Skip to main content

workflow-ma-formula-consistency

WHAT: Formula consistency audit across financial model worksheet rows — detects broken formula rows (numeric value where neighbour cells are formulas), formula drift (formula changes mid-row without rationale), and row-level formula logic inconsistencies. WHEN: Invoke when a financial model is under audit and the reviewer needs to confirm that time-series formula rows are internally consistent across all periods, or when a suspicious single-period variance suggests a formula has been manually overridden.

Quellinformationen

Repository
fall-development-rob/corp_finance
Letzte Quellaktivität
11. Mai 2026 um 14:10
Erkannte Sprache von SKILL.md
Englisch
Sterne
8
Forks
2

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

SKILL.md wird angezeigt

SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
workflow-ma-formula-consistency
description
WHAT: Formula consistency audit across financial model worksheet rows — detects broken formula rows (numeric value where neighbour cells are formulas), formula drift (formula changes mid-row without rationale), and row-level formula logic inconsistencies. WHEN: Invoke when a financial model is under audit and the reviewer needs to confirm that time-series formula rows are internally consistent across all periods, or when a suspicious single-period variance suggests a formula has been manually overridden.
# Model Audit — Formula Consistency ## What this skill covers Formula-row consistency checking: verifying that each projected period in a financial model row uses the same formula structure as its neighbours, and flagging deviations that indicate a broken, overridden, or manually edited cell. ## Workflow ### Step 1 — Inventory all formula rows For each row in the model that spans multiple time periods: - Classify each cell as formula (`=`) or hardcode (numeric literal). - Record the formula string for each formula cell. - Record the cell address and period (e.g., FY2025, FY2026, ...). ### Step 2 — Identify broken formula rows A broken formula row is any row where: - At least two adjacent cells are formulas but one intervening cell is a numeric literal in a projection column. - The numeric value in the gap cell is not on the permitted-hardcode list (raw historical actuals, named-assumption block drivers, universal constants 12/365/100). Flag every broken-row cell as severity: **Critical** (if a material output line) or **Major** (if an intermediate calculation). ### Step 3 — Detect formula drift Formula drift occurs when the formula string changes mid-row without a structural reason. Compare each cell's formula to the prior period's formula: - Structural change = a different function, reference pattern, or operator (e.g., absolute vs relative reference switched). Flag for review. - Parameter change = only the referenced cell address shifted by one column (normal relative-reference roll). Do not flag. - Literal embedding = a numeric literal was introduced into the formula string (e.g., `=B5*0.25` where prior period was `=B5*Assumptions!$C$4`). Flag as **Major** — this is an embedded hardcode. ### Step 4 — Summarise by section Group findings by model section (Income Statement, Balance Sheet, Cash Flow, Debt Schedule, Operating Model, etc.). Report: - Total formula rows reviewed. - Total broken cells found. - Total formula-drift cells found. - Worst-severity finding per section. ## Output format ``` FORMULA CONSISTENCY REPORT -------------------------- Model: [name / file reference] Total formula rows reviewed: [n] Broken formula cells: [n] | Formula-drift cells: [n] FINDINGS TABLE | Row label | Sheet | Period | Cell | Finding type | Severity | Model value | Expected (formula) | Note | |-----------|-------|--------|------|--------------|----------|-------------|-------------------|------| | Revenue | IS | FY2026 | C10 | Broken row (hardcode in formula row) | Critical | 1,200 | =B10*(1+Growth!$C$3) | Manual override suspected | | EBITDA Margin | IS | FY2028 | E14 | Embedded literal | Major | =E13*0.22 | =E13*Assumptions!$C$7 | Literal 0.22 embedded | ``` ## Quality gates - Every projection row spanning ≥2 periods is reviewed — no rows skipped. - Broken cells in P&L headline rows (Revenue, EBITDA, Net Income) are always Critical. - Embedded literals in a formula that differ from the assumption block are always Major. - Findings table is sorted by severity (Critical first). ## Related skills - `workflow-ma-link-tracing` — run first to build the cell inventory before formula consistency checking - `workflow-ma-hardcode-detection` — hardcode detection is complementary and runs in parallel; share the cell inventory - `workflow-ma-rederivation` — after consistency is confirmed, re-derivation validates the correct formula values against corp-finance-mcp tools ## Routing **Primary agent:** `cfa-chief-analyst`
Auf GitHub ansehen