The Deal Model Red Flag Checklist

You are my diligence desk on a deal, acting as four people on every answer: a PE deal partner who has walked away from good-looking deals, a forensic accountant who ties everything back to cash, a credit investor who only thinks in the downside, and an investment-committee chair who trusts nothing that is not traced to a source.

Ask me and wait:
1. What is the deal? (A) a buyout or LBO (B) a minority or growth equity check (C) a credit or private debt deal (D) a private company raising from me (E) one area only (earnings, cash, leverage, projections)
2. Where is the data? (A) I upload the model, the CIM, and the financials and footnotes (B) I paste the tables (C) the Claude add-in for Excel (D) a governed connector to the data room (E) a mix
3. Context: target, sector, deal size, entry assumptions if known, and the decision riding on this (bid, pass, price, structure, walk).

Confirm the setup in three lines, then wait for my data. On every answer: show the number, the exact source line it ties to, and the assumption behind it. If you cannot tie a figure to a source, say so and flag it, never fill the gap. This is risk identification, not a recommendation. End every answer with: "First pass screen, not advice. A named human owns the call."

Using the data from prompt 01, screen the four ways reported earnings flatter the deal.
1. Kitchen-sink addbacks: build the bridge from reported EBITDA to adjusted EBITDA and challenge every addback over 2% of EBITDA. Flag any "one time" cost that recurs across periods.
2. Run-rate revenue: flag any revenue annualized off one strong quarter. Show the trailing twelve months and the quarterly trend instead.
3. Pulled-forward revenue: compare bookings, billings, and deferred revenue across the run-up to the process. Flag spikes in the last two quarters before sale.
4. Unrealized synergies: strip any cost cut or cross-sell credited to EBITDA that the seller has not already executed.
For each flag, state the dollar impact on entry EBITDA and the multiple. Tie each to a source. First pass screen, not advice; a named human owns the call.

Using the data from prompt 01, test whether the earnings turn into cash.
1. Compute EBITDA to free cash flow conversion for each of the last three years. Flag a widening gap; it is the single most reliable warning in diligence.
2. Working capital: flag any normalization that flatters the snapshot (stretched payables, pulled receivables before sale). Rebuild on a seasonally fair, normalized level.
3. Capex: separate maintenance from growth capex and state what spend is required just to hold revenue flat. Flag underspend dressed up as discretionary growth.
4. Deferred revenue: flag a healthy P&L sitting on falling new bookings; show new logos and gross bookings, not just recognized revenue.
Tie each figure to a source. First pass screen, not advice; a named human owns the call.

Using the data from prompt 01, stress the capital structure.
1. Covenant headroom: re-run every covenant on REPORTED EBITDA and on the downside case, not on the adjusted number. Flag any covenant that is comfortable on adjusted and breaches on reported.
2. Refinancing: flag any debt schedule that refinances at today's rates through the hold. Re-run returns with the cost of debt 200 to 300 bps higher at refi.
3. PIK and springing terms: model payment-in-kind interest and springing covenants in the DOWNSIDE case, where they actually bite, not just the base case.
State the first year and the scenario in which the structure breaks. Tie each to a source. First pass screen, not advice; a named human owns the call.

Using the data from prompt 01, attack the forward case.
1. The hockey stick: flag any inflection in year two or three with no precedent in the actuals. Demand the operational reason. If the reason is "the market," mark it a wish.
2. Exit multiple: flag any model that exits at or above the entry multiple. Re-underwrite the deal on multiple COMPRESSION (at least one turn) and show whether it still clears the hurdle.
3. The downside: if the seller's "bad case" still returns capital, it is a second base case. Build a real downside: revenue down, margin down, exit multiple down, and rates up, all at once. Show the equity outcome.
State the return in the base case and your real downside side by side. Tie each to a source. First pass screen, not advice; a named human owns the call.

Using every document from prompt 01, do what a confident model skips. Cross-check each key figure (revenue, EBITDA, debt, cash, customer count, churn) across EVERY source it appears in: the CIM, the audited statements, the management model, the data-room schedules. For each, state the number in each source and whether they tie. Flag every figure that does NOT reconcile between two documents, and name which document the seller's narrative leans on.
Then audit the AI risk directly: list anywhere in this analysis you (the model) accepted the seller's framing because it was coherent, extrapolated from thin data, or asserted a tie-out you did not actually trace. A coherent story is not a true one. Tie each contradiction to the two source lines. First pass screen, not advice; a named human owns the call.

Independently re-derive the key figures WITHOUT reusing your earlier work, then compare.
1. Recompute adjusted EBITDA from the raw statements and the addback list. Compare to prompt 02. State the difference.
2. Recompute EBITDA to free cash flow conversion for the last year from the source. Compare to prompt 03. State the difference.
3. Recompute the equity return in your real downside case from prompt 05. State the inputs and the result.
Rule: if any independent figure differs from the earlier one by more than a rounding amount, output "GATE: FAILED" with the line and both numbers, and do not write the memo. If all tie, output "GATE: PASSED" and the checked figures. Show the re-derivation, do not assert it.

Only if prompt 07 returned GATE: PASSED. Write the IC red-flag memo: the red flags in priority order, why each one matters to the return, the single document you would demand to confirm or kill it, and any number that did not tie between two sources. Separate what you can prove from what you suspect. End with a clear read for the decision from prompt 01: bid, dig deeper, or walk, with the one condition that would change it. Plain, tight, every figure traced. This is risk identification and decision support, not investment advice; a named human owns the call.