# Competency mapping in academies: 2024–26 scales, 3-level split or 5-level verify

Sofia Almeida · November 18, 2026

> Default to a 3-level competency scale for 2024–26 academies; verify on 5 levels with two raters when promotion, pay, or certification decisions depend on a single pass/fail rating.

| Takeaway | Detail |
| --- | --- |
| Default to a 3-level competency scale for 2024–26 employer academies. | Use foundation/practitioner/expert as the default scale. |
| Add 5-level verification only when a level decision has consequences. | Trigger verification for promotion, pay band, or certification outcomes. |
| Apply the two-rater consequence test to every competency. | If a single rater's pass/fail decision triggers promotion, pay, or certification, verify it on a 5-level scale with two independent raters. |
| Operationalized prompts can sharply raise specification adherence. | Across 12 runs per overall prompt condition, mean FCS rose from 9.26% in specification and 1.85% in code for minimal prompts to 97.22% and 82.41% for operationalized prompts. |

For employer academies building 2024–26 competency maps, this guide delivers a default 3-level scale: foundation, practitioner, and expert.

The answer changes when a level decision carries promotion, pay band, or certification consequences, which requires 5-level verification with two independent raters.

![Sunlit academy corridor with pale stone arches polished](https://static.mm-ais.com/article-images-ai/competency-mapping-in-academies-2024-26-ai-180d0965.jpg)
Sunlit academy corridor with pale stone arches polished

## How the 3-level split actually works

The 3-level split works as a program-stage scale, not as a job-architecture scale. In an employer academy, foundation, practitioner, and expert describe where a learner sits in the learning pathway: intake, cohort completion, and certification panel. That distinction matters because the same competency can be mapped to academy progress without implying that “expert” equals a particular title, band, or reporting line. The scale answers, “What can this person demonstrate at this point in the academy?” It does not answer, “What level of job have they reached?”

To build the scale, assign each competency three observable anchors. Foundation means the learner performs with coaching. Practitioner means the learner performs independently. Expert means the learner teaches or reviews others. Then tie each anchor to a named program stage: foundation to intake assessment, practitioner to the cohort capstone, and expert to the certification panel. This gives each level a stable place in the academy workflow and prevents raters from inventing their own stage logic during review.

Write each anchor as an observable work product, not as an adjective. “Strong communicator” is not assessable. A completed audit memo, a coaching session plan, a stakeholder briefing, or a review checklist with annotations is assessable because a rater can hold the artifact and compare it against the anchor. The practical check is simple: if a level cannot be evidenced by an artifact a rater can hold, the level is not assessable in its current form. Rewrite it as evidence before using it in a competency map.

This artifact rule also keeps the three levels from collapsing into personality judgments. At intake, the rater is not asking whether the learner seems promising; the rater is checking whether the submitted work shows performance with coaching. At cohort completion, the rater is checking whether the capstone shows independent performance. At the certification panel, the rater is checking whether the learner can teach, critique, or review the work of others. Each stage has a different proof burden, and the proof should live in the work sample.

The mapping is deliberately narrow. It gives academy teams a common development language across 2024–26 competency maps while keeping the scale close to program design. A learner can move from foundation to practitioner as the cohort progresses, then to expert when they demonstrate review behavior at the panel. That sequence describes academy progression. It should not be read as a substitute for HR leveling, compensation bands, or promotion criteria.

Operationalized prompts and specifications have been shown to improve consistency in AI-related evaluation work: alphaXiv reports mean FCS rising from 9.26% in specification and 1.85% in code for minimal prompts to 97.22% and 82.41% for operationalized prompts across 12 runs per overall prompt condition. The academy lesson is analogous: define the artifact, stage, and anchor precisely enough that different raters are looking at the same thing. A 3-level scale is useful when the decision is developmental because it keeps feedback tied to visible work and visible progress.

![Quiet hillside campus pathway dawn slate steps winding](https://static.mm-ais.com/article-images-ai/competency-mapping-in-academies-2024-26-ai-61cb0dd5.jpg)
Quiet hillside campus pathway dawn slate steps winding

## When 5-level verification earns its cost

5-level verification earns its cost only when a single rater's level decision can change a person's pay, promotion, or certification status. If the decision is developmental only, the default 3-level split remains sufficient; adding raters and levels to non-consequential competencies buys precision that no one acts on.

Run the audit as three consequence questions: Does a level decision change pay band? Does it change promotion eligibility? Does it change external certification status? Any yes moves that competency into 5-level verification; three no's keep it on the 3-level path. Apply the test competency by competency, not by job family, because one academy can have a certification-gated safety competency beside a development-only communication competency.

| Consequence question | Result if yes | Result if no |
| --- | --- | --- |
| Changes pay band? | Move to 5-level verification | Keep 3-level split |
| Changes promotion eligibility? | Move to 5-level verification | Keep 3-level split |
| Changes external certification status? | Move to 5-level verification | Keep 3-level split |

On the 5-level scale, make levels 2 and 4 verification gates rather than ordinary score points. Level 2 should not count as practitioner until a second independent rater confirms the claim; level 4 should not count as expert until a work-product review panel accepts evidence. This keeps the extra levels from becoming labels and forces the consequence-bearing decision through a second judgment.

Cap the verified set to the smallest subset that survives the audit. The cap is a governance control, not a scoring rule: if the audit expands beyond a narrow minority of competencies, revisit whether the consequence questions are being applied too loosely or whether the academy is treating development milestones as HR decisions.

Document the trigger for each verified competency in the map itself: the consequence question that produced the yes, the gate level, and the rater or panel required. That record lets a future reviewer distinguish a deliberate gate from scope creep and prevents the 5-level layer from becoming the default by habit.

## Options compared: split vs verify

Employer academies comparing scale designs in 2024–26 should evaluate each option by the decision it supports, not by the apparent precision of extra levels. A split-versus-verify model treats the competency map as an operating choice: one scale serves routine development, while a stricter scale is reserved for decisions that require defensible ratings. The comparison below names where each option wins and where it creates avoidable cost.

| Option | Rater load | Calibration cost | Best for | Loses when |
| --- | --- | --- | --- | --- |
| 3-level split (winner for development-only competencies) | 1 rater per competency | Quarterly only | Development feedback and cohort placement | The score feeds pay or promotion |
| 5-level verify (winner only for gated decisions) | 2 raters per gated competency plus a panel at level 4 | Per cycle | Certification and promotion gates | Used for routine development feedback (rater cost with no decision value) |

Read the table as a default-versus-exception rule. The 3-level split wins as the default map for development-only competencies because it matches a single-rater review and quarterly calibration to feedback and placement decisions. When a competency is used only to guide learning progress or group learners, the lighter operating load preserves capacity for coaching instead of scoring.

The 5-level verify option wins only for gated decisions, where the added rater load and per-cycle calibration buy decision defensibility. Its value disappears when applied to routine development feedback: assigning two raters per gated competency, plus a panel at level 4, creates rater cost with no decision value. A practical check is to limit verification to the gated subset named in the map and keep every development-only competency on the default split.

To keep the comparison actionable, review the competency map as a portfolio rather than a uniform scale. Most entries should remain development-only and stay on the 3-level split; only the subset tied to gates should move to 5-level verification. This split-by-decision design makes the map easier to administer, easier to explain to raters, and aligned with the actual consequences attached to each rating.

## Costs and the numbers to check

Before you commit any competency to a five-level verification path, run the rater-hours budget. The hybrid model has a fixed arithmetic shape: annual rater hours = (competencies × 3 levels × 0.5h anchor-writing) + (gated competencies × 2 raters × 1h per assessment cycle) + (4h per quarterly calibration session × raters). The first term is a one-time design load spread across the year; the second is the recurring load that only appears where a level decision carries consequence; the third is the maintenance load that keeps both raters reading the same anchors the same way. Do this before you add a single extra level, because the second and third terms are where a competency map quietly becomes expensive.

The calibration-divergence threshold is the early-warning signal for a broken scale. If more than 15% of two-rater pairs diverge by more than one level in a calibration cycle, the anchors — not the raters — are the defect. That is a design finding, not a performance finding. Budget a rewrite cycle for the level descriptors before you add more gated competencies to the map; adding raters or adding levels on top of ambiguous anchors multiplies the disagreement rather than resolving it. The check belongs in the same quarterly session already counted in the formula, so it costs no extra meeting time to run.

Track three numbers per competency across the year: anchor-writing hours actually logged against the 0.5h-per-level estimate, assessment hours against the two-rater × one-hour-per-cycle line, and the share of pairs landing more than one level apart. The first two tell you whether the budget held; the third tells you whether the scale is still fit to gate anything. When the divergence share crosses the threshold, freeze new gated additions for that cycle and rewrite the descriptors. When it stays under, the existing anchors are doing their job and the map can absorb the next consequence-bearing competency without a redesign.

The reason to keep the formula explicit is that specification quality drives consistency. As the alphaXiv specification-adherence figures show above, operationalized specifications sharply improved measured outcomes, and the same direction holds here: a level descriptor that states observable evidence produces tighter two-rater agreement than one that relies on shared intuition. Treat anchor-writing as the highest-leverage half-hour in the model, not as administrative overhead.

Architecture choices follow the same logic. In application design, the useful answer is usually "it depends" on the forces at work, and the recommendation is to let those forces drive the structure rather than reach for a default pattern (microservices.io, "Designing an architecture using dark matter and dark energy"). For a competency map, the force is consequence: where no pay, promotion, or certification outcome rides on the decision, the extra rater and extra levels buy little; where one does, the second rater is the control that makes the decision defensible. The formula tells you what that control costs before you promise it.

## What the evidence does not establish

It is worth being precise about what the evidence base does not yet settle, because that is where most academy design arguments quietly overreach. The table below is the missing-evidence inventory: it lists the claims you should not attribute to published research when you defend a 3-level default or a 5-level verification layer in a 2024–26 competency map. Treat each row as a prompt for local validation, not as a reason to avoid making the design decision.

| Question a team may ask | What no published source in this review establishes | What to do instead |
| --- | --- | --- |
| Do 5-level scales produce better inter-rater reliability than 3-level scales inside employer academies? | No published study establishes this for internal academy settings. | Run your own calibration exercise and compare divergence between raters on the same learner evidence. |
| Is there a universal optimal share of competencies that should be gated? | No evidence establishes one. | Use a method-derived guardrail against consequence sprawl, then test whether your gate count stays defensible. |
| Does adding levels by itself improve fairness? | No published source in this review isolates level count from rater training, rubric wording, and evidence quality. | Operationalize the rubric before changing the scale; specification-driven work reported by alphaXiv showed mean FCS rising from 9.26% to 97.22% in prompts and from 1.85% to 82.41% in code when prompts were operationalized. |
| Is there a research-backed “best” architecture for assessment governance? | No single architecture is established as best. The architecture discussion from microservices.io frames the answer as context-dependent, using dark matter and dark energy as metaphors for competing forces. | Name the forces in your own setting: developmental feedback, consequence exposure, rater capacity, and audit needs. |

The first missing-evidence point is the one most likely to be misstated in a design review. No published study establishes that 5-level scales produce better inter-rater reliability than 3-level scales in internal academy settings. That means level count is a design choice to be tested through your own calibration divergence data. A practical check is to have two raters score the same sample of learner evidence, record where they land on the scale, and examine whether disagreement clusters around particular competencies or particular level boundaries.

The second missing-evidence point concerns gating. No evidence establishes a universal optimal percentage of competencies that should carry promotion, pay band, or certification consequences. The 20% cap used in this approach is a method-derived guardrail against consequence sprawl, not an empirical finding. Use it as a review trigger: if your gated share rises above it, require a written consequence rationale for each added competency rather than assuming the map has become more rigorous.

This is also why “more levels” should not be treated as a fairness upgrade by default. The alphaXiv specification-adherence figures cited above show how operationalized specifications changed measured outcomes; the transferable check is to specify what each level means, what evidence qualifies, and how raters handle ambiguity before you add granularity.

Finally, keep the architecture question separate from the measurement question. The microservices.io architecture discussion makes the general point that design answers depend on context and that there is no silver bullet. For an academy, the context is the consequence attached to the decision. Where no consequence is attached, a missing-evidence claim cannot justify extra levels; where a consequence is attached, the missing evidence still cannot replace local calibration. The defensible position is narrow: choose the simpler scale by default, verify the consequential cases, and document the checks that support both choices.

## And copy-usable worksheet

Here is a copy-usable worked example: a 12-competency manager map run through the consequence test. Start with a full 12-item list and run every item through the two-rater consequence test. In this example, three items come back gated: coaching others, performance decisions, and compliance sign-off. That is 3 of 12, or 25%, which sits above the 20% cap for gated items. On review, compliance sign-off is demoted to checklist-verified rather than competency-verified, because the sign-off itself is a document control step, not a judgment about a person's capability. That brings the gated count to 2 of 12, or 17%, inside the cap.

The worked map is reproduced below so you can copy the structure. The two gated competencies are the only ones that receive 5-level verification with two independent raters. The other ten stay on the default 3-level split with one rater. Note that the demoted item did not disappear from the map; it moved to a checklist control, which is a different verification mechanism and does not consume a rater slot.

| Competency | Consequence test result | Scale | Raters | Verification route |
| --- | --- | --- | --- | --- |
| Coaching others | Gated: promotion and pay band | 5-level | Two independent | Competency-verified |
| Performance decisions | Gated: promotion and pay band | 5-level | Two independent | Competency-verified |
| Compliance sign-off | Initially gated; demoted on review | Not applicable | None for competency | Checklist-verified |
| Other nine competencies | Developmental only | 3-level | One | Competency-verified |

Checkpoint 1 is intake. All 12 competencies are scored at 3 levels by one rater, including the two that will later be gated. This gives the academy a baseline for every learner without spending verification capacity up front. The single rater at intake is enough because no level decision at this stage changes pay, promotion, or certification status.

Checkpoint 2 is the capstone. Only the 2 gated competencies are scored by two independent raters on the 5-level scale. The other 10 remain at 3 levels with one rater. The academy's rule is simple: if a level decision at the capstone triggers a promotion, pay band, or certification outcome, verify it on a 5-level scale with two independent raters; otherwise keep the default.

The arithmetic is worth checking before you adopt the worksheet. Three gated items out of 12 is 25%, above the 20% cap. After the demotion, 2 out of 12 is 16.67%, which rounds to 17%, below the cap. If your own map lands above 20%, look for items like compliance sign-off that are really document controls in disguise and move them to checklist verification. That keeps the gated set small and the verification effort focused where a single rater's decision actually carries weight.

## Decision rules for your 2024–26 map

Apply these rules at map freeze and again whenever a rater panel, assessment task, or AI tool changes. Each rule governs the map itself, not the learner.

If a competency's level decision changes pay, promotion, or certification, verify it on a 5-level scale with two independent raters and an observable work product. If it only guides development or cohort placement, keep the 3-level scale with one rater.

If two-rater calibration diverges by more than one level on more than 15% of pairs in a cycle, freeze new gating and rewrite the anchors before the next assessment cycle. Do not add gates while the anchors are unstable.

If gated competencies exceed 20% of the map, convert the lowest-stakes gate to a checklist verification rule. The cap keeps verification load proportional to consequence-bearing decisions.

AI-exposure re-check rule: if an observable work product can be generated or materially altered by AI, re-check it with a second rater or require a live demonstration before the level decision gates pay, promotion, or certification. Treat AI exposure as a trigger for re-verification, not as a reason to remove the gate.

Record which rule fired for each competency so the next cycle can audit changes without re-deriving the rationale.

| If | Then |
| --- | --- |
| Level decision changes pay, promotion, or certification | Use 5 levels, 2 independent raters, and an observable work product |
| Level decision guides development or cohort placement only | Use 3 levels with 1 rater |
| Two-rater calibration diverges by more than one level on more than 15% of pairs in a cycle | Freeze new gating and rewrite anchors before the next assessment cycle |
| Gated competencies exceed 20% of the map | Convert the lowest-stakes gate to a checklist verification rule |
| An observable work product can be generated or materially altered by AI | Re-check with a second rater or require a live demonstration before the level decision gates an outcome |

## What to do next

| Step | Action | Why it matters |
| --- | --- | --- |
| 1 | Adopt foundation/practitioner/expert as the default 3-level scale for 2024–26 employer academies. | Keeps the map simple unless a level decision has consequences. |
| 2 | Run the two-rater consequence test on every competency: determine whether one rater's pass/fail level decision triggers promotion, pay band, or certification. | Separates competencies that need verification from those that can stay on the 3-level split. |
| 3 | For triggered competencies, verify on a 5-level scale with two independent raters scoring against an observable work product. | Prevents a single rater's consequence-bearing decision from standing unverified. |
| 4 | Keep non-triggered competencies on the 3-level split with one rater and a quarterly calibration review. | Preserves efficiency while catching rater drift. |
| 5 | Use operationalized prompts rather than minimal prompts when specifying competency criteria. | The article's minimal-to-operationalized comparison shows a sharp rise in mean FCS across 12 runs per condition. |
| 6 | Record which competencies triggered verification and the calibration cadence in the 2024–26 academy map. | Creates an audit trail for later promotion, pay, or certification reviews. |

## Frequently Asked Questions

**What scale should an employer academy use by default for its 2024–26 competency map?**

Default to a 3-level competency scale for 2024–26 employer academies, using foundation, practitioner, and expert as the default scale.

**When does a competency level decision require 5-level verification?**

Add 5-level verification only when a level decision has consequences, specifically triggering verification for promotion, pay band, or certification outcomes.

**What does the two-rater consequence test require for a competency?**

Apply the two-rater consequence test to every competency, and if a single rater's pass/fail decision triggers promotion, pay, or certification, verify it on a 5-level scale with two independent raters.

**How much can operationalized prompts change specification adherence compared with minimal prompts?**

Operationalized prompts can sharply raise specification adherence: across 12 runs per overall prompt condition, mean FCS rose from 9.26% in specification and 1.85% in code for minimal prompts to 97.22% and 82.41% for operationalized prompts.

**What does the 3-level split represent in an employer academy?**

The 3-level split works as a program-stage scale, not as a job-architecture scale.

**Where do foundation, practitioner, and expert sit in an employer academy learning pathway?**

In an employer academy, foundation, practitioner, and expert describe where a learner sits in the learning pathway: intake, cohort completion, and certification panel.

## Quick answers

| What should employer academies default to for 2024–26 competency mapping? | Employer academies should default to a 3-level competency scale for 2024–26. |
| --- | --- |
| What are the levels in the default scale? | The default scale is foundation, practitioner, and expert. |
| When should 5-level verification be added? | Add 5-level verification only when a level decision has consequences. |
| Which outcomes trigger verification? | Verification is triggered for promotion, pay band, or certification outcomes. |
| What did operationalized prompts do to FCS across the reported runs? | Across the reported runs per overall prompt condition, mean FCS rose from the figures named 9.26% and 1.85% for minimal prompts to the figures named 97.22% and 82.41% for operationalized prompts. |

Also worth reading: **Skills data portability 2026: xAPI adoption checklist for employer academies**: [Skills data portability 2026: xAPI](https://lpi.academy/blog/skills-data-portability-2026-xapi-adoption-checklist-for-employer-academies.php) · **Competency-based academies: verify time-to-proficiency claims before committing**: [Competency-based academies: verify time-to-proficiency claims](https://lpi.academy/blog/competency-based-academies-verify-time-to-proficiency-claims-before-committing.php) · **How competency-mapped manager academies reduce promotion failure rates by 30%: A 2024 LPI benchmark analysis**: [How competency-mapped manager academies reduce](https://lpi.academy/blog/how-competency-mapped-manager-academies-reduce-promotion-failure-rates-by-30-a-2024-lpi-benchmark-analysis.php)

### Related reading

- [How competency-mapped manager academies reduce promotion failure rates by 30%: A 2024 LPI benchmark analysis](https://lpi.academy/blog/how-competency-mapped-manager-academies-reduce-promotion-failure-rates-by-30-a-2024-lpi-benchmark-analysis.php)
- [LMS Competency Mapping: 3 Platforms, xAPI Depth vs SCORM Fallback 2026](https://lpi.academy/blog/lms-competency-mapping-3-platforms-xapi-depth-vs-scorm-fallback-2026.php)
- [Audit manager competency models for AI exposure: task scores vs 20% cutoff](https://lpi.academy/blog/audit-manager-competency-models-for-ai-exposure-task-scores-vs-20-cutoff.php)
- [How to fix manager mistakes: 90-Day Competency Rule vs. Delayed Review](https://lpi.academy/blog/how-to-fix-manager-mistakes-90-day-competency-rule-vs-delayed-review.php)
- [Skills data portability 2026: xAPI adoption checklist for employer academies](https://lpi.academy/blog/skills-data-portability-2026-xapi-adoption-checklist-for-employer-academies.php)
- [Corporate training deadlines 2026: 412 academies cut 90 to 30 days, no gain](https://lpi.academy/blog/corporate-training-deadlines-2026-412-academies-cut-90-to-30-days-no-gain.php)

### Latest

- [Workers' comp safety programs in 2026: is the 30% fewer-claims claim worth...](https://lpi.academy/blog/workers-comp-safety-programs-in-2026-is-the-30-fewer-claims-claim-worth-funding.php)
- [Skills Taxonomy Refresh Timing: Annual vs Biennial at 2.5-Year Half-Life](https://lpi.academy/blog/skills-taxonomy-refresh-timing-annual-vs-biennial-at-25-year-half-life.php)
- [Competency-based academies: verify time-to-proficiency claims before committing](https://lpi.academy/blog/competency-based-academies-verify-time-to-proficiency-claims-before-committing.php)

Canonical: https://lpi.academy/blog/competency-mapping-in-academies-202426-scales-3-level-split-or-5-level-verify.php
Markdown: https://lpi.academy/blog/competency-mapping-in-academies-202426-scales-3-level-split-or-5-level-verify.php/index.md
