Project Step: Problem Brief and Harm Statement
Unit ID: ML-M01-U06 Estimated active time: 30-45 minutes
Bring the pieces together
You have now seen the parts of a strong ML problem frame.
You can answer:
- What kind of task is this?
- What is the target?
- What is the unit of analysis?
- When is the prediction made?
- What action follows?
- Who uses the result?
- Who is affected?
- What can go wrong?
- What simpler options should we compare?
- Is the project feasible enough to continue?
Now you will write the first problem brief.
This brief is not a long report. It is a thinking tool. It helps you avoid building the wrong model.
Problem brief template
Use this structure.
| Field | Write this |
|---|---|
| Practical goal | What real-world process are we trying to improve? |
| Task type | Prediction, description, automation, decision support, or a combination |
| Target | The value the model would predict |
| Unit of analysis | What one prediction represents |
| Prediction time | When the prediction is made |
| Outcome window | What future period the target covers |
| User | Who uses the result |
| Action | What the user does with the result |
| Affected people | Who may be helped or harmed |
| Baseline | The simple method the model must beat |
| Success criterion | What improvement would make the model worth using |
| Unacceptable failure | What should stop or redesign the project |
| Feasibility decision | Continue, revise, or stop |
Worked example
Here is a first brief for the learner-support example.
| Field | Example answer |
|---|---|
| Practical goal | Help learners receive support early enough to continue Module 1. |
| Task type | Prediction used for decision support. |
| Target | Whether a learner completes Module 1 by day 10. |
| Unit of analysis | One learner's first attempt at Module 1. |
| Prediction time | End of day 3 after the learner starts Module 1. |
| Outcome window | Completion by day 10. |
| User | Course mentor. |
| Action | Offer an optional support message on day 4. |
| Affected people | Learners who are flagged or not flagged. |
| Baseline | Message learners who have not opened Module 1 for three days. |
| Success criterion | Find more learners who need support than the simple rule, without sending many unnecessary messages. |
| Unacceptable failure | The score is used to punish, shame, block, or rank learners. |
| Feasibility decision | Revise or continue, depending on label quality and mentor capacity. |
Notice the careful wording.
The brief does not say:
The model will identify why learners fail.
It says:
The model may help mentors decide where to offer support.
That difference matters.
Harm statement
A harm statement names the main risk in plain language.
For our example:
This project could harm learners if a support score is treated as a judgement about ability, if learners are contacted in a way that feels intrusive, or if the model misses groups whose learning patterns are not well represented in past data.
A harm statement is not written to stop all projects. It is written so the team can design safeguards.
Your submission
Create your own problem brief for one of these cases.
- Learner support in an online course.
- Support-ticket priority for a small education company.
- Course-resource recommendation for learners who ask for extra practice.
Do not use real personal data. Use the case description only.
Your answer should include:
- the full problem brief table;
- a harm statement in plain language;
- one baseline option;
- one reason to continue, revise, or stop.
Reflection
Finish with two sentences.
- The model should not be built if...
- The model result should not be used to...
Takeaway
The problem brief is the first safety check in an ML project. It makes the work clear before algorithms, code, and metrics arrive. In Module 2, you will inspect the data behind the brief.