Chapter 8. Assessment, Levels and AI-Matching

5. Test assignment: the "one evening" rule — paid or simplified

A test assignment should provide a work-related signal, not extract free labour. A practical rule for most roles: if the assignment cannot be completed in one evening, it must be paid, simplified or replaced with another assessment method. "One evening" means a limited scope with a clear time constraint, not a hidden 12 hours of work.

Assignment type When to use Quality rule What is prohibited

Short work sample Need to verify a specific 60–180 minutes, rubric, not Using the result in the

professional skill work employer's business

Paid assignment Need deep verification of a high- Payment, boundaries, IP rules Calling the payment a "bonus"

impact skill and usage of result without terms

Simplified case Need to assess thinking structure, Synthetic case with limited Requiring a full strategic

not finished work data plan

Live case Reasoning process matters Brief, with equal conditions Applying work-unrelated stress

6. Agency and in-house recruitment: process variants

In an agency context, motivation and signal reliability are especially important because the agency manages the trust of three parties: the candidate, the client and its own team. In an in-house recruitment context, the main risk is often that the hiring manager mistakes a polished signal for readiness for the role or oversells the candidate on a role that is better than it actually is.

7. How to implement in 1 day, 1 week and 1 month

Timeframe Action Artifact Owner

1 day Add 5 screening questions on Motivation snapshot Recruiter

motivation and

constraints

1 day Add a confidence field next Scorecard update Recruitment lead

to the rating

1 week Rewrite test assignments Assignment rubric Hiring manager +

using the time-limit / recruiter

paid / simplified rule

1 week Set up the debrief question Debrief agenda Recruiter

"which signal is unreliable?"

1 month Audit rejections and offer Rejection analysis Recruitment lead

declines for motivation /

signal issues

1 month Update interviewer training Training notes HRD / lead

on personal-contribution

probing questions

Implementation checklist:

Every role has 3–5 motivation risks linked to the actual work.

Notes separate facts, interpretations and unresolved questions.

Test assignments have a time limit, rubric and usage rule.

For each must-have criterion there is at least one reliable signal.

AI summaries are not used without verifying the original assessment facts.

Roles and responsibilities

Role What they are responsible for What they must not do

Recruiter Motivation snapshot, constraints, communication, missing Set diagnoses or press on personal topics

assessment facts log

Hiring manager Role realism, threshold risks, decision on acceptable Do not oversell the role without its difficult

mismatches aspects

Interviewer Assessment facts against assigned criteria, personal- Assess motivation on sympathy alone

contribution probing questions, confidence

Recruitment lead Rubrics, assignment policy, calibration and audit Make exceptions without documentation

HR / legal Personal data, reference and assignment policy Provide generic advice without context

owner jurisdiction review

Weekly implementation pack

Artifact Minimum version Where to store

Motivation snapshot Drivers, constraints, critical constraints, role realism risk ATS notes / scorecard

Signal reliability field High / medium / low + reason Scorecard

Assignment rubric 4–5 criteria with scale anchors Interview kit

Authorship questions 5 reusable questions Interview guide

Pre-check terms checklist Decision criteria, offer risks, timeline Offer stage notes

8. Measuring: motivation metrics and signal

reliability

Metric Formula / observation What it shows How to use it

Motivation mismatch rate Rejections / leavers / How honestly the team Improve role preview and

declines with reason mismatch / all completed portrays the role screening

processes

Offer decline mix Reasons for offer Where late-stage risk was Strengthen pre-check

declines visible earlier on terms

Low-confidence decision Decisions with low How often the team decides Add a next-step rule

rate confidence / all decisions on weak assessment facts

Test completion rate Completed assignments / Respectfulness and realism Shorten or pay for

sent assignments of the assignment assignments

Assignment defence mismatch Strong written work + weak defence Risk of unattributed Add a defence stage

decisions

Using HarmonyATS as an example: the rejection report helps you see rejection reasons by stage; the funnel with details shows where candidates drop out after an assignment or a manager interview; the candidate movement report shows delays between a signal and a decision. These reports are only useful alongside qualitative notes — the numbers alone do not tell you whether the signal was genuinely reliable. Diagnostics from reports:

For each vacancy, the recruiter records a motivation snapshot: motives, constraints, critical constraints, compensation expectations, availability and role realism risks. For must-have criteria, the interviewer must state assessment facts and confidence. A

Frequent "other reasons" Rejection reasons not Rejection taxonomy Update the reason list and

structured training

9. Good and difficult cases

Situation Risk What to check Recommended How to document Metric / next

action step

Strong candidate who Early departure Task expectations, Realistic job preview before "Motivation risk: Early attrition / offer decline

wants "strategy" but tolerance for offer strategy-to-execution

the role is 80% execution operational work mix"

execution

Candidate asks for a Title expectations above What decisions and influence Discuss scope, level "Title expectation Decline reason: level /

high title the role's scope they need and growth plan clarified" title

Candidate cannot show No assessment facts Can an anonymised case Use case defence / "Portfolio limited by Confidence after

portfolio due to NDA be discussed? references confidentiality" alternative

assessment

60% of candidates do Assignment overload Time, clarity, stage, Simplify or pay "Assignment redesign Completion rate

not return the test compensation needed"

assignment

Hiring manager says Vague suspicion Which fact triggers doubt Translate into personal- "Concern reframed into Vague-feedback rate

"I just don't believe it" contribution / consistency assessment question"

probing questions

AI-Matching gives a Keyword overvaluation Assessment facts against Human check; do not "AI score not sufficient" AI override log

high score but the the criterion advance on score alone

notes are weak

10. Regulation fragment: motivation and

signal reliability

For each vacancy, the recruiter records a motivation snapshot: motives, constraints, critical constraints, compensation

expectations, availability and role realism risks. For must-have criteria, the interviewer must state assessment facts and

confidence. A decision may not rely solely on a CV, AI score, general feedback or an unverified test assignment. If signal

reliability is low against the threshold criterion, the responsible person assigns a next step or records a rejection on

work-related grounds.

A test assignment is permitted only when a purpose, time constraint, identical brief, evaluation rubric and usage rule are in

place. If the expected time exceeds one evening, the assignment must be paid, simplified or replaced with another

assessment method.

Chapter AI prompts

The full prompts are in Appendix H to keep the chapter readable while the working instructions are collected in one

place. AI helps to structure data and drafts but does not make decisions on hiring, rejection, level, compensation or

legal obligations.

Task Full prompt Human check

Work through motivation risks Appendix H, H4.9 Check facts, privacy, tone and decision

owner before using the result.

Assess signal reliability Appendix H, H4.10 Check facts, privacy, tone and decision

owner before using the result.

Redesign a test assignment Appendix H, H4.6 Check facts, privacy, tone and decision

owner before using the result.

11. Maturity model

Level What it looks like

Chaos Motivation is assessed by enthusiasm alone; signal reliability is never discussed

Managed process There is a motivation snapshot, personal-contribution probing questions, time-boxed assignments

Assessment-based Assessment and confidence are separated; the debrief discusses missing assessment facts

Fair and respectful Assignments are short / paid, role preview is honest, notes are work-related

AI-supported operations AI helps to find gaps and questions; the decision stays with people

Key takeaways from the chapter / where to go next

What you need Where to find it

Terms and short definitions Appendix B. Mini-glossary

Ready-made regulations and SLA Appendix D. Ready-made regulations and processes

Working templates Appendix E. Template library

End-to-end examples BDM and Backend Appendix F. End-to-end cases BDM and Backend

Full AI prompts for the chapter Appendix H. AI prompt library, section H4. Role, criteria

and assessment

In short: this chapter covers motivation, hidden information and signal reliability. If you need a ready-made artifact,

take it from the appendices; the chapter itself covers the application logic.

How to use this chapter

Working scenario. Two candidates look almost equally strong. One Backend Engineer writes code quickly,

but waits for detailed requirements and rarely considers architectural consequences. The other is slower at live

coding, but identifies risks independently, proposes trade-offs and understands how a decision affects support. Both

CVs say "Senior". In practice, their levels may differ greatly.

What has actually gone wrong here: the team tries to assign a level by job title, years of experience or confidence

in the interview. A level is not seniority or a fancy title. It is the expected complexity of work, autonomy, quality of

decisions, communication, influence and business thinking in a specific company.

This chapter is useful to the recruitment lead, HRD, hiring manager, agency account manager and interviewers.

It can be used as a standalone working module: take the universal matrix, adapt it to your company, set up a scorecard,

bring in AI Studio / AI-Matching as a hypothesis and introduce a human check before the final decision.

Remember in one phrase: a level describes the expected complexity of the work, not the impressiveness of the CV.

Example type How to use it What to record

External candidate Check market language against internal Past job title vs actual scope

level

Internal mobility case Compare internal and external Same criteria, different assessment fact

standards sources

Workshop: 90-minute level calibration workshop

The workshop is not for HR to "bring a ready-made matrix" but for hiring managers to agree on what the levels

mean. If the matrix has not been walked through on real examples, it quickly becomes a pretty table that

only gets opened before a performance review.

Time Block What the facilitator does What the team does Output

0–10 min Role scope Reminds the team of role Clarifies which roles fall inside / Boundary of the matrix

families and the business task outside

10–25 min Real tasks Asks for 8–12 real tasks of Groups tasks by complexity Task complexity map

the role

25–40 min Level scale anchors Shows the universal Describes Junior / Middle / Draft scale anchors

axes through work Senior via examples

examples

40–55 min Calibration profiles Gives 3 anonymised Rates level along each axis Discrepancy log

profiles

55–70 min Threshold Asks what cannot be Compensates identifies Threshold list

compensated must-have criteria and

trainable gaps

70–82 min Interview assessment Links scale anchors to Decides where to check each Assessment facts plan

facts questions and assignments axis

82–90 min Decision rule Records the responsible Records the approval rule Level decision rule

person and review cycle

A ready-to-use agenda for the invitation:

Subject: level calibration for [role family].

Objective: agree on how Junior / Middle / Senior differ in practice so that recruitment, interviews and offer decisions are

based on the same criteria.

Preparation required: each participant brings 2 Middle-level task examples and 1 example of an error when a candidate's /

employee's level was assessed inaccurately.

Meeting outcome: draft level matrix, thresholds, assessment facts plan for interviews and the decision owner.

Facilitation rule: if a participant says "a Senior should be strong", the facilitator returns to the question "in

which task is that visible?". If it sounds like "we just feel it here", the facilitator asks for observable behaviour.

If the argument turns on years of experience, the facilitator redirects the debate to autonomy, complexity,

influence and business impact.

Level calibration across different functions

The universal matrix must be consistent in logic but alive with examples. What follows is not a "market truth" but a

way to explain to managers that a level is defined by the real scope of the work.

Function Middle looks like this Senior looks like this Typical assessment mistake

IT / software engineering Owns a feature / service area, Considers system trade-offs, Considers someone a Senior because

understands edge cases, writes influences engineering they know more technologies but

maintainable solutions practice does not hold architectural decisions

4. Competencies: how to link skills and levels

A competency is a measurable work behaviour or skill that can be observed through assessment facts. In Chapter 6, skills are described via output, assessment facts and scale anchors. Here we add the level: at which level a person should demonstrate the skill.

4. Competencies: how to link skills and levels

A competency is a measurable work behaviour or skill that can be observed through assessment facts. In

Chapter 6, skills are described via output, assessment facts and scale anchors. Here we add the level: at which level

a person should demonstrate the skill.

Competency Junior Middle Senior

Communication Reports status and asks Manages expectations and risks Creates shared clarity among

questions in their zone stakeholders

Accountability Carries out tasks and reports Delivers results in their zone Changes the system so failures

problems do not recur

Judgement Follows clear rules Chooses a reasonable Makes decisions under

compromise uncertainty and explains the

consequences

Ability to learn Learns from feedback Builds their own learning cycle Helps others learn and refreshes

practices

Domain professional skill Applies basic methods Solves routine and moderately Designs approaches for complex

complex tasks tasks

Business thinking Understands the task Sees impact on the team / Links decisions to capacity,

customer risk, revenue or quality

Rule: not all competencies need to be equally strong. For an expert individual contributor, deep professional skill

may carry more weight; for a manager, it is decision maturity, feedback, conflict maturity and influence.

But reliability and ethical behaviour cannot be fully compensated by strong expertise.

Examples by role family

Role family More important at Junior More important at Middle More important at Senior

Software engineering Basic coding discipline, learning, Independently delivers features, Designs trade-offs, improves

asking for debrief debugs, sees local risks systems, mentors, handles

ambiguity

Sales Follows process, learns Runs discovery, manages funnel, Builds account strategy, handles

product, qualifies simple leads negotiates typical deals complex stakeholders, protects

value

Marketing Executes campaigns and basic Reads funnel, runs Connects positioning, channels,

reports experiments, adjusts budget analytics and business

or content results

Finance / operations Follows control procedures, Owns the process, finds Design controls, risk model and

executes work accurately errors, improves reporting decision-ready insights

Recruitment Screens per rubric, keeps Manages role funnel, criteria Design hiring process, calibrates

notes and SLA and hiring manager discipline managers, reads analytics

Customer success Handles work instructions and Owns client portfolios, Builds retention system, escalation

tasks and customer notes diagnoses adoption and risk and stakeholder strategy

Assessment facts by level

Axis Junior assessment facts Middle assessment facts Senior assessment facts

Autonomy Knows when to ask for help Structures own tasks and Defines ambiguous work for

dependencies self and others

Complexity Handles known cases with Handles mixed constraints and Handles unclear problems with

exceptions system impact

Influence Cooperates within a task Coordinates adjacent Changes process, mentors or

stakeholders aligns groups

Judgement Follows rules Explains trade-offs Chooses approaches under uncertainty

Business thinking Understands the immediate goal Connects work to team Connects work to customer, revenue,

outcomes risk or quality

How to distinguish "ready for the level" from "can grow into the level"

One of the most common hiring mistakes is to hire a candidate as a Senior because they could potentially become

a Senior. Potential matters, but the offer, compensation, onboarding and expectations must match the currently

proven readiness.

Signal Ready for the level Can grow into the level Decision risk

Autonomy Has previously led a similar scope Reasons well but has not held Needs a responsible onboarding and

without constant oversight responsibility lower initial scope

Complexity Worked on real complex cases Worked on academic / limited Must not be sold to the hiring

examples manager as a proven Senior

Practical rule: if a candidate is "almost a Senior" but the assessment facts still show Middle+, this is not a bad candidate. It is a good reason to honestly discuss a different level, a growth plan and the first 90 days. The bad decision is not the lower level itself, but a silent mismatch: when the candidate is sold one thing, the hiring manager expects another and onboarding does not close the gap.

5. Scorecard for level decisions

Candidate level hypothesis: [Middle / Middle+ / Senior]. Strong assessment facts: [criteria and examples]. Weak or missing assessment facts: [criteria]. Reliability concerns: [personal contribution, context, confidence]. Decision: [advance / reject / pause / different level]. Responsible person: [name].

Candidate level hypothesis: [Middle / Middle+ / Senior].

Strong assessment facts: [criteria and examples].

Weak or missing assessment facts: [criteria].

Reliability concerns: [personal contribution, context, confidence].

Decision: [advance / reject / pause / different level].

Responsible person: [name].

End-to-end example: level decision

Field BDM in B2B SaaS payments Backend Engineer

Level hypothesis Middle+ / Senior BDM Middle+ / Senior Backend Engineer

Strong assessment facts High discipline in qualification, good lost- Good work accountability, strong

deal analysis, can explain stakeholder map idempotency example, mature incident

learning

Weak assessment facts Payment domain knowledge still at a Code review lacked edge-case

conceptual level, no deep examples tests; payment-specific consistency

not yet verified

7. Where to use and where not to use the level matrix

Use the matrix when there are several levels within a role family, a pay band, a career path or disagreement

between interviewers. Do not use it as rigid bureaucracy for a single hire if the role does not have a tiered structure.

In that case, decision criteria and onboarding support level are sufficient.

8. AI Studio and

AI-Matching: what can be automated Using HarmonyATS as an example, AI Studio and AI-Matching can serve as a structuring layer: describe criteria, weights, assessment facts, red flags, questions and a recommendation hypothesis. This is not a replacement for the interviewer, nor is it a legally safe automatic filter.

High-volume Standardise minimum AI / keyword sorting replaces Use the same

thresholds assessment criteria and review

7. Where to use and where not to use

the level matrix

Use the matrix when there are several levels within a role family, a pay band, a career path or disagreement

between interviewers. Do not use it as rigid bureaucracy for a single hire if the role does not have a tiered

structure. In that case, decision criteria and onboarding support level are sufficient.

Use Do not use

Junior-Middle-Senior hiring When the level does not affect scope and compensation

Interviewer calibration To justify a decision already taken

Offer alignment To haggle over a job title without changing the work

Internal mobility To block progression without assessment facts

AI-Matching methodology To automate the final decision

8. AI Studio and AI-Matching: what can be

automated

Using HarmonyATS as an example, AI Studio and AI-Matching can serve as a structuring layer: describe criteria, weights,

assessment facts, red flags, questions and a recommendation hypothesis. This is not a replacement for the interviewer,

nor is it a legally safe automatic filter.

1. Define Upload only work-related criteria and their Hiring manager + recruiter

criteria weights

2. Add level scale Describe expectations for each level on each axis Recruitment lead

anchors

AI-Matching governance checklist AI-Matching is especially useful in two cases: when you need to quickly break down a CV by criteria and when interviewers find it difficult to keep the level matrix in mind. But the higher the decision impact, the tighter the control must be. The AI output must be visible, verifiable and correctable.