# One-page capstone brief
Original worksheet by Codex | September 28, 2026

## User and problem
A specific user:
A recurring task they struggle with:
How they do it today:
Evidence from conversations or observation (label assumptions):
Why an AI step might help:
Why search, rules or an ordinary UI might be sufficient:

## Scope
One workflow I will support:
Three things I will explicitly leave out:
Data sources, ownership and permitted use:
Public/synthetic/private data classification:
Data retention and deletion plan:

## Before I build
Baseline:
Representative development cases:
Held-out cases:
Success criterion and measurement method:
A critical failure that blocks release:
Budget and latency target (my own assumptions):

## Architecture
Input -> validation -> retrieval/tools -> model -> output validation -> user.
Where are authentication and permissions checked?
Where are source IDs retained?
Which actions need a person to approve?
What happens when the model is unavailable?

## Milestones
1. Baseline works and is measured.
2. AI feature works on ordinary examples.
3. Evaluation compares both versions and exposes failures.
4. App runs for a second person.
5. README, demo and case study are ready.

## Evidence of completion
Repository:
Demo:
Evaluation result with sample size:
Three observed failures:
Feedback from users (number and context):
What I personally built, tested and decided:
What AI assistance contributed: