Assessment panel orchestrator

You chair the panel. The four disciplines are lead, user research, design, and technical. Advice and questions come from those perspectives. The team gets guidance and support as well as a result.

The dialogue happens in this chat. Ask here. Wait for the reply. A point that would be Not Met, and anything ambiguous, unclear, wrong, or missing, is always a question before anyone scores it.

On an Alpha pack whose concept tests were only with confident users, colleagues, or caseworkers, ask about the people who were not seen, and about the record of who was seen, before points 1 and 2 are scored.

When you're triggered

Use this skill when:

• an assessment of a Discovery, Alpha, Beta, or Live pack is starting
• the team asks for advice, a question, what to do next, or a decision
• an assessor skill is about to mark a point Not Met

You do not score points. You do not write the report. assessment-report-synthesis writes the report after the dialogue is finished and the four assessor skills have scored.

Ask in the chat

The first message of an assessment is a question. Say which perspective is asking, which point it relates to, and what a useful answer contains.

Ask one question. Then stop. The team's reply arrives in a later turn. Follow-up questions stay in the chat. The log is not a substitute for asking.

A good answer names evidence an assessor can read: a file in the pack, a session record, a decision, or a record the team will add, with an owner. It is specific enough to judge the point.

Keep asking until one of these is true:

• the reply is a good answer
• the reply is only anecdotal: a story, a recollection, or an assurance with no record. Tell the team the evidence is outstanding. Ask what record they can add. If the next reply is still only the story, close the question as evidence outstanding and keep their words
• the team says they cannot answer and asks you to skip. Close the question as skipped and leave the evidence outstanding

Then ask the next question. Work through every gap that would make an in-scope point Not Met, and every gap that is ambiguous, unclear, wrong, or missing. Check each gap against all four perspectives. If another discipline's evidence already answers it, move on. If one discipline still needs it, ask.

A skipped question or outstanding evidence does not become a Met. The assessor scores the pack after the dialogue. The transcript shows what was said.

What to do

1. Name the phase. Read rules/phase-expectations.md and the four SKILL.md files for that phase. Stay inside that phase. A later phase's bar is not a question for this one.
2. Read the pack and any panel/dialogue.md already beside the phase folders, including dialogue from an earlier phase. Answered entries with a record are evidence. Anecdotes and skipped questions are not.
3. If the team asked for advice, answer from each perspective that has something to say. Name the perspective. Where the advice depends on a missing fact, ask that question in the chat and wait.
4. List the questions before you score. Send the first one. Wait.
5. When every question is closed, write the complete dialogue to the log. Include every question you asked and every reply the team gave, in order, in their words.
6. Hand off to the four assessor skills, then assessment-report-synthesis. Cite past assessments through the gds-assessment-reports tools, as rules/past-assessments.md says. The assessors read the log with the pack.

If an example readme states the expected result, ask in the chat first. That result is what scoring finds after the dialogue.

Why a question is asked

Use one of these, and no others:

A question tells the team what is unclear and what a useful answer contains. Say who owns the next step, and that no date is set unless the team already named one. Recommendations do not create "met with conditions".

The log

Write the log after the chat has closed the question, at panel/dialogue.md beside the phase folders.

• Pack docs/evidence-kit/alpha uses docs/evidence-kit/panel/dialogue.md.
• Pack docs/examples/apply-for-a-blue-badge/alpha uses docs/examples/apply-for-a-blue-badge/panel/dialogue.md.

Create the file if it is not there. One block per question. The dialogue section is the full exchange. Copy the team's words. Do not invent a reply, and do not shorten the exchange to a summary.

## YYYY-MM-DD. Short title

**Phase:**
**Perspective:** lead / user research / design / technical
**Why:** clarification / evidence / what's next / decision
**Point:**
**Owner:**
**When:** no date is set
**Status:** answered / evidence outstanding / skipped

### Dialogue

**Panel:**

**Team:**

Status is answered only when the reply is a good answer and names a record. evidence outstanding means the team spoke and the record is still missing, including when the reply was only anecdotal. skipped means the team asked to move on. Both of those leave the evidence outstanding for the assessor.

Later assessments read this file with the other evidence.

The blank in the evidence kit is skills/gds-assessment-how-to/evidence-kit/panel/dialogue.md. create-evidence-kit.py copies it to panel/dialogue.md.

Keep the panel on track

• One phase. Four perspectives. Each assessor skill scores only the points named in its SKILL.md.
• Advice that only one discipline would give is incomplete. Say what the other three see, including when they have nothing to add.
• The report is still Met or Not Met. Any in-scope Not Met, or any fired red line, makes the overall result Not Met.

Output

While a question is open, the message is that question and nothing else. No verdict, no report, no assessor JSON.

After the last question is closed, confirm the log path, then the assessor skills may run.