The Live Reporting Pack

I want to turn my finance reporting into one living workspace the whole firm reads off a single share link, instead of emailing a static dashboard or a stale spreadsheet every quarter. Walk me through it step by step, do not skip anything. Treat me like a CFO, controller, IR lead, or PE operator who has never built a Claude Artifact and does not write code. Define every term once.

This is NOT a Terminal or coding install. There is no `git clone`, no `npm install`, no command line. It all happens inside the Claude app (claude.ai in a browser, or the Claude desktop app). The whole thing runs in my own Claude account and my numbers stay in my tenant.

## What a live reporting workspace actually is
Two real Claude features do all the work here, no third tool required:

- A **Claude Project** is a private workspace that holds my data and remembers it across chats. I drop my numbers in once (the actuals export, the budget, the portfolio packs), and every prompt in this vault reads from the same place.
- A **Claude Artifact** is the report itself, built as a live interactive dashboard, not a screenshot. When I hit **Publish**, Claude gives me a share link. Anyone with the link opens the dashboard in their browser, sees the numbers, filters them, and on an interactive artifact, types their own questions against the data I loaded. No login, no Tableau seat, no PDF attachment.

So "live" means this: the board, the LP, or the CFO opens one link and works the numbers themselves. When the numbers change, I paste the new figures into the same Project, ask Claude to refresh the artifact, hit Publish again, and the same link now shows the latest. That is the honest mechanism. It is not a database that polls itself every second; it is a workspace I refresh in thirty seconds and a link that always shows the current version. (If I later want true hands-off auto-refresh on a schedule, that is a connector or a managed agent on top, and the consultance.ai team wires that part.)

## The setup, once, about 10 minutes
Walk me through ONE step at a time, waiting for me to confirm each before moving on:

1. **Confirm the plan.** Real financials are firm data, so I want a **Team or Enterprise** Claude plan, which keeps my numbers in my own tenant. Tell me how to confirm that under my account. The free plan is fine for testing the first prompt on dummy numbers, but not for real board or LP data.
2. **Create the workspace.** In claude.ai, in the left sidebar click **Projects**, then **Create Project**. Name it for the reporting cadence, for example "Q3 Board Pack" or "Live Reporting Desk". Success looks like an empty Project with a Knowledge panel on the right.
3. **Load the numbers.** In the Project's **Knowledge** panel, click **Add content** and upload what the prompts will read: the GL or actuals export, the budget or plan, the portfolio packs, the prior board deck or LP letter. Excel, CSV, and PDF all work. If my data lives in Google Drive or Notion, I can connect it through the connector in Claude settings instead of uploading, zero install.
4. **Pin the house rules.** In the Project's **Custom Instructions** box, paste this: "You build live finance dashboards as interactive Artifacts supervised by a named human owner. Every figure must tie to a source I loaded into this Project. Never invent a number. Label every assumption. Show the downside beside the base case. Before you publish any artifact, list any figure you could not tie to a source. Flag every LP or performance figure for compliance review."
5. **Preflight the data, do not skip this one.** Before I trust any dashboard, prove Claude can actually read the numbers. In a normal chat inside the Project, ask it to read three specific figures back to me from the files I loaded (last quarter revenue, total cash, one portfolio company's EBITDA). If it reads all three correctly, the dashboards will be right. If it cannot, fix the file or the access now. This is the single mistake that wastes a first run: a clean-looking dashboard that is quietly wrong because Claude never had the data.

Then for each report below: open a new chat **inside the Project**, paste the prompt, replace each `{{TOKEN}}` with my real input, run it, read the output, then hit **Publish** and copy the share link.

## How to read each prompt
Each of the ten gives me the same five things: the **report** it builds, the **data** it reads, the **rubric** Claude grades its own output against before it publishes, a **ready-made prompt** I paste as one block, and the **proof check** that tells me the live link is right before I send it to anyone. Replace the `{{TOKENS}}`, run it in the Project, publish, share.

## The vault, ten live reporting prompts

### 1. Board Pack Builder (build the board pack from raw numbers like a Goldman analyst)
- **Report:** a one-page board dashboard, revenue, margin, cash, the three KPIs that matter, each versus plan and versus last period.
- **Data:** the actuals export and the budget in Project Knowledge.
- **Rubric:** every figure ties to a loaded source, base case and variance shown side by side, no number left unsourced.
- **Prompt to paste:** "Build me a live board pack as an interactive Artifact from the actuals and budget in this Project. Show {{TOP_KPIS, e.g. revenue, gross margin, cash, EBITDA}} for {{PERIOD}}, each against plan and against the prior period, with a one-line driver on anything off by more than {{THRESHOLD, e.g. 5%}}. Make it a clean dashboard a board member reads in ten seconds, with a filter to switch period. Before you finish, list any figure you could not tie to a source. Do not include a number you cannot source."
- **Proof check:** pick one tile, open the source file, confirm it ties to the cent. Then hit Publish and open the link in an incognito window to confirm it loads with no login.

### 2. Quarterly Deck to Live Link (turn the quarterly deck into a live link like a Blackstone IR lead)
- **Report:** the quarterly story as an interactive page, not a 40-slide PDF, with the headline numbers, the trend charts, and the commentary in one scrollable artifact.
- **Data:** the prior quarterly deck plus this quarter's actuals.
- **Rubric:** every chart is built from loaded data, the narrative matches the numbers, nothing is carried over stale from last quarter.
- **Prompt to paste:** "Rebuild my last quarterly deck ({{DECK_FILE}}) as a single live interactive Artifact using this quarter's actuals. Keep the same sections, refresh every number and chart, and write the commentary in plain language that matches what the numbers actually show. Flag anything that contradicts last quarter's deck. Add a period selector at the top."
- **Proof check:** read the commentary against the charts. If a sentence says "up" and the chart shows down, the narrative did not refresh. Fix before publishing.

### 3. Portfolio Rollup that Refreshes (refresh the portfolio view on every data change like a Carlyle ops director)
- **Report:** one portfolio dashboard across all holdings, each company's revenue, EBITDA, and status, with a total roll-up at the top.
- **Data:** the per-company monthly packs in Project Knowledge.
- **Rubric:** every company present or flagged as missing, the roll-up sums correctly, each figure cites the company it came from.
- **Prompt to paste:** "Build a live portfolio dashboard as an Artifact from the monthly packs in this Project. One row per company with {{METRICS, e.g. revenue, EBITDA, vs plan}}, a portfolio total at the top, and a flag on any company whose pack is missing or contradicts last month. Make it sortable by any column. When I paste new packs later, you will refresh this same dashboard."
- **Proof check:** add the company rows by hand once and confirm they equal the roll-up total. Then next month, paste the new packs, ask for a refresh, republish, confirm the same link updated.

### 4. Self-Serve LP Model (hand an LP a model they query themselves like a KKR reporting partner)
- **Report:** an interactive artifact the LP opens and drives, with the fund or deal numbers and inputs they can move (commitment, entry multiple, hold) to see their own outcome, without emailing me.
- **Data:** the fund model or deal model, plus the inputs that are safe to expose.
- **Rubric:** the LP can only move the inputs I allow, every output recomputes from the loaded model, no private deal detail leaks into a public link.
- **Prompt to paste:** "Build an interactive Artifact an LP can use themselves from {{MODEL_FILE}}. Expose only these inputs as sliders or fields: {{SAFE_INPUTS}}. Everything else stays fixed. When they change an input, recompute {{OUTPUTS, e.g. net IRR, MOIC, distributions}} live. Add a one-line note that this is illustrative and a named human at the firm confirms any figure. Do not expose {{ANYTHING_CONFIDENTIAL}}."
- **Proof check:** move every exposed slider to its extreme and confirm nothing breaks or reveals a hidden number. Treat the published link as if a stranger has it, because once shared, they do.

### 5. Twelve-Entity Consolidation (roll up 12 entities into one number like a fund controller)
- **Report:** a consolidation dashboard that takes many entities and shows one consolidated number, with drill-down into each entity and the eliminations.
- **Data:** each entity's trial balance or P&L.
- **Rubric:** intercompany eliminations shown explicitly, the consolidated total equals the sum of entities minus eliminations, every entity accounted for.
- **Prompt to paste:** "Consolidate the {{N}} entities in this Project into one set of numbers as a live Artifact. Show the consolidated {{P&L / balance sheet}} at the top, a drill-down to each entity, and the intercompany eliminations as their own line. Prove the math: consolidated equals sum of entities minus eliminations. Flag any entity whose file is missing or whose currency I have not told you how to convert."
- **Proof check:** confirm consolidated = sum of entities minus eliminations on screen. If it does not foot, it is wrong, do not publish.

### 6. Living IC Scenario (walk the IC through a living scenario like a Bridgewater strategist)
- **Report:** an interactive scenario artifact for the investment committee, base, upside, downside, with the levers live so the committee can test "what if" in the room.
- **Data:** the model and the assumptions that drive the scenarios.
- **Rubric:** each scenario built from the same model, the levers are real model inputs, the downside is a genuine downside not a second base case.
- **Prompt to paste:** "Build a live scenario Artifact for an IC from {{MODEL_FILE}}. Show base, upside, and downside on {{KEY_OUTPUTS}}, with these levers adjustable in the room: {{LEVERS, e.g. revenue growth, exit multiple, rate path}}. Make the downside a real one: revenue down, margin down, multiple compression, rates up, at the same time. Show all three side by side and recompute as a lever moves."
- **Proof check:** drag the downside levers to a real bad case and confirm the returns actually fall. If the "downside" still clears the hurdle comfortably, it is not a downside, tighten it before the meeting.

### 7. Variance Flagger (flag the variance the second it moves like a Deloitte audit senior)
- **Report:** a variance dashboard versus budget that highlights every material line in red the moment the actuals change.
- **Data:** the GL actuals and the budget.
- **Rubric:** total variance reconciles to actuals minus budget, every line over threshold has a plain-English reason, nothing left as "higher than budget".
- **Prompt to paste:** "Build a live variance Artifact versus budget from the actuals and budget in this Project. Show every line off by more than {{THRESHOLD}} in dollars and percent, sorted biggest first, each with a real one-line reason. Put a three-line summary at the top a board member reads in ten seconds. Prove total variance equals actuals minus budget. When I paste new actuals, refresh this same dashboard and re-flag."
- **Proof check:** confirm the total variance on screen equals actuals minus budget before reading a word of the commentary.

### 8. Plain-English Q&A Layer (answer the CFO's question in plain English like an FP&A director)
- **Report:** a published workspace where the CFO types a question in plain English ("why is cash down this month?") and gets an answer off the live numbers, with the figures that support it.
- **Data:** everything already loaded in the Project.
- **Rubric:** every answer cites the figures behind it, says "I cannot find that" rather than guessing, never invents a number.
- **Prompt to paste:** "Set up this Project so anyone I share it with can ask finance questions in plain English and get an answer from the numbers I loaded. For every answer, show the figures behind it and the source. If a question cannot be answered from the loaded data, say so plainly instead of guessing. Start by listing the five questions a CFO is most likely to ask this period, with the answer ready."
- **Proof check:** ask it a question you already know the answer to. If it matches and cites the source, the layer is trustworthy. Ask it something not in the data and confirm it says so rather than inventing.

### 9. Forecast Stress Test (stress test the forecast on the fly like a Citadel risk lead)
- **Report:** an interactive forecast artifact where I move the assumptions and watch the forecast and the worst week move with them.
- **Data:** the forecast model and its assumptions.
- **Rubric:** week one ties to real opening cash, every assumption is a live lever, the lowest projected week is always called out.
- **Prompt to paste:** "Build a live stress-test Artifact from {{FORECAST_FILE}}. Start from real opening cash. Make these assumptions adjustable: {{ASSUMPTIONS, e.g. collections timing, churn, FX, rate path}}. As I move them, recompute the forecast and always show the lowest projected {{week/month}} and its date in red. Add a one-click 'severe but plausible' button that moves several assumptions against me at once."
- **Proof check:** confirm period one equals today's real balance before trusting any later period. Then hit the severe button and sanity-check the worst week is genuinely worse.

### 10. Close-the-Books Workspace (close the books on a workspace not an inbox like a controller at a $500M family office)
- **Report:** a live close tracker as an artifact, every close task, its owner, its status, and the open items, instead of a chain of emails.
- **Data:** the close checklist and the current actuals.
- **Rubric:** every task has an owner and a status, blocked items surfaced at the top, the close is not marked done while anything is open.
- **Prompt to paste:** "Build a live month-end close workspace as an Artifact from {{CLOSE_CHECKLIST}}. One row per task with owner, status, and due, blocked or overdue items pinned at the top, and a progress bar for the whole close. Pull the current actuals to auto-tick any task whose number is already in. Do not show the close as complete while any task is open or any figure is untied."
- **Proof check:** confirm a task you know is not done shows as open, and the close does not read 100% while it is. Share the link with the close team instead of starting an email thread.

## The trust gate, the one that makes the rest safe
Before any of these ten links goes to a board member, an LP, or the CFO, run this in the Project: "Re-check the artifact you just built against the sources I loaded. Re-derive the three biggest numbers. Confirm every figure ties and every assumption is labeled. List anything that does not tie, and do not tell me it is ready to publish until the list is empty." A published link is public to anyone who has it. The gate is what stands between "live and right" and "live and wrong in front of an LP".

## Build the whole workspace in one weekend
- **Saturday morning:** confirm the plan, create the Project, load your three core sources, run the data preflight (step 5). Do not skip the preflight.
- **Saturday afternoon:** build prompts 1, 7, and 8 (board pack, variance flagger, plain-English layer). Publish each, open the links in an incognito window, tighten the rubric, republish.
- **Sunday:** build the report your audience actually opens first, the portfolio rollup (3), the LP model (4), or the IC scenario (6). Run the trust gate on every link before you share it.
- **The following week:** add the rest as the data sources come online, and the first time the numbers change, practice the refresh: paste new figures, ask for a refresh, republish, confirm the same link updated.

Build one, publish it, open the link as an outsider, tighten the test, then send it. The rubric is the report. Everything else is just the data and the share button.

## Rules for walking me through this
- One step at a time. Tell me exactly what to click and where it is in the Claude app: the **Projects** section in the left sidebar, **Create Project**, the **Knowledge** panel and its **Add content** button, the **Custom Instructions** field, and the **Publish** then **Share link** button on an artifact.
- Define jargon once: Project, Project Knowledge, Custom Instructions, Artifact, interactive artifact, Publish, share link, connector, `{{TOKEN}}`, rubric, base vs downside.
- If the UI looks different (web vs desktop, Team vs Enterprise), give me the variant. Do not tell me a step is "not possible". Tell me to look for a "Projects" or "Workspaces" section, an "Add content" or "Add files" upload, a "Custom Instructions" field, or the "Publish" and "Share" control on the artifact panel. If a control is genuinely missing, fall back to building the report in a chat and exporting it.
- Never tell me to install anything via Terminal. The whole thing runs from a browser tab or the desktop app.
- Be honest about "live": it is a workspace I refresh and a link that always shows the latest version, not a self-polling database. If I want hands-off auto-refresh on a schedule, tell me that is a connector or a managed agent on top, and point me to consultance.ai to have it wired.
- Anti-pattern call-outs: if I say "my IT blocked claude.ai", tell me about the Team or Enterprise plan with SSO, or the desktop app. If I say "I do not have a paid plan", tell me the free plan tests the first prompt but real board and LP work needs the firm tenant so the data stays in my own account. If I say "I do not want the LP to see everything", point me back to prompt 4's rule about exposing only safe inputs and treating the link as public.

## Permission to use
Prompt set authored by consultance.ai. Built on Claude Projects and Claude Artifacts; no affiliation with Tableau, Power BI, or any named firm implied. Everything runs in your own Claude account and your numbers never leave your tenant. These build the report; a named human owner reviews and ties out every figure before any link reaches a board, an LP, or the CFO. LP and performance figures carry Marketing Rule (Rule 206(4)-1) and GIPS obligations; your compliance and counsel sign off. Not financial, legal, or investment advice. If it replaced a quarter of static-dashboard work, a repost is the only thanks I will ask for.