business skills / behavioural factors grouping, used where a criterion spans several factors (e.g. commercial awareness plus communication plus planning). SFIA itself still describes this five-way grouping, so the umbrella remains faithful.
• A specific factor code is preferred where one factor clearly fits — e.g. SFIA-IMPM-2 (Improvement mindset) for improvement-seeking behaviour.
• SCPE (Security, privacy and ethics) is new in SFIA 9 and the most load-bearing behavioural factor for an AI consultancy — criterion authors should prefer it over SFIA-BUS for ethics- and privacy-flavoured expectations.
• Attribute and behavioural-factor levels follow SFIA’s own 1–7 responsibility scale, per the anchor map.

Calab custom skills (CALAB-*)

Custom skills are defined only where SFIA 9 has no adequate coverage. Each has a named steward accountable for its level descriptors and for retiring or remapping the skill if SFIA later covers the ground. Stewards: guild-technology for the three AI skills (CALAB-AGE, CALAB-CTX, CALAB-EVL); guild-administration for CALAB-ADM.

The three AI skills were re-tested against SFIA 9 (2026-08-13): SFIA 9’s AI coverage comprises Machine learning (MLNG), AI and data ethics (AIDE), Data science (DATS), Data analytics (DAAN), Data engineering (DENG), and Analytical classification and coding (ANCC) — none covers agent workflow design, context/prompt engineering, or model-behaviour evaluation, and SFIA’s own AI skills guidance maps AI activity onto existing skills without addressing these three disciplines. All three are therefore retained. CALAB-ADM was validated the same way (2026-08-14): the SFIA 9 A–Z defines no professional skill for personal time management, timesheeting, leave planning, or administrative self-discipline — the closest, ADMN (Business administration), is a skill for administrative-services roles, not engineer self-discipline.

CALAB-AGE — Agentic engineering

Designing and operating agentic solutions: task decomposition for agents, guardrails, critical review of agent output, and agent tooling.

CALAB-CTX — Context engineering

Designing the context AI systems consume: prompt and instruction assets, retrieval design, and knowledge-as-code for AI consumption. The term and scope align with current industry usage (e.g. Anthropic’s context engineering guidance, which frames context engineering as curating system instructions, tools, MCP, retrieval, and message history).

CALAB-EVL — AI evaluation

Judging and assuring the quality of AI output: evaluations, regression harnesses for model behaviour, and human-in-the-loop review design.

Boundary notes to avoid double-counting: SFIA TEST (Functional testing) and NFTS (Non-functional testing) own deterministic software test mechanics; SFIA ANCC (Analytical classification and coding, levels 2–6) covers the dataset-labelling and ground-truth slice of evals work and may be cited as a complement; SFIA AIDE (AI and data ethics) owns ethics practices, not quality harnesses. CALAB-EVL covers what none of these do — evaluating non-deterministic model behaviour.

CALAB-ADM — Operational discipline

Managing one’s own administrative obligations — time recording, leave planning, expenses, policy acknowledgements — and, at higher levels, improving the administrative practices that keep the organisation running. Validated 2026-08-14: SFIA 9 defines no professional skill for personal administrative self-discipline (see custom-skill validation note above). Personal work organisation in the general sense maps to the PLAN behavioural factor; CALAB-ADM covers the Calab-specific administrative execution that PLAN does not.

Advisory SFIA category routing

Routing guidance for authors and reviewers — which guilds typically reference skills from each SFIA 9 category. This is not ownership: any guild may reference any skill through its own criteria, and sign-off authority always follows the criterion owner.

Engineer skill coverage map (derived)

Derived from the Skill columns of the Engineer L1 and L2 criteria files across all contributing guilds. Regenerate whenever criteria change. L3–L5 will be added when those levels are revised.

SFIA professional skills

Remap history (2026-08-13 SFIA 9 validation): HCDE retired in SFIA 9 → HCEV (which now means UX design; evaluation moved to USEV); RLMT (levels 4–7) → CEXP at L1/L2; CNSL (levels 4–7) → BUSA-3; EMRG (levels 4–6) → METL-3; INOV (now Innovation management, levels 5–7) → IMPM behavioural factor; RELM (now Release management) → DEPL for hands-on CI/CD and deployment work. RLMT-4, CNSL-4, and EMRG-4 remain natural references for the L3+ criteria pass.

SFIA responsibility attributes and behavioural factors

Calab custom skills

Maintenance

• Criterion-to-skill mappings live in the criteria files and are owned by the criterion’s owning guild — sign-off authority at evaluation follows the criterion owner, and this registry’s coverage map is only a derived summary.
• CALAB skill stewards maintain descriptor text, watch SFIA releases, and retire or remap a custom skill when SFIA adequately covers it.
• SFIA version pin: SFIA 9 (validated against live sfia-online.org pages, 2026-08-13). The version pin and anchor map are org-wide policy owned through governance. Re-validate on SFIA 10 release.

Known coverage gaps (candidates for future criteria)

SFIA 9 skills defined at levels 2–3 that a modern AI-era engineering ladder would normally assess but no Calab criterion references yet. Adding any of these requires authoring a criterion first — the registry never lists a skill without a referencing criterion.