Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

FINANCE.md

The standard that encodes financial conventions for AI agents and modeling tools.

Like CLAUDE.md encodes a project's conventions for Claude, like AGENTS.md encodes agent capabilities — FINANCE.md encodes an organization's financial conventions for any agent or tool that models.

📄 Read the standard on the web · 📖 Why it exists


The problem

Every time you start a session with an AI agent on a financial model, you repeat yourself:

"Amounts are in thousands. Costs are positive. Books close on the 10th. We report on a management basis, not IFRS. 'Marge brute' here means revenue minus purchases only — it's not gross profit."

This context lives in your head, gets pasted into prompts, and evaporates between sessions. There is no persistent, machine-readable layer that tells agents how this organization thinks about money.

FINANCE.md is that layer.


What it is

A FINANCE.md file is a short YAML front matter block (machine-readable) followed by a free-prose Markdown body (human-readable). It encodes:

  • Metadata in the front matter: model name, currency, language, sources, owners, glossary — the small set of facts an agent needs at a glance.
  • What the model is for in the body: what it produces, for whom, by when. An agent building a lender-facing budget behaves differently from one filling in a template.
  • How the statements are built: the sections and key lines of the P&L, cash flow and balance sheet, and how they connect — net income into retained earnings, closing cash back to the cash flow. This is what stops an agent from inventing a structure nobody asked for.
  • Conventions and rationale: units, sign, closing calendar, and any accounting policy the model actually has (EBITDA basis, net debt perimeter, working capital), and why.

It is not:

  • A generic modeling playbook (that's your cookbook)
  • A model definition (that's your MDK or schema)
  • A data format (that's your API)

The format is intentionally compact — a 100-line file beats a 1 000-line YAML every time. Conventions belong in prose; the YAML carries only what is genuinely structured.


Format (v0.1 — draft)

---
finance_md: "0.1"
model_name: "Distrib SAS — FY2026 budget"
currency: EUR
secondary_units: [tonnes]
language: fr
owners: ["cfo@distrib.example"]
sources: [pennylane, manual_excel]
glossary:
  Marge brute commerciale: "Revenue minus purchase cost only. Excludes logistics."
  Tonnage: "Volume delivered, in tonnes. Capacity-planning metric, not a billing unit."
---

# Conventions

## EBITDA

Excludes IFRS16 — French GAAP entity. Operating leases stay above EBITDA as
rent expense. Sponsor convention; this is the bank-facing definition.

## Working capital

DSO on revenue (45 days target), DPO on purchases (50 days). Inventory tracked
separately in days of COGS.

Required front-matter fields: finance_md, model_name, currency. Everything else is optional. See SPEC.md for the full reference.


Hierarchy

A FINANCE.md can declare a parent via inherits_from. The hierarchy is typically:

Sector       FINANCE.md  ← published reference (PE LBO, SaaS FP&A, infra…)
  └── Organization  FINANCE.md  ← group-wide
        └── Model    FINANCE.md  ← deal- or model-specific

Inheritance is declarative: a child file extends or contradicts its parent, but consumers are responsible for merging.


Using FINANCE.md

Claude — system prompt

Paste the file contents (front matter + body) into your system prompt:

You are a financial modeling assistant. The model's financial conventions
are defined below. Always apply them unless the user explicitly overrides one.

<<< FINANCE.md >>>

Claude Code — CLAUDE.md

Reference it in your project's CLAUDE.md:

@FINANCE.md

Claude Code picks the conventions up at session start.

Cursor — project rules

Place FINANCE.md at the root of your financial-modeling project. Cursor auto-injects project-level files. Add an explicit reference in .cursor/rules:

Always read and apply FINANCE.md before any financial modeling task.

MCP servers

Any MCP server can expose a read_conventions tool that returns the FINANCE.md for a given model. Layerz, for example, ships three tools (layerz_get_finance_md, layerz_set_finance_md, layerz_generate_finance_md) that let an agent push its local FINANCE.md into a Layerz model at session start, then operate under those conventions.


Placement

  • Organization-level: root of a financial-modeling repository (e.g. models/FINANCE.md)
  • Model-level: alongside a specific model (e.g. models/ma/deal-x/FINANCE.md)

Note: FINANCE.md belongs in a financial-modeling project, not at the root of a generic company repository where the filename could be confused with expense policies or HR finance docs.


Examples

Use case File
SME — monthly budget examples/sme-budget/FINANCE.md
LBO / M&A examples/lbo/FINANCE.md
SaaS FP&A examples/saas-fpa/FINANCE.md
Infrastructure / Concession examples/infra/FINANCE.md

Start from TEMPLATE.md for a blank scaffold.


Status

This repository is a 0.1 draft. The format is being explored in the open: field names, defaults, and validation rules may change before a 1.0 is cut. Treat the current schema as a working proposal, not a stable contract.


Roadmap

  • Initial draft (v0.1, this repo)
  • Schema 1.0 stable
  • MCP reference implementations
  • Validator CLI (npx finance-md validate)
  • Convention linter (validate agent outputs against FINANCE.md)
  • Translations / locale-specific conventions

Where this fits

FINANCE.md is one of a small family of open entry points to Layerz:

Project Role
finance-md (you are here) The standard: how an organization encodes its financial conventions for any agent
cookbook The recipes: prompts and patterns for doing financial work with AI agents
Layerz MCP The tool: how an agent drives Layerz to build structured, versioned models
slides-for-claude The presentations: turn a model or a topic into a self-contained HTML deck

The standard tells an agent what your numbers mean. The tool gives it a place to build that does not drift. The cookbook shows what to ask for in the first place.


Contributing

This is an open standard. Contributions welcome:

  • Additional examples for common deal types
  • Schema extensions
  • Tool integrations and use-case guides

Open a PR or a discussion.


License

MIT — see LICENSE.


Initiated by Layerz

About

The standard that encodes financial conventions for AI agents and modeling tools

Topics

Resources

Contributing

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors