Dimension D2 of 6

Evidence sourcing

Did they demand evidence for the claim that mattered?

The probe

What the session puts in front of them

The assistant asserts claims that matter to the answer, with nothing behind them until someone asks.

Each live bank ships twelve authored cases and a role opens on one of them, so a leak costs that case rather than the bank. Rotation within a role is the next integrity work; everyone invited to one role still meets the same case, and we would rather name that than imply otherwise.

Evidence

Enacted, not narrated

Whether sources are demanded for a specific claim rather than requested in general — and whether at least one is actually opened.

The finding is anchored to a moment you can open: a capture timestamp, an AI transcript turn, an artifact diff, or a debrief answer.

The two outcomes

What a pass and a fail look like

No composite. Each dimension resolves on its own evidence, and you can read the moment it turned.

Fail

When it goes wrong

Claims flow into the deliverable as facts, with nothing asked for behind them.

Pass

When it goes right

A source is demanded for a named claim, opened, and what it did or did not support is stated in the deliverable.

Never scored

What we ignore either way

Prose quality, prompt syntax, tool trivia, speed, tone of voice, and how much AI the candidate used. Volume of usage is not a virtue; judgment about the output is the test.

How it is scored

From session to D2 finding

The same four steps for every dimension, so a result means the same thing across roles.

The situation is set up

The assistant asserts claims that matter to the answer, with nothing behind them until someone asks. Each live bank ships twelve authored cases and a role opens on one of them; rotation within a role is roadmap, so until it lands everyone invited to one role meets the same case.

The session is captured

Workspace events, the AI exchange, the deliverable's history and — where the candidate allows it — tab-scoped screen capture and microphone, recorded together so a reviewer can read them against each other.

A reviewer writes the finding

A person reads the session against the bank's answer key and writes the finding themselves, with the evidence excerpt attached. There is no automated scorer in the product — not a weak one, none — so this step is the scoring rather than a check on it.

And releases it deliberately

A release is refused until all six dimensions are written and the reviewer has confirmed the report says nothing the candidate should not read about themselves. It is enforced in the database rules rather than by policy: no client can write a result, and no client can read one that is not released.

Previous: Problem framing · Next: Delegation boundary

See D2 scored on a real session

The sample report shows all six dimensions with the evidence excerpts attached, exactly as an employer receives them.

Open your first role Ten attempts a month against a live item bank, with a human-written report on every one.