Skills taxonomy levels are where career records most often overstate themselves without meaning to. A rating like mid or senior looks like a measurement and reads like a measurement, but it has no agreed scale behind it — the same word means something different in the next company, the next country and the next profession.
This article sets out how skills are usually grouped, why a self-assessed level and a credential are two separate facts, what a rating without a named standard actually asserts, and how to store a skill so that filtering on it means what it says.
How can skill levels be described without overstating them?
By attaching the level to a named standard, or by dropping the level and letting the credential speak.
An unqualified rating is a claim whose meaning cannot be checked by anybody, including the person who made it. The word mid on its own asserts a position on a scale that was never defined, and two readers will place it differently. A rating that names its framework — this person’s own assessment against a published skills framework, at this version — is at least a statement that can be compared to another statement made the same way.
The plainest option is often the most useful. For many skills, the honest description is a qualifier rather than a rank: has used this in production, has studied this, is currently learning this, trained others in this. Those describe evidence instead of position, and evidence survives being read by somebody who does not share your vocabulary.
The thing to avoid is the composite. A single number that merges a self-assessment, a certificate and a number of years is arithmetically meaningless and rhetorically impressive, which is exactly the wrong pair of properties for a hiring record.
Is a self-assessed level the same as a certificate?
No, and they cannot substitute for one another in either direction.
A self-assessment is a claim about a person made by that person. It is useful, it is cheap, it covers everything, and it is only as good as the person’s calibration. A credential or certificate is a claim made by somebody else, against criteria that were written down before the assessment happened. It is expensive, it covers a narrow slice, and it is verifiable by a third party.
Three practical differences follow. The self-assessment exists for every skill, while the certification exists only where a certification body has decided the skill is worth certifying — and those decisions follow commercial and regulatory logic as much as they follow what matters in the work. The self-assessment goes stale slowly, while the credential has whatever currency its issuer gives it. And the self-assessment is cheap to revise, while a credential cannot be revised at all, only replaced.
Treating a certificate as an upgrade to a rating therefore loses the rating, and treating a rating as equivalent to a certificate invents evidence. A record that keeps both, clearly labelled, is strictly more informative than one that keeps either alone.
What kinds of skill are usually recorded?
Four groupings cover the ground, and they are grouped because they cannot be described with the same vocabulary.
- Language ability — measured in named proficiency scales or in specific communicative tasks, and it belongs to a framework rather than to a rating.
- Technical and tool skills — named tools and technologies, where the vocabulary changes quickly and where a version or a context is often part of the meaning.
- Industry and professional competence — the domain knowledge of a trade, frequently tied to a licensure or a professional body.
- General and interpersonal skills — communication, organisation, collaboration and the like, where almost no vocabulary is shared and where a rating is least meaningful.
Using one rating scale across all four is the common error. A scale that reads sensibly for a programming language reads absurdly for collaboration, and a scale built for interpersonal skills cannot say anything useful about which version of a database engine somebody has worked with.
| Group | Best described by | Rating works badly because |
|---|---|---|
| Language ability | A named proficiency framework | The frameworks already exist and are more precise |
| Technical and tool skills | Named technologies, with context | The name carries more signal than any rank |
| Professional competence | Issuer-backed credential plus experience | The credential determines what counts as competence |
| Interpersonal skills | Observed behaviour or examples | There is no scale anyone shares |
Why does the same level word mean different things in different companies?
Because organisations define their own ladders, and they define them for internal purposes: pay bands, promotion criteria, review cycles. None of those definitions is intended to be portable, and none of them is published in a form another organisation could use.
The result is that the same word occupies different positions in different ladders. Where one company reserves a word for people who set technical direction, another uses it for people who work unsupervised. The word is stable; the rung it names is not. Adding a region to the label does not fix this either, because the variation is within regions as much as between them.
There is a second-order effect that catches people out. Because the words carry prestige, they drift upwards over time. A word that once described a small minority of staff comes to describe a comfortable majority, and the drift is invisible from inside any single organisation. A record that stores a level word without recording where the word came from is therefore storing a fact that decays.
The only durable answer is to store the level together with the standard that produced it — the framework name, the version, and who applied it. Where that is unavailable, store no level, and let the credential and the described experience carry the claim.
How do synonyms break skill searches?
They break them quietly, one missed match at a time. A recruiter filters for a technology by its common abbreviation, and the candidate who wrote the full name is not returned. The filter reports success, the list is short, and nobody knows anything was lost.
The variation comes from four directions. Abbreviations and expansions of the same term. Different generations of a product name, where both old and new names are in active use by different people. Translation, where a term is rendered into a local language by some writers and left in its original form by others, sometimes within the same document. And plain variants in how people write a compound term, with or without a space or a hyphen.
Normalising the vocabulary solves the matching problem and creates a different one, because normalisation is a decision about what two things are the same, and that decision is not always correct. Merging two spellings of one technology is safe. Merging two closely related but distinct technologies is not, and the difference is exactly the kind of judgement that should be visible and reversible rather than buried in a lookup table.
For developers: skill fields and normalisation
Model a skill as a statement about a person rather than as a value in a column. The statement has a subject — the skill itself, identified independently of how it is written — and a claim attached to it, which is either a rating with its framework or a credential with its issuer.
Four habits keep the model honest. Keep the skill identity separate from its display text, so a synonym can be added without rewriting any record. Keep the rating’s provenance in the record, including which framework and version produced it. Never store a rating where no framework applies — leave the claim type empty instead, and let the presence of a credential fill the gap. And treat the industry and the country as filters on the vocabulary rather than as decoration, because a skill list containing a discipline unrelated to the role is a strong signal that either the record or the generator is drawing from the wrong pool.
For test data, the value is in the awkward cases: a skill written as an abbreviation in one record and in full in the next, a language claim tied to a framework, a credential whose name is common but whose issuer differs by region. Ratings in the sample records on this site are demonstration values used to exercise display and filtering logic. They are not an assessment of anybody’s ability, and the labels attached to them describe nothing about a real person.
Next steps
Take the skill filter in your product and run the same search twice, once using an abbreviation and once using the term written in full. If the two searches return different sets, you have a normalisation gap and it is costing you candidates rather than failing loudly. The career profile tool generates records whose skills are drawn from the industry and country of the record, which is the quickest way to see what a mismatched vocabulary looks like. The credential side of the same problem is taken apart in the certification data article, and which skills a role title is expected to bring with it is the other half of the matching.