Key terms
Workerbee’s vocabulary — the concepts you’ll see across the Work Intelligence Console (WIC), the API, and every AI connector.
| Graph | A database of occupations, skills, and education connected as nodes and relationships, so the platform can reason across related concepts instead of doing flat keyword lookups. |
| Canonical Occupation / Skill | A normalized, deduplicated representation of a job title or skill, so "Software Engineer" and "SWE," or "JS" and "JavaScript," are recognized as the same thing. |
| Structuring | The step where an unstructured resume or job description is parsed into consistent fields (skills, titles, education) — not chunked-and-embedded, but converted into facts on the Graph. |
| Success Profile | A standardized definition of what "good" looks like for a role — required and nice-to-have skills, experience, and other criteria — so every candidate is evaluated against the same bar. |
| Worker Passport | A person's structured profile record: parsed resume data, canonical skills, work history, and other attributes, normalized so it can be matched against roles. Each capability keeps its evidence, with source and level. |
| Evidence | Any record that links a person to a capability: a document, a conversation, an assessment result, a manager's input, or work output. Every piece of evidence keeps its source and carries an evidence level. |
| Evidence level | How strong the proof is that a person has a capability. Five levels, weakest to strongest: Inferred, Self-reported, Manager-observed, Assessed, Demonstrated. Each piece of evidence is shown with its level and its source. The level is set by what the evidence shows, not by where it came from. |
| Confidence | Shown as strong, moderate or weak, not a percentage. It combines the evidence level, how important the requirement is in the Success Profile, and how recent the evidence is. Planned |
| Outcome | What happened after a decision: hired, advanced, moved, retained, exited. Recorded against the evidence and the Success Profile version used. An outcome is not an evidence level, and Workerbee does not claim a recommendation caused it. |
| Talent Intelligence | The layer that compares a Worker Passport against a Success Profile (or runs broader workforce queries — mobility, succession, skill gaps) and returns ranked, evidence-backed results. |
| Job Match / Ranking | The output of comparing candidates against a Success Profile: a rank, a score, and a human-readable rationale — never a black-box number. |
| Decision Audit | A retrievable record of what was asked, what was answered, and what evidence was used for a given ranking or query — built for compliance, not added after the fact. |
| Candidate Anonymization | Network/external candidate identities are withheld (shown as initials) until the candidate consents to being identified. A product guarantee, not a demo limitation. |
| k-anonymity suppression | Small result sets are suppressed so a query can't be used to single out one individual by elimination. |
| MCP (Model Context Protocol) | An open protocol that lets AI assistants and agents call Workerbee's capabilities as typed tools, instead of a one-off custom integration per client. |
| Work Intelligence Console (WIC) | Workerbee's web app for managing a company's understanding of its work: explore the Work Graph's departments, roles and people, edit and publish Success Profiles, upload files, and manage access and API keys. |
| Connector | The authorized integration point that lets an AI client (Claude, ChatGPT, Gemini, an MCP-compatible IDE) access Workerbee on a user's behalf, set up through an OAuth-style consent flow. |
| Integration (ATS / HRIS / LMS) | A live connection to an external system that keeps a customer's Work Graph in sync automatically, as opposed to a one-time file upload. |
The evidence ladder
Every piece of evidence about a person’s capability carries one of five levels, weakest to strongest:
Inferred → Self-reported → Manager-observed → Assessed → Demonstrated
The level is set by what the evidence shows, not by the channel it came through.
| Level | What it means | Typical sources |
|---|---|---|
| Inferred | The Work Analyst infers the capability from related information; nobody has stated it. Always labelled as the weakest level. | Job title, occupation, adjacent skills in the Work Graph |
| Self-reported | The person says they have it. | Resume, profile, statements in an interview, course or LMS completion |
| Manager-observed | Someone responsible for the person’s work says they have it. | Performance reviews, manager input, references |
| Assessed | A structured test designed to measure the capability says it is met. | Validated assessments (partner instruments), graded tests, a structured live exercise in an interview |
| Demonstrated | The capability shows up in the person’s actual work. | Work systems such as GitHub, Jira, CRM or work orders |
How the level is set
- By what the evidence shows, not where it came from. An interview transcript can hold Self-reported evidence (“I’ve used dbt for two years”) or Assessed evidence (a structured live exercise).
- Performance reviews. A performance review is a manager’s account of the work, so it counts as Manager-observed.
- Courses. Course or LMS completion records exposure, not the skill, so it counts as Self-reported. A graded test inside the course counts as Assessed. Applying the skill in real work afterwards is Demonstrated.
- Assessed and Demonstrated. A test measures the skill in controlled conditions. Demonstrated shows it in real work.
- Internal people and external candidates. The same ladder applies to both. Internal people simply have more sources available, such as work systems and manager input.
How evidence progresses
A capability usually starts as a claim and gains strength as better evidence arrives. Each new piece of evidence is added, not replaced, so the record shows the whole progression with every source visible.
A candidate for Acme’s Senior Data Analyst role lists Python on their resume (Self-reported). In the interview they describe a forecasting pipeline they built (still Self-reported: it’s their account). They then complete a Python assessment (Assessed). After joining, their merged pull requests show Python in daily use (Demonstrated). Their Python evidence moves from Self-reported to Demonstrated, and every step stays visible with its source.
Finishing a Python course adds Self-reported evidence; it does not by itself show the skill.
The same applies to someone already working at the company.
Wei · Power BI. Today it’s Inferred, from his role and related skills, which is why the record says “Verify Power BI before advancing.” His manager confirms he builds the team’s reports: Manager-observed. He passes a Power BI assessment: Assessed. His dashboards show up in Acme’s work systems: Demonstrated.
When evidence conflicts, both pieces are kept and shown with their sources.
Level vs confidence Planned
The level says what kind of proof exists. Confidence weighs that level together with how important the requirement is in the Success Profile and how recent the evidence is. Confidence is shown as strong, moderate or weak, not as a percentage.
Levels don’t decay. Confidence does.
Taxonomy sources
The Graph’s occupations and skills are built from hundreds of thousands of job descriptions and resumes, in addition to public taxonomies from O*NET, ESCO and Lightcast.
Last updated