Trust & Assurance
Explainable by design.
CareerPathIQ uses advanced AI where contextual reasoning adds value, deterministic logic where rules provide greater certainty, and evidence as the foundation for both.
Controlled intelligence
Advanced AI inside an evidence-based and auditable framework.
The model contributes reasoning. It does not own the evidence, the deterministic rules, the governance architecture or the final decision.
Operating principles
Twelve commitments for responsible use.
These principles guide how CareerPathIQ separates evidence, rules, interpretation and human accountability.
Evidence before AI opinion
Recorded evidence is the foundation. AI does not manufacture the facts it is asked to interpret.
Rules before generative reasoning
Known checks use deterministic logic where that provides greater certainty and repeatability.
AI for interpretation, comparison and explanation
Advanced reasoning is used where context matters, within defined evidence and role boundaries.
No manufactured candidate evidence
Employers, achievements, dates, qualifications and outcomes must come from the person’s record.
Explain material findings
Important findings should expose their evidence basis and the reasoning applied.
Make uncertainty visible
The platform distinguishes a genuine gap from missing, ambiguous or conflicting evidence.
Separate fact from interpretation
Direct evidence, derived conclusions and suggested actions are labelled differently.
Meaningful human oversight
People can review, challenge and decide rather than merely approve an automated output.
Record material activity
Material evidence, findings, reviews and changes can be recorded for accountability.
Govern models and model changes
Models are controlled components, tested before material changes are introduced.
Protect personal information
Access and use remain scoped to purpose, permission and appropriate governance.
Test and monitor continuously
Performance, limitations and unintended effects require ongoing review, not a one-off assertion.
Architecture
The model is one governed layer, not the whole system.
Evidence and rules constrain the reasoning layer; explainability and human review govern what happens next.
Decision-support architecture
- 01
Career Evidence
- 02
Structured Career Record
- 03
SOC + ESCO + Role Intelligence
- 04
Deterministic Rules
- 05
Evidence Analysis
- 06
Advanced AI Reasoning
- 07
Explainable Finding
- 08
Human Review
- 09
Auditable Outcome
Fact / Interpretation / Recommendation
Different kinds of output should never be blurred together.
FACT
Directly supported by stored evidence.
INTERPRETATION
A conclusion derived from the available evidence.
RECOMMENDATION
A suggested action for human consideration.
Model-enabled, not model-dependent
Govern the capability, not just a model name.
CareerPathIQ governs models as components within the platform. Providers or models may change after testing without changing the evidence boundaries, deterministic controls, review requirements or governance architecture around them.
What remains controlled
- Permitted purpose and data boundaries
- Evidence and source precedence
- Deterministic rules and repeatable checks
- Testing, review and change controls
- Human accountability for consequential decisions
Uncertainty is information
A responsible system should know when the evidence cannot support certainty.
CareerPathIQ is designed to surface uncertainty rather than hide it behind confident language.
Fairness
Assess relevant evidence, not presentation advantage.
A polished CV can make relevant evidence easier to find, but CV quality is not candidate quality. CareerPathIQ keeps presentation quality separate from what the person’s recorded experience, skills and qualifications actually support.
Fairness requires boundaries
- Use requirements relevant to the stated purpose
- Distinguish missing evidence from missing capability
- Keep self-insight profiling out of third-party selection
- Review limitations and unintended effects over time
Human review and contestability
People must be able to question and improve the record.
Meaningful oversight includes a practical route to correct, add, challenge and reassess information.
Correct inaccurate data
Add missing evidence
Question a finding
Request human review
Reassess after information changes
Responsible AI should stand up to questions.
Ask about the evidence model, deterministic controls, model governance, human review or assurance material.