What does missing evidence mean in CV screening?
Missing evidence means the material available to the reviewer does not establish the capability being assessed. It does not prove that the candidate lacks that capability. A CV is a selective summary, a cover letter has limited space, and an application answer responds only to the question asked.
That distinction does not require a recruiter to progress every uncertain candidate. It requires the reviewer to describe the uncertainty accurately. “Not evidenced” is a statement about the submitted material. “Does not meet the requirement” is a conclusion that needs a defined criterion and sufficient evidence.
Articles 13 and 14 of the EU AI Act address transparency about the capabilities and limitations of high-risk AI systems and effective human oversight. They do not prescribe this four-state framework. The framework is Skilltage's editorial method for helping reviewers interpret uncertainty without turning an absent signal into a negative fact.
Which evidence states should reviewers keep separate?
Use four states because each calls for a different reviewer action:
| State | What the reviewer can observe | Safe interpretation | Next action |
|---|---|---|---|
| Supported | The source contains relevant, credible evidence for the criterion | The submitted material supports the capability to the stated extent | Record the source and continue review |
| Not evidenced | The source does not address the criterion, or the detail is too vague | Capability is unknown from the available material | Inspect other sources or clarify if the criterion matters |
| Contradicted | A source clearly conflicts with the criterion or another submitted source | There is a real conflict, not merely silence | Show both sides and resolve the conflict through human review |
| Analysis failed | A document was unavailable, unreadable, or not processed reliably | The system cannot assess the source | Retry, inspect manually, or obtain a usable source |
“Not evidenced” and “analysis failed” concern information quality. “Contradicted” concerns conflicting information. None of these states should silently become an automatic rejection.
How do these states appear across candidate inputs?
The same requirement can produce different states depending on the source:
- CV: a certification is not listed. That is not evidenced unless another source confirms that the candidate does not hold it.
- Cover letter: the candidate says they have led a team, while the CV describes only individual-contributor roles. That is contradictory and needs clarification.
- Application answer: the candidate gives a concrete example of using the required equipment. That may support the requirement even when the CV is brief.
- Failed document: a scanned attachment cannot be read. That is an analysis failure, not evidence about the candidate.
The ICO's recruitment-AI audit work emphasizes data accuracy, minimisation, transparency, and meaningful human involvement. Those concerns are easier to act on when reviewers can see which source exists, what it says, and where uncertainty remains.
When should a reviewer inspect, clarify, or mark a requirement unmet?
Inspect the source first when the evidence is weak, surprising, or summarized by software. Clarify when the requirement matters and a short factual answer could resolve the gap. Continue without clarification when the missing detail is only a preference or can be assessed later at lower candidate effort.
Treat a hard requirement as unmet only when all three conditions hold:
- The requirement was approved before the candidate was reviewed.
- The evidence needed to meet it was defined and proportionate.
- The available source confirms that the candidate does not meet it, or a required clarification has produced that conclusion.
If the role genuinely requires a current licence and the candidate confirms they do not hold it, the requirement may be unmet. If the licence is simply absent from the CV, the state is still not evidenced.
What does a worked review look like?
Consider the requirement: “Can work independently with a five-axis CNC machine after site onboarding.”
| Candidate material | Reviewer state | Reason | Proportionate next step |
|---|---|---|---|
| CV names five-axis setup and operation in two recent roles | Supported | Direct, role-relevant evidence is present | Record the excerpts and continue |
| CV says “CNC operator” without machine type or responsibility | Not evidenced | The wording does not establish five-axis independence | Ask one factual clarification or check another source |
| CV says three-axis only; application answer says five-axis lead operator | Contradicted | The sources conflict on material experience | Ask the candidate to explain scope and dates |
| Uploaded CV cannot be parsed and is not available to the reviewer | Analysis failed | No reliable source can be assessed | Recover the document or review it manually |
The framework does not award capability where none is shown. It prevents the reviewer from replacing “unknown” with an invented negative fact.
What should reviewers check before acting?
Use this checklist for every material gap:
- Is the criterion approved, job-related, and clear enough to assess?
- Which source was expected to contain the evidence?
- Is the information absent, weak, contradictory, or unavailable because analysis failed?
- Can I inspect the original source behind the assessment?
- Would one short clarification resolve a material uncertainty?
- Am I applying the same evidence expectation to every candidate for this role?
- If I mark a hard requirement unmet, what source supports that conclusion?
- Is the final progression or rejection an explicit human action?
Skilltage guidance: show requirement-level evidence, rationale, and gaps to the reviewer, while keeping the candidate's selected status under human control. The current Skilltage review UI uses met, partial, and not-met assessment labels; the four-state model above is a reviewer framework, not a claim that the product exposes four matching status values.
What are the limitations and decision boundaries?
Missing evidence can still matter. A recruiter may reasonably need proof before progressing someone for a regulated, safety-critical, or genuinely non-negotiable requirement. The principle is not “assume the candidate can do it”; it is “do not claim the candidate cannot do it without evidence.”
This framework does not determine whether a recruitment process is lawful, fair, or suitable. Human review can still be inconsistent, and a well-documented decision can still rely on a poor criterion. Organisations remain responsible for their criteria, information requests, accommodations, decisions, and candidate communication.
Start by defining acceptable evidence before reviewing CVs, then use human-controlled AI screening boundaries when software helps organise the material.
Next step
Take one current screening criterion and write four examples beside it: supported, not evidenced, contradicted, and analysis failed. If reviewers cannot agree on the difference, clarify the criterion and evidence expectation before reviewing more candidates.