Pre-work · ~60 minutes, remote, before the training day

Defensible architecture claims webinar

A short calibration session: confirm Claude Code can read your workspace, then practise deciding what an individual AI-generated architecture claim is safe to become — before it enters a decision process.

This is a live, facilitator-led session. This page is not a self-paced assignment — follow the facilitator's lead and use the blocks below to copy each prompt exactly when it's introduced.

Before you join

Download and extract the webinar starter package below.
Keep the four source documents together in the extracted folder.
Open that folder as your working folder in Claude Code.
Confirm Claude Code is installed and authenticated — terminal, desktop app, VS Code, or any interface you're comfortable with.

01-product-request.md Product idea, desired journey, initial scope — the only file Claude reads during the webinar

02-current-state-brief.md Present, unopened during the webinar — verifies you have the correct folder

03-product-meeting-notes.md Present, unopened during the webinar

05-system-landscape.md Present, unopened during the webinar

Do not rename the files or add other case documents to this folder, and do not read or analyse any of them before the session. During the webinar, Claude reads only 01-product-request.md — the other three sources stay unopened for the full training-day discovery. The case is entirely fictional and contains no customer or confidential data.

Download webinar starter package (.zip)

Optional setup check

Before joining, you can verify your workspace with this prompt. Claude should list the four files above and nothing more.

List the Markdown files available in the current working folder. Do not read
or analyse their contents.

During the webinar

Two short exercises, in the same Claude conversation. The facilitator pastes each prompt into the Teams chat when it's introduced — copy it from here just as easily.

Step 1 · workspace check

Confirms Claude can see the four files. Seeing file names is not the same as reading file contents.

List the Markdown files in the current working folder.
Do not read or analyse their contents.

Step 2 · claim-audit preview

Read the product request and produce a bounded, four-row audit of its claims. Different answers between participants are expected — that's the point of the group comparison that follows.

Read only `01-product-request.md`. Do not read any other files.

Start with:
SOURCE READ: <file name> — <document title>

Then produce a claim-audit preview with exactly four rows.

Use these columns:
- Claim: paraphrase one material claim, maximum 18 words
- Type: stated fact, requested outcome, assumption, or proposed solution
- Support: document section plus a phrase of maximum 10 words
- Weakness: what remains ambiguous or unsupported, maximum 15 words

After the table, list exactly two questions that should be answered before
architecture design begins. Maximum 12 words per question.

Do not propose an architecture.
Do not add an introduction or conclusion.
The complete response must be under 180 words.

Context update

The facilitator will read this once, then pause. Decide for yourself whether it changes any of your four claims — before asking Claude.

Step 3 · revise the analysis

Stay in the same conversation. Check whether Claude's response distinguishes claims that are unchanged, revised, and withdrawn — rather than silently smoothing over what changed.

Use the new context below to review your previous four-row claim audit.

Report only:
- the number of claims that remain unchanged;
- the number that should be revised;
- the number that should be withdrawn;
- one revised claim, maximum 30 words;
- one sentence explaining why it changed.

Do not reproduce the full table.
Maximum 90 words.

New context:
A workspace identifies where a user is active. It does not establish which
legal customer pays, which billing account applies, or whether the user has
purchasing authority.

What this sets up

Today you confirm Claude can read a real source and audit individual claims. The training day scales that same discipline to a full evidence corpus, ownership, blockers, and architecture artifacts — see the full-day schedule.