NewAI Content Creation is now live in Early Access
Turning Point Training Instituteby Training Center
Skip to content
0%
Module 7 · Test the resultLesson 1 of 2
Video lesson

L13 · Test the decision, not the confidence

IntermediateVideo lesson
5 min readIntermediate
Downloads & resources
Download this lesson's materials, ready to use.

Skill packs and prompts are provided in English so they work best with the AI tools. Use them as-is.

Test the decision, not the confidence

Learning time: about 33.3–49.3 active minutes. Choose ONE presentation: the text (about 4.3 minutes, estimated) OR the lesson video (3:20, measured). Then 30–45 minutes of practice and checking (estimated). Do not count both presentations. Setup delays and optional extra study are separate; learner timing has not yet been piloted.

Example prompt

Where you are: Module 7, lesson 1 of 2.

You have a first calculator and a reviewed change. Now you will establish whether its numbers and boundaries support the decision. Your evidence is a test log containing inputs, expected results calculated independently, actual results and the exact tested version. O7 is the main outcome; O4 matters when a reviewer finds a problem.

What you will do

Check a normal result, a boundary, an invalid input and a changed requirement. You do not need to read JavaScript to determine whether 12 seats at 30 produce 360 revenue. You do need to know what the calculation means and which assumptions it excludes.

Teaching

A test begins before you press Calculate. Write what should happen. If you let the same agent build the tool and invent every expected answer from its own output, the check can repeat its mistake. Use paper or a separate spreadsheet for at least two cases. A second agent can help inspect the evidence, but agreement is not independent arithmetic.

Revenue equals attendance multiplied by price. Total cost equals fixed cost plus attendance multiplied by variable cost. Surplus is revenue minus total cost. Break-even asks how many attendees cover costs. With positive contribution per attendee, divide fixed cost by contribution and round upward to a whole person. Do not round the contribution early. Price at or below variable cost needs special handling; dividing by zero is not an acceptable response. Zero fixed cost is another boundary: zero attendance may break even without making the business profitable.

A passing test applies to the saved version you tested. After a change, repeat affected checks and at least one ordinary case. Record a failure before repair so you can show what changed. “All tests pass” without inputs, actual results or version is an assertion.

Worked example

Use a different mini event: fixed cost 90, price 25, variable cost 10 and capacity 8. Contribution 15 means six attendees break even. At five, revenue 125 and cost 140 produce a loss 15. The calculator might correctly show that loss while a memo still says “ready to launch.” That is a decision defect, even if the arithmetic works.

For a capacity of five, break-even cannot be reached under those inputs. You may propose a cost or price change for the owner to consider. You may not silently alter the facts to make the answer positive. Test the memo and tool together.

Do it yourself

Open your Workshop Launch Kit working copy. Before reading the supplied practice feedback, calculate the 12-attendee and 8-attendee results yourself. Then use project-starter/TEST-CASES.md to check your tool. Fill templates/REVIEW-AND-TEST-LOG.csv with actual outputs and the version. Include a blank price, fractional attendance and attendance above capacity. A visible error must clear any old result so a colleague does not mistake stale numbers for the new answer.

Use this bounded reviewer request:

Example prompt

Review my saved calculator and test log as read-only. The project brief supplies the inputs and acceptance criteria. Verify my two independent calculations; inspect normal, invalid and changed-input results. Report the exact case, observed behavior, expected behavior and consequence. Do not repair or publish. Distinguish a tested result from code inspection and identify any missing evidence.

Apply the changed-capacity case only after preserving the initial version. Ask the builder to fix confirmed defects within the calculator and memo only. Re-run affected checks and give the reviewer the new version.

Feedback and retry

Pause before checking your work. At 12, the supplied workshop earns 360 against 276 costs, leaving 84. At 8, it earns 240 against 244, losing 4. Break-even is 9. Capacity 8 makes that break-even infeasible. An invalid input should produce an explanation, not a plausible number. If your result differs, check which input, unit or formula caused it before rewriting everything.

For a fresh retry, use fixed cost 100, variable cost 5, price 25 and capacity 6. Test attendance 4 and 7. Expected: four gives a loss of 20 (revenue 100, cost 120) because break-even is five; seven is rejected because it exceeds capacity. Explain why a calculated surplus for seven would still be an unacceptable tool result.

Check your understanding

Choose an answer and write one sentence explaining why before you read the answers.

  1. A tool passes all its numeric tests, but the memo treats 12 interest responses as 12 bookings. Can you accept the project?
    • A. Yes: the numbers are correct.
    • B. Yes, if a second agent agrees with the memo.
    • C. No: the interpretation exceeds the evidence; the tests need a data and decision check as well as numeric checks.

Answers

  1. C. No. The interpretation exceeds the evidence. The tests need a data and decision check as well as numeric checks.

Summary

Keep independent expectations, actual outputs, failures and retests together. Accept a result only within the assumptions and version checked.

Sources

Course-owned synthetic calculations and acceptance model. Technical review boundaries: T12 in the course source register (Official technical research, in the course resources); Git state and review distinction: T17/T19. The reference calculator tests are recorded in the course production record (not a learner download); those tests do not prove your own copy works.