# Finance Forge Shared Workflow

Use this reference for finance modeling and analysis skills that need a common contract.

## Core Principles

1. Ground before calculating: read the actual source files, filings, exports, workbook tabs, or user-provided facts before building claims.
2. Separate fact, derivation, assumption, and interpretation. Never let a hard-coded assumption masquerade as source data.
3. Preserve provenance. Every important number should trace to a source, a model formula, or an explicit assumption.
4. Make units and time periods visible. Record currency, scale, fiscal/calendar period, cohort, geography, and any normalization.
5. Prefer decision-useful outputs. A model or analysis should state what decision it supports and what would change the answer.
6. Verify with finance-native checks: tie-outs, subtotal checks, accounting identities, variance bridges, sensitivity sanity, and source freshness.
7. State uncertainty plainly. Use ranges, scenarios, and open questions instead of false precision.

## Artifact Ladder

Use the smallest artifact that can honestly answer the request.

- Intake note: problem frame, decision, sources, missing inputs, and recommended route.
- Analysis brief: answer, evidence ledger, calculations, assumptions, sensitivities, and open questions.
- Model plan: driver tree, source map, assumptions register, outputs, checks, and implementation units.
- Model artifact: spreadsheet or code model plus source ledger, checks tab/output, and verification log.
- Review memo: findings ranked by decision risk, with exact cell/file/source references where available.

## Evidence Ledger

For analysis or model work, maintain a ledger with these fields when material:

- ID: stable label such as `S1`, `A3`, or `D2`
- Type: source fact, derived calculation, assumption, interpretation
- Value or claim
- Source path, URL, tab/cell, filing section, or calculation reference
- Period and units
- Confidence: high, medium, low
- Notes: caveats, transformations, or stale-data risk

## Model Contract

A model is not ready to build until these are explicit:

- Decision supported
- Primary outputs
- Time horizon and grain
- Source data and refresh expectations
- Driver tree
- Assumptions and scenario knobs
- Required tie-outs and validation checks
- Known non-goals
- Handoff format: workbook, notebook, script, memo, or dashboard

## Verification Checks

Choose checks based on the work:

- Source tie-out: raw inputs match the cited source totals or exported records.
- Formula coverage: calculated lines are formula-driven, not accidental hardcodes.
- Unit consistency: currency, scale, and period conversions are correct.
- Accounting identity: statements, margins, cash flow, or working-capital relationships reconcile where applicable.
- Bridge integrity: start + drivers = end for variance, ARR, headcount, capacity, or cash bridges.
- Scenario sensitivity: upside/base/downside move in the expected direction and magnitude.
- Edge cases: zero volume, negative growth, missing data, partial periods, acquisition/disposal, churn, FX, one-time items.
- Presentation check: outputs answer the decision question without burying assumptions.

## Output Discipline

When reporting back:

- Lead with the answer or artifact created.
- Name the highest-risk assumptions.
- Mention verification performed and what did not get verified.
- Do not overclaim. Say "illustrative", "derived", "source-disclosed", or "assumption-driven" when that distinction matters.