Forecast & Scenario

Forecasting feels like the scariest power-skill and is one of the most valuable. The trick isn’t fancy maths — it’s making assumptions explicit so the number is trustworthy and adjustable.

What you’ll build

A Skill that takes historical data + your assumptions and returns:

• The base case — the forecast, with the method stated in one line.
• The assumptions — listed plainly, each one changeable.
• Scenarios — best / base / worst, and what drives the gap.
• The sensitivity — which single assumption moves the answer most.

How to build it

1. Keep your standing assumptions in assumptions.md (growth rate, conversion, seasonality).
2. Decide the data source (file or connector) and the horizon.
3. Encode the method and the “show your working” rule in SKILL.md. Use the code/analysis tools so the maths is real, not vibes.

The SKILL.md

---
name: forecast-scenario
description: Produce an explainable forecast with base/best/worst scenarios and a sensitivity check.
---

When asked to forecast:
1. Read the historical data and the standing assumptions in assumptions.md.
2. State the method in one sentence (e.g. "trailing 3-month run-rate with stated growth").
3. Compute the base case using the analysis tool — show the calculation, don't assert a number.
4. Produce best / base / worst by flexing the key assumptions; show what changed.
5. Report which single assumption the answer is most sensitive to.
6. Flag clearly that this is a model, not a promise; list every assumption the user can adjust.

Watch-outs

• Explainability over precision. A simple model you can defend beats a black box.
• Make assumptions editable in one place — that’s what lets non-analysts own the forecast.
• This is the Skill where Diligence matters most: never present a model as a guarantee.