Part I
Why one rubric works for every role
The rubric does not score what someone knows about their job. It scores how they work with AI, and that is the same capability whether they sell, market, hire, analyse, build product, or review contracts.
A salesperson and a finance analyst face the same underlying question when they pick up an AI tool: do they aim it precisely, connect it into a system, catch where it is wrong, calibrate it to the risk, build something reusable, and actually ship it? Those six behaviours are universal. The domain knowledge sits on top, but the capability beneath is shared. That is why one rubric can score the whole organisation on a single scale.
What changes by role
The challenge content. Sales gets 25 cold accounts. Marketing gets a campaign brief. Finance gets a messy dataset. HR gets a role to fill. And the expected level: a leader is held to a higher bar than a junior.
What stays the same
The six dimensions and their anchors. Direction, orchestration, discernment, calibration, leverage and build depth are scored identically no matter the function. The capability is the constant.
So the rule is simple: build the challenge by function, score it by these six dimensions. A People team and a Revenue team run different challenges and land on the same comparable scale.
Part II
The two axes
The rubric is not a flat average. It works on two axes, because raw output and trustworthy output are different things.
- Build sets the band. How far they actually took it (R1 build depth, and I1 leverage) determines the tier band. You cannot reach the top by describing a brilliant idea you never built.
- Judgment positions and gates. The quality dimensions (A1 direction, A2 orchestration, J1 discernment, J2 calibration) place them within the band, and critically can only pull the tier down, never up.
This is deliberate. A polished one-shot output cannot outrank a working system, and an impressive but dangerous build cannot outrank careful, clean work. Impressive is not the same as trustworthy, and the two-axis design is what keeps the two apart.
Part III · the core
The six dimensions, in detail
Each is scored 0 to 4 against the anchors below, with evidence cited from the submission. For each, here is what it measures, why it is universal, the full scale, how it looks across functions, and what a scorer looks for.
A1Direction
Applied capability
weight 0.20
Did they aim AI before firing? Whether the person framed the problem precisely and briefed the AI deliberately, rather than typing a vague request and accepting whatever came back.
Why universal: every role lives or dies on framing the right problem. The person who briefs AI like they would brief a sharp junior gets a sharp result; the person who types "write me X" gets generic mush, in any function.
The scale
0
No framing. A vague, generic request. Takes the first output as-is.
1
Minimal framing. A little context, but broad and undifferentiated.
2
Basic segmentation. Breaks the problem into a couple of parts, one angle each.
3
Considered. Clear problem definition, a segmented approach, some prioritisation of what matters.
4
Sharp thesis. A precise, prioritised framing and a tailored brief per segment. Deliberately deprioritises low-value paths. Aims before firing.
Across functions
Sales
Defined the ICP and segmented the accounts before any outreach.
Marketing
Fixed the audience, positioning and single message before generating.
Finance
Framed the exact question and the decision it informs before touching the data.
People / HR
Defined the real success criteria for the role before screening.
Scorer looks for: a stated target or thesis, a specific brief to the AI, and evidence they deprioritised something. Vague in, vague out.
A2Orchestration
Applied capability
weight 0.15
A connected system, or one-off prompts? Whether they chained AI into a multi-step process where each stage feeds the next, or fired isolated prompts and stitched the results by hand.
Why universal: leverage in any role comes from systems, not single acts. The person who connects steps into a pipeline produces at a different scale from the person doing everything one prompt at a time.
The scale
0
Single prompt. One prompt, copy the output, done.
1
A few prompts. Manually stitched, no real flow between them.
2
Loosely linked. Research and output connected, some sequencing.
3
Connected process. A multi-step flow where one stage's output feeds the next.
4
Chained pipeline. Branching and conditional logic; handles variation and edge cases on its own.
Across functions
Sales
Research, enrich, message, follow-up that branches on reply.
Marketing
Brief, draft, channel variants, repurposing, as one flow.
Finance
Pull, clean, analyse, narrate, chained end to end.
People / HR
Parse, screen, rank, tailor outreach, as a pipeline.
Scorer looks for: how many steps actually connect, whether there is branching, and whether it could run on a fresh input without being rebuilt.
Did they catch where AI was wrong? Whether they caught fabrications, hallucinations and low-quality output, and kept them out of the final work, or shipped them as if true.
Why universal: AI confidently invents facts in every domain. The capability to not trust it blindly is the single most important safety behaviour anywhere, and it is what the gate keys on.
The scale
0
Shipped fabrications. Invented facts or hallucinated detail passed straight through. No verification.
1
Caught the obvious. Got the worst errors, missed subtler fabrications. Errors still present.
2
Reviewed. Caught most, but verification was inconsistent.
3
Systematic checking. Treats AI output as a draft; most errors caught by habit.
4
Verification built in. The process itself confirms sources and facts; AI is never trusted blind.
Across functions
Sales
Checked that each personalised claim about an account was actually true.
Marketing
Verified every stat and claim before it could be published.
Finance
Validated AI's calculations and assumptions against the source numbers.
Legal
Confirmed AI-cited cases and clauses actually exist and apply.
Scorer looks for: an explicit verification step, and whether anything false slipped into the final work. A score of 0 or 1 here triggers the gate.
Did they match AI to the context and the risk? Whether they judged the right level of aggression, polish, disclosure and care for the audience, the brand and the stakes.
Why universal: every function has a line that AI makes easy to cross at scale, spammy outreach, off-brand content, false precision, unfair screening. Calibration is knowing where the line is and designing inside it.
The scale
0
Tone-deaf. Blunt and mass; ignores risk, compliance, audience, sustainability.
1
Some awareness. Misjudges the line in places.
2
Mostly appropriate. Generally calibrated, a few misjudgements.
3
Well-calibrated. Matched to audience and risk, sustainable at volume.
4
Sophisticated. Compliance, audience and sustainability designed in from the start. Scales without harm.
Across functions
Sales
Honest, sustainable volume, deliverability and compliance considered.
Marketing
On-brand, audience-appropriate, no over-claiming.
Finance
Flagged uncertainty, no false precision, appropriate confidence.
People / HR
Fair, bias-aware, respectful of candidates.
Scorer looks for: evidence that risk, audience and sustainability were considered, and whether this would scale without causing harm. A score of 0 caps the tier.
Reusable, or one-time? Whether they built something that compounds and transfers to others, or produced a single output that helps once and is gone.
Why universal: the difference between a strong individual and a force multiplier, in any team, is whether their work makes everyone else better. Leverage is what turns a good operator into someone worth deploying across the team.
The scale
0
One-time. A single output, no reuse.
1
Fragments. A saved prompt or two they could reuse themselves.
2
Personal templates. Reusable components for their own work.
3
Shareable asset. Something a teammate could use with a little guidance.
4
Team system. Documented and transferable; the whole team can adopt it. Multiplies others.
Across functions
Any role
Could a teammate pick this up tomorrow and run it? Is it documented? Does it compound, or does it die when this person is on leave?
Scorer looks for: reusability, documentation, and whether the value extends beyond the individual. I1 = 4 is also a requirement for the top tier.
How far did they actually take it? The distance from describing an intention to shipping a system that runs. Demonstrated over declared.
Why universal: talk is identical across every function; output is not. This dimension is what stops a confident description outscoring a real, working thing. It is the backbone of the whole instrument.
The scale
0
Described. "I would do X." Intent only.
1
A plan. An outline or approach, not yet executed.
2
One-time deliverable. A real output, produced once by hand.
3
Repeatable workflow. A defined, re-runnable process, run manually.
4
Running system. An automation that executes on its own.
Scorer looks for: what actually exists in the submission. Did they build it, or describe building it? Build depth and leverage together set the tier band.
Part V
The gates
Judgment can only pull a tier down. A build that fails on safety is a liability, not an achievement, and the gates enforce that no matter how impressive the system looks.
J1 = 0 ships fabrications. Cap at Tier 2. Flag: do not deploy.
J1 = 1 errors amplified by automation. Cap at Tier 3. Flag: needs a verification layer before live work.
J2 = 0 scorched-earth, compliance or audience ignored. Cap at Tier 3. Flag: needs guardrails.
The logic is the same in every function: an automation that confidently produces false or harmful output at scale will damage the business faster than no automation at all. It must rank below the careful person who shipped something clean.
Part VI · the role layer
How the same tier reads by role
The score is role-independent. The meaning is not. The same Tier 4 is a rising star in a junior and the baseline expectation in a leader. This is the layer that turns a score into an action.
Expected tier by level
Level (any function)
Expected
Because
Junior / entry IC
Tier 1 to 2
Competent, judicious use of AI is the bar.
Mid IC
Tier 2 to 3
Should be structuring and starting to systematise.
Senior IC
Tier 3
Should own a repeatable workflow.
Manager / Lead
Tier 3 to 4
Should be building leverage for the team, not just self.
Director / Leader
Tier 4
Should think in systems and multiply the org.
The action is the gap
Compare demonstrated tier against expected tier. The off-diagonal cases are where the value is:
- Two or more above expected: deploy. A junior at Tier 4 or 5 is hidden capability. Roll out their system and fast-track them.
- Two or more below expected: develop. A manager at Tier 1 or 2 is a real gap, more urgent because of the seat they hold.
- On the diagonal: on track. Performing at the level the role predicts. Maintain.
This is why the assessment is sold to the organisation as a map. The tier alone is data; the tier against the role is a decision.