Interview prep — one judgment card will stay behind
Three questions. We will not ask you for a document.
What happens next: ① the AI interviews you (voice works) → ② you review the judgment card it builds → ③ you decide who may see it. It stays yours.
Already have cards here? → Open my cards
1. What should we call you?
Demo identity — no authentication, by design: a lightweight switcher lets judges test the senior and the junior roles immediately, on synthetic data. In production, owner access, visibility, and verified usage are bound to corporate SSO. This demo focuses on the knowledge-extraction and reuse engine.
No password. This is the name your alter will carry, and how you come back to your own cards.
2. How you talk — your alter will speak like this
3. How long are you here? (optional — so we dig the urgent areas first)
This statement is yours to hand to HR — a bill you present, not a screen they watch (the demo has no sign-in; production puts this behind SSO, owner-only). It counts only citations and explicit field reports (no views, no likes), so it can serve directly as a settlement basis for a knowledge-royalty policy. Rates and payment are HR policy, not product. "Did not hold" is listed too — an honest invoice negotiates better.
…
🗂 What you have left behind
🗺 Work to hand over
How would you like to leave something today?
🪜 Just ask me
You answer, we do the rest. A question is already waiting — a junior's, a document's, or today's opener.
🔨 I have something to leave
Think of one moment in the last six months when this team would have gone badly wrong without you.
We take it from here — the next questions follow what you just said, and you never have to pick a method.
📦 Here, take my stuff
Hand over a procedure or just ramble — we find what is missing and turn it into questions. Nothing becomes a card until you answer.
📄 📄 Red-pen a procedure
🎙 Or just talk
Ramble while you work. We mine the judgment said in passing.
The document is not stored and never becomes cards by itself — what it is missing becomes questions in your queue, and only your answers become cards.
🧰 Ways to record — pick one, or use the suggestion above
No real names needed — say the role instead ("the HR lead", "a logistics client"). Names you do say are kept out of the card.
Questions follow the cognitive-task-analysis playbook — ACTA knowledge audit & task diagram, CDM timeline, vague-word pinning, member checking. (docs/elicitation-protocol.md)
The judgment taking shape
—
🌡 Does this go into words?
Who may see this?
It stays.
This judgment now answers even when you are not there.
Watch your alter use it
This preview is not recorded anywhere — your ledger counts only real juniors.
What you actually said — kept verbatim while this card lives
Fix this card (what you write wins over the machine)
Title
Situation
Cues — one per line
The call
Actions — one per line
Why it works
When it does not hold — one per line
When it went wrong
Not in words (delete lines the machine invented — this is your record, and only you edit it)
Rules you approve here run deterministically on the protocol screen. Unknown answers never downgrade — they escalate.
Rules — ALL must be true for this judgment (one per line; empty = reading only)
Rules — must ALL be absent
Priority on conflict (0 = reading, higher wins)