model-review

name	model-review
description	7-pass review of financial models — assumptions, mechanics, integrity (BS balances, CF reconciles), hardcodes, error-checks, scenarios, documentation. Used by `financial-modeler` (primary) and `valuation-analyst` for DCF integrity checks.

Skill: model-review

Purpose: 7-pass review pattern for financial models. Used by financial-modeler (primary) and valuation-analyst for DCF integrity checks.

When to use

• Reviewing a model built by someone else before relying on it
• Pre-publication review of a model going to a board, lender, or external party
• Triaging a model that "doesn't tie" or produces unexpected results
• Periodic refresh check on a long-lived model

The 7 passes (in order)

Don't skip ahead. Each pass depends on the prior ones being clean.

Pass 1 — Assumptions

• Is there a dedicated Assumptions / Inputs tab?
• Are inputs color-coded (blue) and visually distinct from formulas (black) and links (green)?
• Does every load-bearing input have a documented source? (Cite: where, when, who.)
• Are inputs labeled with units (%, $, x, days)?
• Are growth-rate / margin / ratio assumptions reasonable vs. history?

Pass fails if: any load-bearing input has no source, OR inputs are scattered across multiple tabs.

Pass 2 — Mechanics

• Do the three statements tie? (Net income flows to retained earnings; cash flow ties to BS cash; FCF reconciles.)
• Working capital mechanics: do DSO / DPO / DIO drivers produce the AR / AP / inventory balances they should?
• Debt schedule: principal amortization + interest expense + ending balance all tie?
• Tax: effective rate applied? Loss carryforward / NOL mechanics?
• Equity: share count + stock-comp + buybacks all reconciled?

Pass fails if: the three statements don't tie, OR any major mechanical relationship is broken.

Pass 3 — Integrity checks

• Is there a balance-check row on the BS (= A − L − E, should equal 0)?
• Is there a cash tie-out (ending cash on BS = ending cash on CF)?
• Are there debug rows / consistency checks anywhere else (sum of segments = consolidated, etc.)?
• Are there any #REF!, #NAME?, #DIV/0!, #VALUE! errors in the model?

Pass fails if: any integrity check is missing or returns non-zero, OR any error value exists.

Pass 4 — Hardcodes

• Are there hardcoded numbers buried inside formulas? (Search for digit patterns inside formulas — *0.21, +12000, /365.)
• Are there hardcoded dates that should be driven from a date schedule?
• Are there hardcoded sheet references that should use named ranges?

Pass fails if: any hardcode is material (drives the answer) and isn't justified.

Pass 5 — Error-checks (IFERROR audit)

• Are there IFERROR wrappers masking actual errors?
• Are there IF statements that silently route bad data to a default?
• Are there 0 defaults where the right behavior is to fail visibly?

Pass fails if: any error-suppression wrapper hides a real bug.

Pass 6 — Scenarios

• Is there a scenario switch? (A single input that drives the whole model's scenario.)
• Are there at least three scenarios (base / upside / downside)?
• Do all three actually flow through the model end-to-end?
• Is there a sensitivity table on the top 2-3 drivers (revenue growth, margin, WACC for DCF)?

Pass fails if: scenarios exist but don't actually switch the model, OR sensitivity tables are missing on a model going external.

Pass 7 — Documentation

• Is there a Documentation / Cover tab?
• Does it state: version, last refresh date, owner, purpose, scope, key assumptions, known limitations?
• Are the tabs in a sensible order (Cover / Inputs / P&L / BS / CF / DCF / Outputs)?
• Are circular references either eliminated or documented (e.g., interest-on-cash-sweep)?
• Is the model self-explanatory to a new reader?

Pass fails if: there's no documentation tab, OR the model has undocumented circulars.

The deliverable

After running all 7 passes, the review produces:

Pass	Status (✅ / ⚠️ / 🔴)	Findings	Severity (P0 / P1 / P2)

Plus:

• Top 3 fixes (the highest-leverage corrections)
• Reasonableness check on key outputs (do they pass the smell test?)
• Recommendation (use as-is / use with caveats / do not use)

Severity guide

• P0 — model produces a wrong answer; fix before use.
• P1 — model is fragile / opaque, not currently wrong; fix before next refresh.
• P2 — model is correct but improvable (documentation, polish, structure); fix during next major rebuild.

See also

• Template: ../../templates/model-documentation.md
• Agent: ../../agents/financial-modeler.md
• Agent: ../../agents/valuation-analyst.md