Gbiz Proof Project Brief

---
name: gbiz-proof-project-brief
description: Scopes a two-week proof project on public data, producing one real business question, the deliverable, a scope and day plan, a source list, a skills plan and the lanes for what Claude will and will not do. Use for "run gbiz-proof-project-brief", "I have no experience to show", "give me a portfolio project idea", "scope a business project on public data", "what project can I do to get a graduate job", "plan a side project for my CV", "proof of work for a business analyst role", part of the Claude for Business Graduates Pack by Polar Bear.
---

# Proof Project Brief

## When To Use
You have no work experience to point to and want something real to show. Every graduate ad asks for evidence and your CV has a degree and a part-time job. This brief answers one question: which single piece of business work, on public data, can you finish in two weeks and defend line by line in an interview?

## When Not To Use
If you already have a live question from a job or placement, plan it with the Issue Tree instead; this brief is for a project you choose yourself. If the project is finished, go to the Portfolio Case Study.

## Inputs
- Two or three job ads for the roles you want (pasted text)
- Any idea you already have: a company, a local market, a sector trend
- Your university's rules if the project overlaps course work, and the hours you really have
If you have none of this, I start from one job ad and mark the output as a first draft.

## Approach
A work sample is the most direct evidence of skill: an employer can read it and ask about it. The question is scoped the way an issue tree starts, with a problem statement that is specific, measurable and owned by an imagined decision-maker, and it rests on public sources such as filings at Companies House. The failure it prevents: the project that grows from "one company's pricing" into "the future of retail", runs for two months and ends unfinished, with nothing to show.

## Workflow
1. Ask three questions: which roles you are targeting, how many hours a week you have for two weeks, and whether any of it touches course work (if so, it stays within your university's rules and the brief says so).
2. Read the job ads and pull out the two or three tasks they share (company research, a model, a deck). Propose three candidate questions that would exercise those tasks, each answerable from public data. You pick one; I do not pick for you.
3. Test the chosen question as a problem statement: specific (one company or market), measurable (a number or a yes or no at the end), with a named imagined decision-maker by role ("the finance director of a fictional competitor"). Rewrite until it passes all three.
4. Fix one deliverable: a short paper (executive summary and business case style) or a short deck. One, not both; two half-finished formats prove less than one finished one.
5. Set the scope: what is in, what is out, a plan by day block across two weeks, and a stop rule (the date you ship, whatever state it is in). List sources up front, public only; nothing confidential from a job or placement unless you have written permission.
6. Plan the skills from Domains 2 to 5 you will run, in order, and draw the delegation lanes: what Claude will do, what you will do, what Claude will not do. Start an AI Work Log on day one, not day ten.
7. Check the brief against the ads: if an interviewer read only the question and the lanes, would they see the job's tasks? If not, change the question, not the ads.

## Output Format
```markdown
# Proof Project Brief
## Question
[One sentence problem statement] | Decision-maker: [role, fictional] | Answer looks like: [number or yes/no]
Tasks from the job ads it exercises: [task: project step; task: project step]
## Deliverable and scope
Deliverable: [short paper OR short deck] | In: [list] | Out: [list] | Stop rule: [ship date]
| Day block | Work | Skill used | Output |
|---|---|---|---|
| [days 1 to 2] | [work] | [skill display name] | [output] |
Sources, public only: [filing, report or dataset: what it gives, date checked]
## Delegation lanes
| Claude does | You do | Claude never does |
|---|---|---|
| [task] | [task] | [task] |
Course overlap: [none / within your university's rules: rule quoted]
## Decision
[You confirm the question and stop rule by [date]; AI Work Log entry one is written that day.]
```

## Done When
- The question passes the specific, measurable and decision-maker test
- One deliverable, a stop rule and a day plan that fits your real hours
- Every source is public and listed with a date
- The "Claude never does" column is filled in, and the work log is started

## Quality Bar
- Companies appear only through public filings and published reports; no data about private individuals.
- No invented figures in the brief; unknowns are placeholders to find.
- The question comes from the job ads, not from what is easy to search.
- Two weeks means two weeks: cut scope before you extend the date.
- The project is your own work on public data; Claude helps, you decide and do the thinking.

## Next
Run gbiz-ai-work-log (AI Work Log) to record the work as you do it.

## About the makers

This pack is made by Polar Bear, a consultancy built by ex-McKinsey founders with a dream to make AI work for People, not instead of them. We help our clients build people systems and AI-first ways of working, and we run our own company on Claude. If your team has outgrown the self-serve version, message Pauline (linkedin.com/in/paulinebertry).