A Vendor Certification Is Evidence of a Stack, Not the Skill

The spreadsheet is called certification_inventory_FINAL.xlsx.

It has a column called skill.

A column called credential.

And a formula that copies the second into the first whenever the first is blank.

That formula has been there for a while.

It survived the HRIS migration.

It survived the cloud migration.

It survived the meeting where somebody briefly proposed that every engineer become an “AI Pathfinder.”

Now a platform changes.

The certificate catalogue changes with it.

And a competent employee, who has designed, operated and repaired systems through several perfectly real incidents, becomes a red cell in your skills matrix.

Because the current badge name is absent.

The employee did not become unqualified.

The data model did.

A vendor certification is evidence of a stack. A skill is the capability to do work under stated conditions.

They meet constantly.

They are still two different things.

The badge has an issuer, a boundary and a stopwatch

This gets easier to see once you treat the credential as the specific thing it is.

Candidate Information Bulletins and examination content outlines exist for exactly that purpose.

Item counts.

Passing standards.

Prerequisite rules.

CompTIA Security+ is not an abstract proposition that somebody “has security.”

It is exam SY0-701.

Up to 90 questions in 90 minutes.

Multiple-choice items mixed with performance-based ones.

Scored from 100 to 900, with a 750 cut.

No mandatory experience gate.

That record says something useful and checkable about an assessment administered by CompTIA.

It does not say a person can design every security programme, diagnose every incident, or exercise security judgment in every organisation.

Security+ can satisfy U.S. Department of Defense compliance for certain IAT and IAM roles.

Even that use has a stated context.

Which would be a great deal to learn from a badge image.

AWS Certified Solutions Architect, Associate, is even more explicit about its boundary.

SAA-C03 runs 65 questions over 130 minutes, at 720 out of 1,000.

Four domains: resilient, high-performing, secure and cost-optimized architectures.

It assesses architectural trade-offs and service configurations on Amazon Web Services.

It does not require live scripting.

That is good evidence of AWS implementation readiness.

The platform is not an embarrassing detail to scrub out of your data.

It is the point of the credential.

Red Hat makes the contrast pleasantly impossible to ignore.

RHCE EX294 requires an active RHCSA credential.

Then a four-hour performance examination, on a live Red Hat Enterprise Linux system.

Candidates write Ansible playbooks to configure managed nodes, storage, security and network services.

The resulting system state is graded.

One credential samples architecture questions.

The other samples live-system automation.

Both can support a capability claim.

Neither dissolves the difference between cloud architecture, Linux automation, AWS and Red Hat Enterprise Linux, however much your talent profile would prefer fewer fields.

So: record the credential as evidence of a bounded assessment, then map it to the capability it may support.

Do not use the credential name as the capability’s primary key.

The discipline comes first, the stack comes after

A usable model has layers.

This is inconvenient only if the objective was to keep one column.

The foundational layer holds systems thinking, analytical reasoning, communication, risk judgment, domain knowledge.

The discipline layer holds cloud architecture, cybersecurity, systems administration, data engineering.

The platform layer holds AWS service configuration, Ansible on Red Hat Enterprise Linux, and the vendor-specific methods people actually use to do the work.

A person can be strong in a discipline and new to a platform.

Another can be immediately deployable on a platform while still developing the discipline.

Those are two very different development conversations.

Collapsing them produces a rather theatrical “gap.”

CISSP illustrates the useful middle ground.

Its adaptive English examination runs 100 to 150 questions over three hours, at 700 out of 1,000.

It also requires five years of professional experience, across two or more of its eight domains.

A candidate who passes without the experience becomes an Associate of (ISC)², with six years to verify the work history.

That is a substantive issuer relationship with a real experience gate. It keeps the assessment result and the unverified work history distinct.

It is still not a universal security ontology.

Those eight domains are the certifying body’s structure for its own credential. Not an instruction to make every security role, learning plan and incident review conform to the same list.

Project management makes the parallel point from another discipline.

PMP eligibility pairs 36 months of project leadership with 35 contact hours of education, for a candidate holding a four-year degree.

The high-school route asks for 60 months.

CAPM asks for 23 contact hours and no field experience at all.

Strip the badge name off its route, prerequisites and assessment mode, and you have erased the information the credential carried.

Your skills matrix wants a noun.

A mature system lets the noun keep its paperwork: exam code, issuer, assessment date.

The credential is doing real work

None of this argues for sending everyone back to a folder of project screenshots called proof_of_cloud_expertise_NEW.zip.

Credentials make implementation readiness visible.

They give a distributed team common terminology.

They create structured learning paths.

They give a worker a signal that is quicker to interpret than a narrative of every production change they have survived.

Pay-index research tracks cash premiums for more than a thousand certified and non-certified technical skills, which is one reason employers treat a legible credential as valuable.

That matters most when a team is adopting a specific platform and needs people who can use it now.

Credentials can also record meaningful gates.

PMP runs 180 questions over 230 minutes, across people, process and business environment. Roughly half the examination reflects agile or hybrid approaches.

The PHR mixes scored and pretest items, with experience requirements that vary by education.

The SHRM-CP mixes knowledge items with situational-judgment items.

These are not decorative logos.

They are operational artefacts, with an issuer, an assessment method, a scope, and usually a maintenance relationship.

Most of them also run on a renewal cycle, with continuing-education requirements and a maintenance fee attached.

The renewal record belongs in your system, because currency is a claim about the credential relationship.

It does not make the issuer the custodian of your entire definition of cybersecurity, cloud architecture or project leadership.

When the catalogue becomes the ontology

The failure arrives quietly.

The HR platform imports badge names, because badge names are clean data. The learning platform recommends courses from the same catalogue.

The skills graph treats the badge as a skill, because the vendor filed it under skills.

Then the product changes.

Or the exam code changes.

Or the content outline changes.

And a mapping update looks like a workforce capability event.

Nothing has happened to the person’s durable ability to reason about systems, manage risk, write automation or improve an operating environment.

An EX294 result is still evidence of Ansible work on Red Hat Enterprise Linux, even when a catalogue manager prefers a new label.

The evidence has not changed.

The wrapper moved.

This is the point where a badge inventory becomes a dependency map. You have stapled your skills architecture to a product roadmap, whose long-term calm and stability are of course beyond dispute.

So review the mapping whenever the product, exam code, issuer, content outline or renewal rule changes.

Keep the older evidence intact: issuer, credential title, product or exam version, issue date, assessment mode, result, renewal status.

Then decide whether the old assessment still supports the capability relationship, supports it at a narrower scope, or needs more evidence.

That last option is not a failure by the certificate holder.

It is an ordinary finding about evidence.

Much like a CISSP candidate who passes before completing the required experience, and is recorded as an Associate rather than a full designee.

AI cannot see a stack you did not store

For years this lived in a cupboard with the learning-transcript export.

An enablement lead knew, from context, that an architecture exam and a live-system performance exam were different kinds of evidence.

A manager could ask what the person actually built. Somebody who had been paged during an incident carried an informal but useful credibility.

Then organisations started asking systems to infer skills, recommend learning and allocate internal work.

A model that sees AWS Certified Solutions Architect, Associate and writes “cloud architecture: verified” has converted product-specific evidence into a general capability assertion.

A model that sees no current badge and writes “capability gap” may just be reporting a catalogue gap.

Both outputs are tidy.

Neither is automatically true.

So the AI record needs what a careful analyst needs.

Evidence type.

Issuer.

Product or exam version.

Assessment mode.

Date.

Renewal status.

Capability mapping.

Scope.

Confidence.

Review status.

A renewal record is renewal evidence.

It is not a machine inference that a whole discipline was refreshed.

The model may propose a relationship.

It should not erase the conditions because JSON has no field called “this is more complicated than it looks.”

Evidence is plural because assessment is expensive

The obvious response is to test the durable capability directly, every time.

This has the advantage of being conceptually clean, and the minor disadvantage of operating in the world.

Low-fidelity assessments, standardized multiple-choice tests and automated coding tests, cost between $10 and $50 per candidate at scale.

Multi-rater assessment centres, with role-plays, simulations and structured rubrics, cost between $3,000 and $15,000 per candidate.

That is roughly a three-hundred-fold difference at either end of the range.

Certification is useful partly because it is a workable evidence proxy.

A serious system does not discard it.

A CompTIA score, an RHCE live-system result and a PMP experience gate each support a different kind of claim.

Put them beside your other evidence and state what each one can carry.

Performance assessment offers a model here.

Some universities evaluate authentic deliverables through anonymous faculty scoring against multi-trait rubrics.

Software applications.

Financial audits.

Clinical care plans.

Then they run proctored standardized examinations alongside those.

The formats differ because the claims differ.

And there is a limit no badge taxonomy solves. Polanyi’s paradox describes the tacit knowledge experts use without being able to state it.

Most expert decision-making in novel, unstructured settings runs on schemas nobody has written down.

A certification, a work sample and a manager review each see a different slice of that.

None of which is an invitation to give up on skills data.

It is a reason not to let the cleanest credential_name field in the export claim to be the whole person.

Keep the evidence, keep the capability

The operational routine is not glamorous.

Define foundational, discipline and platform capabilities separately.

Maintain the certification as an evidence object: issuer, credential and product version, assessment method, result, issue date, renewal.

Map that evidence to capability claims, with scope and confidence attached.

Accept validated work, portfolios, performance assessments and project history alongside it.

Review the mappings when the product or the credential changes. And put the ambiguous ones in front of a person who understands the work.

The alternative is an automated system that notices a renamed exam faster than it notices an employee who has been doing the job.

Vendor certifications stay valuable throughout.

They are just evidence of a stack.

Not a portable definition of the skill underneath it.