# Skills Taxonomy Refresh Timing: Annual vs Biennial at 2.5-Year Half-Life

Sofia Almeida · November 12, 2026

> At a 2.5-year half-life, annual refresh is required; classify every domain by estimated half-life and refresh biennially only when half-life exceeds 3 years.

| Takeaway | Detail |
| --- | --- |
| Annual refresh is required for domains at a 2.5-year half-life. | At 2.5 years, annual cadence is the only one that keeps competency data under 50% decayed. |
| Biennial refresh is defensible only when half-life exceeds 5 years. | Domains with half-life >5 years may use biennial refresh; otherwise use annual. |
| Classify every domain by estimated half-life using a 3-year cutoff. | Refresh annually if half-life ≤3 years; refresh biennially if half-life >3 years. |
| High-decay domains include AI, cybersecurity, cloud, and software engineering. | These fields have technical skill half-lives compressed to as little as 2.5-5 years. |

At a 2.5-year technical skill half-life, an annual review leaves about 24% of skill definitions stale at 12 months, while a biennial review leaves about 43% stale at 24 months; both are under 50% at those review points, and 50% is reached at 30 months.

Biennial refresh is defensible for domains whose estimated half-life exceeds 3 years; domains at or below 3 years should use annual refresh.

![Skills Taxonomy Refresh Timing](https://static.mm-ais.com/article-images-ai/skills-taxonomy-refresh-timing-annual-vs-ai-0e8d085d.jpg)

## How Half-Life Drives Taxonomy Decay

Taxonomy decay follows the same logic as skill obsolescence: the longer a competency framework sits without review, the larger the share of its nodes that no longer reflect current practice. In high-decay domains such as artificial intelligence, cybersecurity, software engineering, and cloud computing—fields where Morson Edge reports the pace of obsolescence is fastest—a short skill half-life compresses the window in which definitions remain reliable. The mechanism is straightforward: each node’s currency erodes continuously between refreshes, so the staleness you face on review day is a function of elapsed time, not just the act of reviewing.

You can estimate that staleness with a simple decay fraction. The share of nodes still current at a review date is 0.5^(months since last refresh / half-life in months), and the share that is stale is 1 - 0.5^(months since last refresh / half-life in months). At a 2.5-year half-life (30 months), a domain reaches the midpoint exactly on schedule: after 30 months, 0.5^(30/30) = 0.5, meaning 50% of its skill definitions have lost currency.

Applying the same formula to common review intervals shows how cadence determines the condition of the taxonomy at the moment of refresh. For a domain with a 30-month half-life, an annual review occurs after 12 months, while a biennial review occurs after 24 months:

| Months since last refresh | Remaining current (0.5^(months/30)) | Stale at review |
| --- | --- | --- |
| 12 (annual) | ~75.8% | ~24.2% |
| 24 (biennial) | ~57.4% | ~42.6% |
| 30 (half-life) | 50.0% | 50.0% |

Under a biennial cycle for a domain with a 2.5-year half-life, a review at 24 months arrives with about 43% of its nodes stale, forcing the refresh to function more like a reconstruction than a routine update. An annual cycle interrupts that curve earlier, leaving about 24% stale at 12 months, so the taxonomy carries a smaller burden of outdated definitions into each review.

To apply this mechanism, classify each domain by its estimated half-life, convert that half-life and your planned cadence into months, and compute the decay fraction before committing to a schedule. The result is a direct, comparable measure of how much of your competency data will be stale on review day, giving you a concrete basis for judging whether a given cadence keeps the taxonomy within a tolerable decay threshold.

![How Half-Life Drives Taxonomy Decay — Skills Taxonomy Refresh Timing](https://static.mm-ais.com/article-images-ai/skills-taxonomy-refresh-timing-annual-vs-ai-5f770fd9.jpg)

## The Evidence on Skill Decay Rates

The evidence base for technical skill decay draws on three Morson publications. A LinkedIn post by Morson Edge states that "the half-life of many technical skills has compressed to as little as 2.5-5 years" in artificial intelligence, cybersecurity, software engineering, and cloud computing. Morson Praxis (January 27, 2026) frames "talent obsolescence" as a productivity cost driven by "rapid skill decay," signaling that employers treat decay as a measurable loss rather than a metaphor. Morson Group (December 11, 2025) adds that "technical half-life" and "skill decay" are "reshaping industries and disrupting productivity." Together, these publications describe the same compression from employer, productivity, and industry perspectives.

In this context, "half-life" is a planning interval, not a laboratory constant. It marks the period over which a substantial share of currently relevant technical competencies in a domain can lose currency as tools, platforms, threats, and practices change. The 2.5-5 year range is a band for estimation, not a single expiry date for every skill. Some competencies decay faster, especially those tied to specific products or frameworks, while foundational concepts may remain useful longer. The range's value is that it gives taxonomy owners a defensible outside boundary for high-decay domains.

| Source | Date | Contribution |
| --- | --- | --- |
| Morson Edge LinkedIn post | February 2026 | Names the 2.5-5 year half-life and four high-decay domains |
| Morson Praxis | January 27, 2026 | Frames talent obsolescence as a productivity cost of rapid skill decay |
| Morson Group | December 11, 2025 | Links technical half-life and skill decay to industry reshaping and productivity disruption |

The convergence matters because it is cross-source and close in time. A single post could be promotional; three related publications across roughly eight weeks, each emphasizing productivity consequences, are better treated as a directional employer signal. That signal is strongest for the four domains named in the LinkedIn post: artificial intelligence, cybersecurity, software engineering, and cloud computing. It should not be automatically extended to domains whose tooling, standards, or threat models change more slowly.

The evidence also has limits. It is employer-focused, it does not provide a peer-reviewed measurement method, and it does not state how much of a taxonomy becomes stale at any specific review date. Therefore, the 2.5-5 year range should be used to classify domains and set review cadence, not to claim an exact percentage of obsolete nodes. Where a domain's observed change rate is slower than the band, the longer estimate should govern; where it is faster, the shorter estimate should govern.

Practical check: estimate a domain's half-life by asking how long its current core tools, certifications, threat models, or platform conventions remain standard practice. If the honest answer falls inside the 2.5-5 year band, classify the domain as high-decay and refresh it on the shorter cycle. If the answer is clearly longer, classify it as lower-decay and use the longer cycle. This keeps the evidence's range tied to observable practice rather than to a calendar assumption.

![The Evidence on Skill Decay Rates — Skills Taxonomy Refresh Timing](https://static.mm-ais.com/article-images-pixabay/skills-taxonomy-refresh-timing-annual-vs-75a356e0.jpg)

## Annual vs Biennial: Cadence Compared

Choosing between an annual and a biennial taxonomy refresh is not a matter of preference but of decay class. As Morson Edge reports, technical skills in artificial intelligence, cybersecurity, software engineering, and cloud computing now turn over on a compressed cycle, which means the right cadence depends on how quickly a domain’s competencies lose relevance. For fast-decay domains, an annual refresh is the clear winner because a two-year gap allows more than a third of mapped competencies to decay before the next review. For slow-decay domains, a biennial refresh wins by preserving analyst capacity for the areas where change is actually occurring.

If a domain lacks two cycles of edit data, mark its half-life estimate as provisional and revisit it at the next review; use the best available estimate to set a provisional cadence rather than leaving the domain outside the classification.

![Annual vs Biennial: Cadence Compared — Skills Taxonomy Refresh Timing](https://static.mm-ais.com/article-images-pixabay/skills-taxonomy-refresh-timing-annual-vs-84350573.jpg)

## And Refresh Worksheet

For a copy-usable cadence assignment, start with a classification pass: estimate half-life per domain in your taxonomy, assign a decay class, and set review frequency against that class. As a worked example, an employer academy with 12 domains assigned decay classes as follows: 5 fast (AI, cloud, cybersecurity, data engineering, software engineering), 4 medium, and 3 slow. Applying an annual refresh to fast-decay domains and a biennial refresh to the remaining 7 yields 5 + 7/2 = 8.5 domain-reviews per year, cutting review load from 12 to 8.5 without applying a single cadence across the taxonomy.

Use three checkpoints to test that assignment against a 2.5-year half-life, a figure cited by Morson Edge for high-decay technical domains. **Checkpoint 1 (time-based currency):** at 12 months post-refresh, a fast-decay domain at a 2.5-year (30-month) half-life retains roughly 76% currency (0.5^(12/30)); at 24 months, roughly 57% (0.5^(24/30)); at 30 months, exactly 50% (0.5^1) — the point at which a biennial interval reaches the half-life threshold for a 2.5-year domain.

**Checkpoint 2 (edit pressure):** before committing to a slot in your calendar, count substantive edits to each domain at its due review point against a fixed sample of nodes (not total word count). For domains due annually, that sample should be reviewed at 12 months; for domains due biennially, review the same sample at 24 months and explicitly record the share changed. If that recorded share approaches the 50% mark at or before 24 months for a domain you placed in the biennial class, reclassify it to annual — no blanket percentage override is assumed here.

**Checkpoint 3 (capacity fit):** convert the class counts to a projected review load (domains due this year ÷ 1 + domains due next year ÷ 2) and compare it to verified reviewer capacity for the year, not an idealized average. Using the 12-domain example above, 8.5 domain-reviews per year is the figure to test against capacity; if capacity falls below that projected load after classification, the correct lever is reclassification by half-life, not stretching a biennial interval for a domain with a 2.5-year half-life.

## Decision Rules for Your Refresh Calendar

The refresh calendar should be a decision table rather than a policy statement: every domain carries an estimated half-life, a cadence, a last-review date, and the evidence behind the estimate. Do not apply one cadence to the whole taxonomy; classify every domain first, then apply the threshold rules below. No domain may keep an exception without a documented reclassification.

If a domain's estimated half-life is three years or less, schedule an annual refresh; if it exceeds three years, schedule a biennial refresh. The estimate and its basis belong in the calendar entry, and any change in cadence requires a documented reclassification rather than an informal override. No domain-level exceptions are allowed without that documentation.

If a major tool or regulation change lands mid-cycle, run an out-of-cycle patch review within 48 hours for the affected nodes instead of waiting for the scheduled refresh. The 48-hour clock starts when the change is confirmed, not when it is announced; if impact is unclear, patch the likely affected nodes first and record the assumption. Scope the patch to changed nodes and their dependencies, and treat it as a supplement to, not a replacement for, the full scheduled review.

If a domain classified as slow records more than 25% substantive node edits in its last review, reclassify it to annual at the next planning cycle. Count only substantive edits—nodes added, removed, or materially changed—not wording or formatting changes. Treat that edit rate as evidence that the half-life estimate was too long, and update the estimate before the next calendar is approved.

At each planning cycle, re-run the classification before carrying forward last year's cadence, because a domain's half-life estimate can move in either direction. Keep the three rules visible in the planning artifact so reviewers can test each domain against the same conditions, and resolve any failed rule before the calendar is finalized.

| Estimated half-life of 3 years or less | Annual refresh |
| --- | --- |
| Estimated half-life greater than 3 years | Biennial refresh |
| Major mid-cycle tool or regulation change | Patch review within 48 hours |
| More than 25% substantive edits in a slow domain | Reclassify to annual |

## Frequently Asked Questions

**For a domain with a 2.5-year half-life, which refresh cadence is required to keep competency data under 50% decayed?**

Annual refresh is required for domains at a 2.5-year half-life, and at 2.5 years annual cadence is the only one that keeps competency data under 50% decayed.

**When is biennial refresh defensible under the 5-year threshold?**

Biennial refresh is defensible only when half-life exceeds 5 years.

**How should domains be classified using the 3-year cutoff?**

Classify every domain by estimated half-life using a 3-year cutoff, refreshing annually if half-life is 3 years or less and biennially if half-life exceeds 3 years.

**At a 2.5-year technical skill half-life, what stale share does a biennial review leave at 24 months?**

At a 2.5-year technical skill half-life, a biennial review leaves about 43% of skill definitions stale at 24 months.

**Which fields are named as high-decay domains?**

High-decay domains include AI, cybersecurity, cloud, and software engineering.

**How does taxonomy decay relate to skill obsolescence?**

Taxonomy decay follows the same logic as skill obsolescence: the longer a competency framework sits without review, the larger the share of its nodes that no longer reflect current practice.

## Quick answers

| What refresh cadence is required for domains at a 2.5-year half-life? | Annual refresh is required for domains at a 2.5-year half-life. |
| --- | --- |
| When is biennial refresh defensible according to the article? | Biennial refresh is defensible only when half-life exceeds 5 years. |
| What cutoff is used to classify domains by estimated half-life? | Classify every domain by estimated half-life using a 3-year cutoff. |
| Which domains are identified as high-decay domains? | High-decay domains include AI, cybersecurity, cloud, and software engineering. |
| How does taxonomy decay relate to skill obsolescence? | Taxonomy decay follows the same logic as skill obsolescence: the longer a competency framework sits without review, the larger the share of its nodes that no longer reflect current practice. |

Also worth reading: **2026 Peer Coaching Academy: Architecture, Mistakes & Tactics**: [2026 Peer Coaching Academy: Architecture,](https://lpi.academy/blog/2026-peer-coaching-academy-architecture-mistakes-tactics.php) · **How to Use Assessment Data to Close Leadership Skill Gaps**: [How to Use Assessment Data](https://lpi.academy/blog/how_to_use_assessment_data_to_close_leadership_skill_gaps.php) · **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)

### Related reading

- [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)
- [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)
- [When to Replace One-Off Workshops: 90-Day Manager Cohort vs One-Off: Replace or Keep?](https://lpi.academy/blog/when-to-replace-one-off-workshops-90-day-manager-cohort-vs-one-off-replace-or-keep.php)
- [Employee Onboarding Platform 2026: 90-Day Role-Readiness Gate Before You Buy](https://lpi.academy/blog/employee-onboarding-platform-2026-90-day-role-readiness-gate-before-you-buy.php)

### Latest

- [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)
- [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)

Canonical: https://lpi.academy/blog/skills-taxonomy-refresh-timing-annual-vs-biennial-at-25-year-half-life.php
Markdown: https://lpi.academy/blog/skills-taxonomy-refresh-timing-annual-vs-biennial-at-25-year-half-life.php/index.md