AI Capex Trade-Off Calculator + Finance Vault

I want to start running the AI Capex Trade-Off prompts from this bundle against my own P&L and headcount. Walk me through it.

This is NOT a Terminal or coding install. There is no `git clone`, no `npm install`. The interactive calculator opens in a browser tab. The 12 Finance Vault prompts run inside Claude (claude.ai in a browser, Claude desktop app, or the Anthropic Console). Treat me like a CFO, ops lead, or founder who is comfortable with spreadsheets but has never coded.

What I am setting up: a Claude Project that holds my P&L, headcount roster, and SaaS ledger so the 12 prompts can run against my real numbers. Plus the live calculator (separate tab) for the quick math before the board.

Walk me through this, one step at a time. Wait for me to confirm before moving on:

1. Open a new browser tab. Click the "Launch the live tool" button on this page. The AI Capex Trade-Off Calculator opens. Drop in your headcount, fully-loaded comp, current SaaS + infra spend, customer-facing payroll %, and target redeploy %. The Benioff ratio + payback quarters appear instantly. Leave the tab open — we will reference numbers later.

2. Back to claude.ai (or open it now). Sign in. For board-grade work I want Pro or Team (Opus 4.7); free + Sonnet 4.6 covers a first-pass.

3. Left sidebar → "Projects" → "Create Project". Name it "AI Capex Reshuffle".

4. In Project Knowledge, paste / upload these:
   - My P&L for the last 4 quarters + the forward 4 quarters of plan (XLSX or CSV upload works)
   - My headcount roster with fully-loaded comp by role (anonymise names if needed — Claude only needs role + comp + tenure + manager)
   - My SaaS + infra vendor ledger (last 12 months)
   - The bundle's `reality-check.md` ceiling note + `audit-compliance-overlay.md` (RBAC, payroll continuity, EEOC for any cut decision)

5. In Custom Instructions, paste: "You are an FP&A lead + chief of staff hybrid. Output only what passes the `<review_gate>` and `<constraints>` blocks in each prompt. Never recommend cutting a role without naming the Claude workflow that replaces 60%+ of the weekly work. CFO + GC + Head of People all sign off before any cut lands."

6. Start with prompt 01 (Benioff Test on your headcount). Copy it from `finance-vault-prompts.md`, open a new chat inside the Project, paste, send. The prompt reads the headcount roster you already pasted into Project Knowledge — no need to re-paste.

7. Read Claude's output. Confirm:
   - Every role-replaceable score names the specific workflow that would replace the weekly tasks
   - The Benioff ratio matches roughly what the calculator showed in step 1
   - No protected-class signals leaked into the scoring

8. Once prompt 01 works, repeat for prompts 02 (AI Capex Trade-Off with P&L), 07 (Headcount-to-GPU), 12 (Internal Memo). Those four are the quarter-changers.

Rules:

- One step at a time. Tell me exactly where Projects sits in the Claude UI, and which menu hides Custom Instructions.
- Define jargon once: Project, Custom Instructions, Project Knowledge, `<review_gate>`, Benioff ratio, GPU-equivalent, customer-facing payroll.
- The live calculator runs in your browser, single HTML file, no install. It is the fast lane.
- Anti-pattern callouts:
  - If I say "I cannot find Projects in claude.ai", check profile menu for "Try Projects" toggle, otherwise paste P&L + headcount into each chat individually (uglier but works)
  - If I say "my P&L is too big to paste", upload it as a file in Project Knowledge — Claude reads XLSX, CSV, PDF
  - If I say "my CFO will not let employee comp data hit Claude", that is a fair gate. Use the calculator standalone first (no data leaves your browser), then take the output back to your CFO with the compliance overlay before pasting any P&L

First message: ask "Are you on claude.ai web, the desktop app, or Anthropic Console — and is your P&L data approved for AI use yet?" Then start with step 1.

When prompts 01, 02, 07, 12 run cleanly, switch into "board-pack mode" and help me write the one-page decision memo for the next board meeting using prompt 11 against the outputs we just generated.