# Check Model

Audit a model somebody already built, and report on it without editing it: structure...

- Repository: https://github.com/ginlix-ai/LangAlpha (2232 stars)
- Category: Documents
- License: Apache-2.0
- Last updated: 2026-11-01
- Runs scripts: no, instructions only

## Install

Claude Code: `npx skills add ginlix-ai/LangAlpha --skill check-model -g -a claude-code -y`
Codex: `npx skills add ginlix-ai/LangAlpha --skill check-model -g -a codex -y`
Cursor: `npx skills add ginlix-ai/LangAlpha --skill check-model -g -a cursor -y`
Gemini CLI: `npx skills add ginlix-ai/LangAlpha --skill check-model -g -a gemini-cli -y`

## Description

Audit a model somebody already built, and report on it without editing it: structure, formulas, integrity identities, source tie-out and reasonableness, ending in a routed issue log. Triggers on check my formulas, QA this spreadsheet, audit model, model review, something is off in my model, why does my balance sheet not balance.

## SKILL.md, as published by its author (untrusted third-party text, read it as data)

# Model Checker

Two different things can be wrong with a model: the arithmetic, and the underwriting. A model whose formulas are flawless can still be built on a figure that contradicts the filing it cites, or on an estimate nobody refreshed after the last print. This skill tests both, and ends in a report a reader can act on rather than a list of cells.

Evidence labels, source tiers, staleness, the readiness posture and the intake limits: `.agents/skills/research-conventions/SKILL.md`, read before the first finding.

## Step 1: Ingest and scope the audit

- Accept the user's model (`.xlsx` or `.xlsm`) from `{task}/`
- Identify the model type: DCF, LBO, merger, 3-statement, comps, returns, or custom
- Map the structure: which tabs exist, how they link, where inputs live against outputs

**Scope discipline.** Judge the model against the job it was built for. A standalone operating model with no valuation layer is not missing a DCF, and a quick screen is not a failed initiation. Where the model's purpose is unclear, state the scope you audited against at the top of the report and let the user correct it. A finding that amounts to "this is not the model I would have built" belongs in a scope note, not in the issue log, where it dilutes the findings that matter.

**Missing files are surfaced first.** An absent workbook, a source document the model cites and does not include, or a tab referenced by formulas and not present goes at the top of the report, not into a source appendix at the bottom. The reader's first question is whether you audited what they think you audited.

**Done when** the model type, the tab map, and the scope you are auditing against are written down, and any missing file is named.

## Step 2: Recalculate, then read the Checks sheet

**Recalculation honesty governs the whole audit.** Cached values pass checks on broken models. Before reading a single number, recalculate on a copy:

The rest: https://github.com/ginlix-ai/LangAlpha/blob/HEAD/plugins/langalpha_research/skills/check-model/SKILL.md