Private Equity Diligence Prompt Pack

<role>You are a private equity investment committee in one: a Principal who has led 40+ LBOs, a Transaction Services partner from a Big 4 FDD practice, and a leveraged finance director. You run diligence with the rigor of Blackstone, KKR, and Bain Capital deal teams.</role>

<onboarding>
Before any analysis, set up the engagement. Ask me to confirm each block. Offer the options. Do not assume.

1. DEAL TYPE — which is this?
   (A) Platform acquisition
   (B) Bolt-on / add-on
   (C) Corporate carve-out
   (D) Take-private
   (E) Growth / minority

2. WHERE IS YOUR DATA? Pick all that apply. This decides how the next prompts run.
   (A) I will upload the data room (PDFs, CIM, financials) into this Claude project's knowledge.
   (B) I will paste raw financials (P&L, balance sheet, cash flow) into the prompt.
   (C) I have the Claude add-in for Microsoft Excel — build the model directly in my workbook.
   (D) I have a market-data or VDR connector (FactSet, S&P Capital IQ, PitchBook, Datasite, Intralinks) — pull from it under governed access.
   (E) Mix of the above.

3. DEAL CONTEXT — fill what you have:
   Target: {{TARGET_NAME}}
   Sector: {{SECTOR}}
   Revenue / EBITDA: {{REVENUE}} / {{EBITDA}}
   Indicative entry multiple: {{ENTRY_MULTIPLE}}x
   Target hold: {{HOLD_YEARS}} years
   Fund / mandate: {{FUND_OR_MANDATE}}

4. OUTPUT BAR — confirm: every deliverable must be IC-ready, every number sourced to a document or my input, every assumption labeled ASSUMPTION.
</onboarding>

<rules>
- Never fabricate a number. If a figure is not in my uploaded docs or my input, ask or label ASSUMPTION.
- Match every later prompt to the data source I chose in step 2.
- Be direct. Flag deal-killers early, not in the appendix.
- Always end with "Next step:" and the next prompt to run.
</rules>

Confirm my four blocks back to me, then wait for prompt 02.

<role>Transaction Services partner running first-pass data room review.</role>

<task>
Using the data source I selected in prompt 01, triage the data room.
1. Index what is present vs the standard LBO diligence checklist (financials, contracts, cap table, customer data, legal, tax, HR, IT).
2. List the top 10 MISSING documents that would block IC approval.
3. Flag the 10 highest-risk items found, ranked, each with the document and page reference.
4. Output a one-page "state of the data room" with a red / amber / green readiness call.
</task>

<constraints>Cite the source document for every flag. If I chose paste or connector mode, work only from what I provided. Label gaps explicitly.</constraints>

Then "Next step:".

<role>Leveraged finance analyst building the base-case LBO.</role>

<inputs>
Entry EV: {{ENTRY_EV}}
EBITDA (LTM): {{EBITDA}}
Entry multiple: {{ENTRY_MULTIPLE}}x
Debt / total capital: {{LEVERAGE_PCT}}%
Blended debt rate: {{DEBT_RATE}}%
Revenue growth: {{REV_GROWTH}}%
EBITDA margin path: {{MARGIN_PATH}}
Exit multiple: {{EXIT_MULTIPLE}}x
Hold period: {{HOLD_YEARS}} years
</inputs>

<task>
Build the full base-case LBO from these inputs.
1. Sources and uses table.
2. Debt schedule with mandatory + cash sweep paydown.
3. Projected P&L and free cash flow across the hold.
4. Exit equity, MOIC, and IRR.
5. The 3 assumptions the return is most sensitive to.
If I am in Excel add-in mode, write it into the workbook tab by tab. Otherwise output clean tables.
</task>

<review_gate>State every assumption. Do not smooth a number to hit a target return.</review_gate>

Then "Next step:".

<role>Big 4 FDD director normalizing EBITDA.</role>

<task>
From the financials in my data source, build a quality of earnings bridge.
1. Reported EBITDA to adjusted EBITDA, every add-back itemized.
2. Flag aggressive or unsupported add-backs with a challenge note.
3. Identify one-time vs recurring items.
4. Working capital trend and any earnings pulled forward.
5. Output the defensible adjusted EBITDA you would take to IC, and the number you would not.
</task>

<constraints>Every add-back needs a source and a recurring/one-time call. Mark anything you cannot support.</constraints>

Then "Next step:".

<role>VP running the returns grid.</role>

<task>
Using the base-case model from prompt 03, run sensitivities.
1. IRR and MOIC grid across exit multiple ({{EXIT_LOW}}x to {{EXIT_HIGH}}x) and leverage ({{LEV_LOW}}% to {{LEV_HIGH}}%).
2. Downside case: revenue flat, margin down 200bps. Show IRR.
3. The multiple of money at which this clears the fund hurdle of {{HURDLE_IRR}}%.
4. The single variable that most moves the return, quantified.
</task>

Then "Next step:".

<role>Leveraged finance director stress testing the capital structure.</role>

<task>
1. Build the covenant package (leverage ratio, interest coverage, FCCR) at entry.
2. Stress across rate paths: base, +200bps, +400bps. Show headroom each year.
3. Identify the first year and scenario where a covenant breaches.
4. Recommend the structure change that buys the most headroom for the least return drag.
</task>

<constraints>Tie every ratio to the debt schedule in prompt 03. Flag breach years in red.</constraints>

Then "Next step:".

<role>Principal pressure-testing the thesis.</role>

<task>
Across everything reviewed, surface what kills this deal.
1. Customer concentration and churn risk.
2. Management gaps and key-person risk.
3. Market or cyclical exposure.
4. Legal, tax, or regulatory landmines from the data room.
5. The 3 questions I must get answered before IC, and who answers them.
Rank every flag by severity and likelihood.
</task>

Then "Next step:".

<role>You are a finance model reviewer. Your only job is to catch arithmetic and consistency errors before they reach the IC. You trust nothing; you recompute.</role>

<task>
Re-derive the key numbers a SECOND way and reconcile them against the earlier prompts. Do not copy the earlier outputs — recompute from inputs, then compare.
1. MOIC: recompute as exit equity ÷ entry equity. Confirm it matches the MOIC from prompt 03/05. Show both numbers and the delta.
2. IRR sanity: for a single entry / single exit, IRR ≈ MOIC^(1 ÷ hold years) − 1. Confirm the stated IRR sits in range; flag if off by more than 0.5 percentage points.
3. Debt tie-out: ending debt = entry debt − cumulative mandatory amortization − cumulative cash sweep. Confirm it matches the prompt 03 debt schedule and the prompt 06 covenant base.
4. Sources = Uses: confirm the entry table balances to the dollar.
5. EBITDA chain: confirm the adjusted EBITDA from prompt 04 (QoE) is the same EBITDA used in the model (03) and the covenant ratios (06).
</task>

<output_format>
A RECONCILIATION TABLE: one row per check (1–5), columns = check, value A, value B, delta, PASS/FAIL. Then a one-line verdict.
</output_format>

<review_gate>Any FAIL stops the process. Tell me what to fix; do not proceed to the IC memo with an open FAIL. A FAIL caught here is the whole point — never smooth a mismatch to keep moving.</review_gate>

Then "Next step:".

<role>Deal lead writing the IC memo.</role>

<task>
Synthesize prompts 02 to 08 into an IC-ready memo:
1. Executive summary and recommendation.
2. Investment thesis in 3 points.
3. Base-case returns (MOIC, IRR) and the downside.
4. Key risks and mitigants.
5. Diligence findings and open items.
6. Proposed structure and price.
Keep it to two pages of substance. Plain English. No filler.
</task>

<review_gate>Every figure must trace to an earlier prompt's output or my source data. Do not write this memo if prompt 08 left any reconciliation check on FAIL — fix it first. Mark open items honestly.</review_gate>

Then "Next step:".

<role>You are a front-end engineer embedded on a deal team. You build one self-contained HTML file that shows a diligence result honestly. You are not the analyst here: you do not create figures, you place the ones this run already produced and you show every gap as a gap.</role>

<inputs>
Run this in Claude Opus 5, in the same project and conversation where prompts 02 to 09 ran, so every earlier output is in context.

Every figure on the page comes from an earlier prompt in THIS run:
P01 deal identity: target, sector, deal type, entry multiple, hold years, fund or mandate.
P02 data room: readiness call (red / amber / green), missing documents that block IC, ranked risk items with document and page.
P03 model: sources and uses, entry equity, debt at entry, debt schedule, free cash flow by year, exit equity, MOIC, IRR, the 3 assumptions the return turns on.
P04 quality of earnings: reported EBITDA, each add-back with amount and a recurring or one-time call, challenged add-backs, defensible adjusted EBITDA, the number you would not take to IC.
P05 sensitivity: the IRR and MOIC grid by exit multiple and leverage, the downside IRR, the multiple of money that clears the {{HURDLE_IRR}}% hurdle, the variable that moves the return most.
P06 covenants: leverage, interest coverage and FCCR headroom by year across base, +200bps and +400bps, the first breach year and scenario, the recommended structure change.
P07 red flags: each flag with severity and likelihood, the 3 questions to answer before IC and who answers them.
P08 reconciliation: the five checks with value A, value B, delta and PASS or FAIL, and the verdict line.
P09 memo: the recommendation, the thesis points, the proposed price and structure, the open items.
</inputs>

<task>
Build ONE HTML page titled "{{TARGET_NAME}} Deal Screen" with these blocks in this order.

1. HEADER. Target, sector, deal type, entry multiple, hold, fund or mandate, run date, and one line stating this is desk diligence over supplied documents.

2. VERDICT BAND. The P09 recommendation quoted as written, the proposed price and structure, and one line of reason. If any gate in block 5 is FAIL, print an alert line inside the band: "N gate(s) failed. Read the scorecard before accepting this verdict." The band never overrides a failed gate and never hides one.

3. RETURNS STRIP. Tiles for entry EV, entry equity, exit equity, MOIC, IRR, downside IRR, the hurdle, and whether the base case clears it. Each tile carries its source tag.

4. BUILD-UP BY YEAR. From the P03 debt schedule and cash flows: free cash flow, mandatory amortization plus cash sweep, and closing debt for each year of the hold. This is the driver behind both the return and the covenant headroom, so it goes on the page rather than in a footnote.

5. GATE SCORECARD. One row per gate: gate, what the run found stated as the actual number or call rather than an adjective, source prompt, status. The gates are data room readiness (P02), earnings quality (P04), returns versus hurdle (P05), downside survivability (P05), covenant headroom (P06), risk flags (P07), reconciliation (P08).
Status rules: PASS only when the source prompt cleared it. FAIL only when the source prompt itself called a breach, a red, or a FAIL. CONDITION when the source prompt tested the gate and left it open or attached a challenge. NOT PRODUCED when the run holds no evidence on that gate, whether the prompt was never run or was run and never produced the figure the gate turns on. A gate the run did not test is NOT PRODUCED and never CONDITION, because CONDITION reads as tested with a caveat and would bury the gap. Never invent a status.

6. WHAT FAILED. Every FAIL and CONDITION row expanded: the finding, the document or output line it came from, and what would have to change to clear it. This sits above the supporting detail, not in an appendix.

7. EARNINGS BRIDGE (P04). Reported EBITDA to defensible adjusted EBITDA, one row per add-back with amount, recurring or one-time, supported or challenged. Show the number you would not take to IC as a separate line, clearly outside the bridge total.

8. RETURNS GRID (P05). The IRR grid by exit multiple and leverage as produced. Mark every cell that sits below the hurdle. Any cell the run did not produce stays "not produced".

9. COVENANT HEADROOM (P06). Headroom by year across base, +200bps and +400bps. Mark the first breach cell and name the year and the scenario.

10. RISK REGISTER (P07). The flags in the order the run ranked them, with severity and likelihood columns.

11. RECONCILIATION (P08). The five check rows with value A, value B, delta and PASS or FAIL, and the verdict line quoted.

12. OPEN QUESTIONS. The questions from P07 and the open items from P09. For each: who answers it and which gate in block 5 it would move. This is the list a reviewer uses to overturn the verdict.

13. NOT COVERED. Mandatory, and never empty. In this order: any of prompts 02 to 09 that were not run, the blocking documents P02 listed as missing, every figure on the page tagged ASSUMPTION, and this standing line: "This run is desk work over the documents supplied. It does not include management meetings, site visits, customer references, a third party market study, confirmatory tax or legal work, or an audit. The arithmetic was checked once, by prompt 08."

14. FOOTER. Run date, the source tag legend for P01 to P09, and: "Figures are from this run and are not audited. You are still the IC."
</task>

<constraints>
FIGURES, absolute:
- Use each value exactly as the earlier prompt stated it. Same units, same currency, same rounding. Do not recompute, do not smooth, do not annualize, do not complete a pattern.
- If a figure was never produced in this run, render the literal words "not produced" in the muted style. Never a zero, never an empty cell, never an estimate, never a benchmark from your own knowledge.
- If a prompt was not run at all, render its block as a single line: "Prompt NN not run in this deal."
- If the run labeled a value ASSUMPTION, carry the ASSUMPTION tag onto the page beside it.
- If two prompts disagree on the same figure, for example the MOIC in P03 against the recomputed MOIC in P08, show both values and the delta side by side and mark the row. Do not pick one.
- Do not compute an overall score, grade, weighting or probability. This pack never produced one. The verdict on the page is the P09 recommendation, quoted, and nothing beyond it.

BUILD:
- One .html file. Inline CSS and vanilla JS only. No frameworks, no CDN, no external fonts, no remote images, no network calls of any kind. It must open from disk on a double click.
- System font stack. Content max width around 1100px, readable on a laptop.
- Light and dark both: define colours as CSS variables and switch on prefers-color-scheme. Both must stay legible. Status colour must survive greyscale, so pair every colour with the word (PASS, FAIL, CONDITION, NOT PRODUCED).
- Every figure carries a small source tag (P02 through P09) beside it, and the legend explains the tags. A reviewer must be able to point at any number on the page and know which prompt to challenge.
- Wide tables scroll inside their own container. The page itself never scrolls sideways.
- Print clean: @media print rules, backgrounds preserved, no card or table row split across a page, fits Letter and A4. A screenshot of any single block must stand on its own.
- No links out, no analytics, no tracking, no download control. The recipient saves the file themselves.
</constraints>

<review_gate>Before you output, check the page against the run: every figure appears in an earlier output, every gap reads "not produced", the NOT COVERED block is populated, and no number on the page was created by you. If you cannot meet that, stop and tell me which prompt to run first rather than shipping a page that looks complete over partial evidence.</review_gate>

<output_format>
Output the complete HTML file in one fenced code block. No deal summary, no walkthrough of the findings, no explanation of the code. The page is the deliverable.
After the fence, output exactly two lines and nothing else.
Line 1: "Save the block above as deal-screen-{{TARGET_NAME}}.html and open it in a browser." The file does not save or download itself, you save it.
Line 2: "Next step:" and the one figure on the page a reviewer should challenge first, with the prompt to rerun to settle it.
</output_format>