The setting
Bramble Media is a content studio that produces product descriptions at scale for e-commerce clients. You are a prompt engineer, and this quarter your client is Northglen Outfitters, an outdoor apparel brand.
Every mission on this path happens at the same company, so context carries over the way it does in a real job: the data you cleaned in mission two is the data the finance lead questions in mission four.
The missions
1. task framingstarter
Frame the Northglen product-description task at Bramble Media
Content lead Mei Fong just handed you the Northglen Outfitters account: 500 product descriptions from a data feed, in Northglen's voice, no invented claims. Before writing a single prompt you need to frame the task so nobody argues about done later.
You deliver: A task specification defining inputs, outputs, constraints, edge cases, and acceptance criteria.
Scored on: Output elements, Constraints with numbers, Edge cases named, Checkable acceptance criteria.
Working from: client_brief.md, sample_products.csv.
2. prompt draftstarter
Draft the Northglen description prompt
With the task framed, write the prompt that will generate all 500 Northglen descriptions. It has to bake in the constraints so the model does not drift on word count, voice, or invented specs.
You deliver: A complete prompt: a system prompt plus a per-product template with placeholders and missing-value handling.
Scored on: Encodes the constraints, Voice is set, Field placeholders, Missing-value handling.
Working from: voice_guide.md, product_feed.csv.
3. structured outputcore
Enforce a clean JSON schema on the Northglen outputs
Northglen wants the descriptions delivered as structured JSON so their CMS can ingest them, not as loose paragraphs. Right now the model returns messy output that breaks the importer, and Mei needs it locked down.
You deliver: A structured-output fix: the list of failing examples with violations plus revised prompt and schema guidance.
Scored on: Finds every violation, Explains each failure, Revised prompt guidance, Null over invention.
Working from: output_schema.md, bad_outputs.md.
4. evaluationcore
Pick the winning prompt variant for Northglen and name the regression
You ran two prompt variants against the same products and scored them. Variant two reads better, and the team wants to ship it. But better voice is not the only thing that matters for a client who forbids invented claims.
You deliver: An eval report naming a winning variant, the quantified regression, and a recommendation.
Scored on: Reads every column, Quantifies the regression, Defensible winner, Actionable recommendation.
Working from: prompt_ab_results.csv.
5. reliability passstretch
Harden the Northglen prompt against its production failures
Northglen went live, and production surfaced failures the test set never showed. Mei needs a reliability pass before the next 2,000 products run so these do not repeat at scale.
You deliver: A reliability plan grouping the failure modes with a mitigation for each and a validation checklist.
Scored on: Covers every failure mode, Specific mitigations, Validation checklist, Addresses invented claims.
Working from: failure_log.md.
How the scoring works
Each deliverable is graded against the rubric written for that mission. Separately, every mission on every path is graded on how you used AI, against the same four criteria:
- Understood the task. The learner framed the goal for the assistant clearly instead of pasting the brief and hoping.
- Grounded in the material. The learner directed the assistant into the provided files and based the work on them, not on invented facts.
- Verified the output. The learner checked claims, numbers, or coverage against the source material before submitting.
- Iterated with judgment. The learner refined weak parts of the draft with specific follow-ups rather than accepting the first answer.
Both scores, with the work behind them, go on your proof profile. That is what makes a claim like "I can use AI for prompt engineering" something an employer can check.