Part 3 of 4 · 2 min read
Exercises
Easy
Easy 1: Run it (30 minutes)
Run python -m pytest -q and python run_all.py in code/mini-scope.
Done when: you can explain each table in the results.
Easy 2: Write your scope sentence (45 minutes)
Write "For [user], my agent does [job], measured by [metric] from [baseline] to [target]" for your capstone.
Done when: every bracket holds a specific person, task or number.
Easy 3: Check it (30 minutes)
Put your scope into a Scope object and run check.
Done when: you have either fixed or justified each flag.
Medium
Medium 1: Your estimates (3 hours)
List your features with likely hours and priorities, run the simulation for your available hours and find your cut line at 50, 80 and 90 percent.
Done when: you can say which feature you will drop first and why.
Medium 2: Dependencies (3 hours)
Extend the simulator so that some features cannot start until others finish, and compare the finish time with the independent model.
Done when: you report how much the 80th percentile moves.
Medium 3: Baselines (3 hours)
Build a rules-only and a single-prompt baseline for your task and measure both on 100 labelled items.
Done when: you report both scores with intervals and set your target above them.
Hard
Hard 1: Track actuals (8 hours, over four weeks)
Estimate each task with a low, likely and high figure, log actual hours and compare with the simulated percentiles at the end. (I did not run this, so your numbers are the ground truth.)
Done when: you report where reality fell in your predicted distribution and recalibrate the spread.
Hard 2: Measure your own last mile (8 hours)
Record the score on a fixed evaluation set after each improvement session and fit the error against time. (I did not run this.)
Done when: you report whether the error halves at a constant rate and the fitted halving time.
Hard 3: Staged rollout plan (8 hours)
Write a Crawl-Walk-Run plan for your agent with acceptance-rate gates, and test the gate on a sample of real or realistic cases. (I did not run this.)
Done when: you report the acceptance rate on the sample and the decision it implies.