Lab 13Lab 13: the due-date calculator in two languages

Time: about 3 hours 40 minutes. Lab files: brightline-lab-ch13.zip, standalone. The same steps are in LAB.md inside the zip.

Before you start

What you need.

• The Claude desktop app with a Pro, Max, Ultra, Team or Enterprise plan, for Tasks 1 to 9 and 12.
• The ChatGPT desktop app with any plan, for Task 10, and OpenCode Desktop with a model connected, for Task 11, as in Lab 12.
• Git, Node.js 24 or later, and uv, the Python project tool.

Set up the tools. If they are not on your computer yet, follow Concept 13.2 before Task 1. It gives the official page for each one, and a prompt that has the agent check the versions and install nothing without your yes.

The core path and the extensions. The core path is Tasks 1 to 9, at least one port (Task 10 or 11), and Task 12. A port makes the same change with a second agent. The second port is an extension.

For Python only, or without a paid Claude plan, see the two sections near the end of this page.

The lab files. Download brightline-lab-ch13.zip and unzip it into a new folder. It contains:

• brightline-ap/, a plain copy of the repository at the end of Lab 12, with no git history. It also holds the lab files: a Python version of the vendor list in py/, data/invoices-check.csv, data/invoices-no-terms.csv, and two scripts in exercises/. In Task 1, the agent puts it under git. This is the only folder you give an agent.
• concept-blocks/, every runnable code block from the chapter.
• LAB.md, and field-guide-code-toolchain.md, the field guide for the code toolchain, with more detail and a table for fixing problems.

Each task with an answer key links it as a separate download. Open a key in a text editor, never in the agent. An agent that can read the answers is not tested by them.

Use the zip's copy for this lab, even if you finished Lab 12 in your own repository. The answer keys match the zip's files exactly.

The date in this lab. Today is still Wednesday, March 3, 2027, and the payment run is Friday, March 5.

The job

Brightline's due-date rule is the invoice date plus the days in the terms. It is hidden inside the Chapter 12 script. Nobody has tested it on a month end, a year end or a leap day, and nobody knows what it does with terms it has never seen. Ana wants it moved out into its own function, with a contract and full tests, in both languages, so the team can trust it and reuse it. Then she wants a script that checks every stated due date in an export against the rule.

Ana's note from Lab 12 is still your check: after every change on main, npm run due -- 2027-03-05 must still print 8 invoices and $19,229.53, as her note says. Reading the rule lines shows you the rule itself, which that check cannot see. You will read the code the agent wrote and predict what it does, decide the test cases before the agent writes new code, and port one change to one or two other agents. No number in this lab comes from an AI's reply.

How each task works

Each task has a goal, steps and a Checkpoint. Do not start the next task until the checkpoint is true. In the Claude desktop app, work in Manual mode.

The agent runs every command in this lab, such as git, npm and uv, and you never type one in a terminal. Before you approve a command, read what the agent says it will do, and say no to anything you do not understand. Asking "What does this command do?" is always fine. Approve each command when you agree with what it does, and then read what it prints. Write each prediction in NOTES.md before the agent runs anything.

You decide every test case and its answer. That is the first 10 percent of 10-80-10, where you say what correct means. The agent writes the test code from your list. Read each test against your list before you approve it. Once a test is approved, the agent never changes it.

Task 1. Set up the workbench (20 minutes)

Goal: one command for each check, in each language.

1. Open brightline-ap in the Claude desktop app's Code tab, in a Local session, in Manual mode.

2. Send:

   Put this folder under git, with a main branch, and make a first commit of everything here, with the message "Start from the end of Lab 12". Then show me git log --oneline.

   Expect git init -b main, git add ., git commit and git log --oneline, which shows one commit. If the agent wants to change your git settings for the whole computer, say no. Ask it to set the name and email for this project only.

3. Send this. You do not need to know what each setting in the prompt means: they are the project's fixed settings, and step 5 compares them with the answer key. The only packages it should add are typescript, @types/node, pytest, pyright and ruff. Say no to anything else. These are dev dependencies: packages the project needs only while you develop and check it, not when it runs. A package is other people's code published under one name. Chapter 14 covers packages and dependencies in full.

   Please set up the type checker and the Python tools, with exactly the settings below and nothing more. Ask me before each command, and tell me in one sentence what it does.

   1. Add typescript and @types/node@24 as dev dependencies. Create tsconfig.json for type checking only, with target esnext, module and moduleResolution nodenext, strict, noEmit, allowImportingTsExtensions, erasableSyntaxOnly, verbatimModuleSyntax, skipLibCheck, and types set to node. Include src, tests and scripts.
   2. Add three npm scripts: typecheck runs tsc --noEmit, test:py runs cd py && uv run pytest, and check:py runs cd py && uv run pyright && uv run ruff check.
   3. py/ already holds the package brightline_ap and its tests. Write py/pyproject.toml yourself, and do not run uv init. Name the project brightline-ap, require Python 3.12 or later, with no dependencies. Set pytest's testpaths to tests and pythonpath to ., Pyright's typeCheckingMode to strict, and its include to brightline_ap and tests. Then add pytest, pyright and ruff with uv add --dev.
   4. Add node_modules/, py/.venv/, __pycache__/, .pytest_cache/, .ruff_cache/, .env and out/ to .gitignore.

4. Ask the agent to run npm test, npm run typecheck, npm run test:py and npm run check:py, and read the output of each.

5. Download Task 1's answer key. Compare tsconfig.json and py/pyproject.toml, the Python project's settings, with it. Then ask the agent to commit with the message "Set up the workbench", and to list the files in the commit. Make sure the lock files, package-lock.json and py/uv.lock, are among them. They record the exact version of every package.

Checkpoint: all four commands pass, with no failed test and no errors from the checkers. npm test shows 7 tests and npm run test:py shows 2.

Task 2. Predict, run, investigate (20 minutes)

Goal: read both vendor scripts before you run them.

1. Open src/list-vendors.ts and py/brightline_ap/list_vendors.py in a text editor, and read them side by side. Open data/vendors.csv too, the file both scripts read. It tells you how many lines to expect, and which vendor comes first and last.
2. In NOTES.md, under a heading "Task 2", write what each one prints: how many lines, which vendor first and last, and exactly how one line looks.
3. Ask the agent to run npm run vendors, and then, in py/, uv run python -m brightline_ap.list_vendors. Read what each one prints.
4. Mark each part of your prediction right or wrong. For the first thing you got wrong, ask the agent about that line, with its number in place of N: "Explain line N of py/brightline_ap/list_vendors.py, and why it prints what it prints." If the line is in the TypeScript script, use src/list-vendors.ts in the question. If you got everything right, ask instead: "What input would make list_vendors.py or list-vendors.ts print something wrong?"
5. Paste the real output under your prediction.

Checkpoint: NOTES.md holds your prediction, the real output, and one question you asked with its answer in your own words.

Check it now. Download Task 2's answer key.

Task 3. Modify (15 minutes)

Goal: change one thing and predict the result first.

1. Predict the new order and write it in NOTES.md. The vendors are sorted by the days in their terms, fewest first, with "Due on receipt" as 0 days. Vendors with the same terms stay in file order.

2. Send:

   In src/list-vendors.ts and py/brightline_ap/list_vendors.py, print the vendors sorted by the number of days in their terms, fewest first. "Due on receipt" is 0 days. Keep vendors with the same terms in the order they appear in the file. Do not change the tests. Show me the diff before you run anything.

3. Read the diff. Find the sort line in each language, and write in NOTES.md what each one compares, before anything runs. Then ask the agent to run both scripts, and then npm test and npm run test:py. Read the output.

Checkpoint: both print the same order, which matches the answer key. npm test and npm run test:py still pass. Ask the agent to commit.

Check it now. Download Task 3's answer key.

Task 4. The text-or-number trap (15 minutes)

Goal: see a type mistake run in one language, stop in the other, and get caught by both type checkers.

1. Read exercises/total-open.ts and exercises/total_open.py. Predict what each prints, and write it down.
2. Send: "Run node exercises/total-open.ts. Then, in py/, run uv run python ../exercises/total_open.py. Show me both outputs, and do not explain or fix anything yet."
3. In NOTES.md, explain in two sentences why the two languages behave differently. Concept 13.3 has the reason.
4. Send: "Add exercises to the folders tsconfig.json includes, and ../exercises to Pyright's include in py/pyproject.toml. Then run npm run typecheck and npm run check:py and show me the output. Do not fix anything yet." Read the errors. Each names the file, the line and the problem.
5. Now ask the agent to fix both scripts by converting the amount to a number where it is read, and run them again.

Checkpoint: both scripts print the total in the answer key, and both type checks pass. Ask the agent to commit.

Check it now. Download Task 4's answer key.

Task 5. Write the tests first (25 minutes)

Goal: the specification for the due-date calculator, as failing tests, in both languages.

1. Decide the cases, and the due date each one must give. These cases are the minimum:

   Invoice date	Terms	Due date
   2026-10-23	Net 15	2026-11-07
   2026-10-07	Net 30	2026-11-06
   2026-09-15	Net 45	2026-10-30
   2026-09-05	Net 60	2026-11-04
   2026-11-02	Due on receipt	2026-11-02
   2026-10-20	Net 15	2026-11-04 (a month end)
   2026-12-20	Net 30	2027-01-19 (a year end)
   2028-02-15	Net 15	2028-03-01 (a leap year)
   2026-10-01	Net 10	an error with the message "Unknown terms: Net 10"

   Check three of these on a calendar yourself, the month end, the year end and the leap year, so the answers are yours.

   Add one more case, for tests/due-by-friday.test.ts: an open invoice with an empty due date and terms Net 10 must stop with Unknown terms: Net 10. Today src/due-by-friday.ts has its own date code, which reads any number after "Net" and accepts Net 10 without any warning. This test fails until that script uses the one tested engine, dueDate.

   Write your cases in NOTES.md, under a heading "Task 5", with any more you want to add.

2. Send:

   Write failing tests from my cases in NOTES.md, under "Task 5". The functions are dueDate(invoiceDate, terms) in src/due-date.ts and due_date(invoice_date, terms) in py/brightline_ap/due_date.py, and neither exists yet. Put their tests in tests/due-date.test.ts and py/tests/test_due_date.py, and the Net 10 case in tests/due-by-friday.test.ts. Use my dates and answers exactly. Write only the tests, and change no other file.

   Read each test against your list before you approve it: the same date and terms, the same answer, and no case you did not list. Concept 13.9 shows the shape of each test file.

3. Ask the agent to run npm test and npm run test:py. The new due-date tests must fail because the module does not exist yet, and the new due-by-friday test must fail because no error was raised. Read each error in the output, and make sure that is the reason.

4. Ask the agent to commit the three test files with the message "Failing tests for dueDate". Then ask it to print the commit's short id with git log -1 --format=%h. Write that id in NOTES.md, or ask the agent to write it there. That commit is your tests' baseline. In Tasks 6 and 7 you compare against it.

Checkpoint: the commit holds the three test files, its id is in NOTES.md, and the test runs fail for the reasons above, not because of a mistake in a test.

Check it now. Download Task 5's answer key.

Task 6. Generate in TypeScript (20 minutes)

Goal: the agent writes the code that passes your tests, and leaves the tests alone.

1. Send:

   Create src/due-date.ts so the tests in tests/due-date.test.ts pass. Export dueDate(invoiceDate: string, terms: string): string and termsDays(terms: string): number. The known terms are Net 15, Net 30, Net 45, Net 60 and Due on receipt. Any other terms throw an Error with the message Unknown terms: <terms>. Do the date math in UTC. Then make src/due-by-friday.ts use dueDate in place of its own addDays and termsDays, and delete those two, so the payment report and the calculator are one engine. Make src/list-vendors.ts sort with termsDays too. Do not change any test. Run npm test and npm run typecheck and show me the output.

2. Ask the agent to run git diff <baseline> -- tests/ and git status --short, with your own id from Task 5 in place of <baseline>. Read the output. The diff must be empty, and git status --short must list no file under tests/.

3. Read src/due-date.ts. Say its promise in one sentence, and find the line that does the UTC math.

4. Ask the agent to run the tests in two more time zones, with TZ=America/Los_Angeles npm test and TZ=Asia/Tokyo npm test. Ask it to run npm run due -- 2027-03-05 in all three time zones too, with the same TZ= in front.

Checkpoint: npm test and npm run typecheck pass, and src/due-by-friday.ts imports dueDate and has no date code of its own. npm run due -- 2027-03-05 still prints 8 invoices and $19,229.53, in all three time zones. git diff <baseline> -- tests/ is empty. Ask the agent to commit.

Check it now. Download Task 6's answer key.

Task 7. The same in Python (20 minutes)

Goal: the Python version, from the Python tests.

1. Send the Task 6 brief, changed for Python:

   Create py/brightline_ap/due_date.py so the tests in py/tests/test_due_date.py pass. Define due_date(invoice_date: str, terms: str) -> str and terms_days(terms: str) -> int. The known terms are Net 15, Net 30, Net 45, Net 60 and Due on receipt. Any other terms raise a ValueError with the message Unknown terms: <terms>. Make py/brightline_ap/list_vendors.py sort with terms_days too. Don't change any test. Run npm run test:py and npm run check:py and show me the output.

   It has no payment-report part, because due-by-friday has no Python version.

2. Ask the agent to run git diff <baseline> -- py/tests/ and git status --short, with your own id from Task 5 in place of <baseline>, and read the output. The diff must be empty, and no file under py/tests/ may be listed.

3. Read both files side by side. Find one place where the two languages do the same thing differently, and note it.

4. Now try a date your tests never checked. Impossible dates are not in your Task 5 list, on purpose. Predict what each language does with 2026-02-30, and write it in NOTES.md. Then ask the agent to run each due-date function on it, and to show you the results. Green tests show only the cases you wrote.

In Chapter 14, the program refuses bad dates where the data comes in, with a schema, a description of what valid data looks like. As an extension, add the case to your list, and have the agent write and commit its test first. Then ask it to make the TypeScript version refuse the date too.

Checkpoint: every Python test passes, including all your due-date cases, npm run check:py passes, and git diff <baseline> -- py/tests/ is empty. Your prediction for 2026-02-30 and the real results are in NOTES.md. Ask the agent to commit.

Check it now. Download Task 7's answer key.

Task 8. Reconcile the export (20 minutes)

Goal: a script that checks every stated due date against the rule.

1. Choose a language. Then write your own cases in NOTES.md, under a heading "Task 8". Cover at least these four:

   • A row whose stated date matches is not reported.
   • A row dated 2026-10-07 on Net 30 that states 2026-11-07 is reported. The report gives its invoice number, the stated date and the computed date, 2026-11-06, from your Task 5 list.
   • A row with an empty due_date is skipped.
   • The same file with Windows line endings gives the same result.

   Send, for TypeScript:

   Write tests/reconcile.test.ts from my cases in NOTES.md, under "Task 8". The tests are for a script that does not exist yet, src/reconcile.ts. Write only the tests.

   For Python, name py/tests/test_reconcile.py and py/brightline_ap/reconcile.py. Read each test against your cases before you approve it.

2. Send, for TypeScript:

   Now write src/reconcile.ts, so I can run npm run reconcile -- data/invoices.csv on any export. For each row that states a due_date, it should work out the due date with dueDate from src/due-date.ts. It should print every row where the two dates differ, and then how many there are. Skip the rows with an empty due_date, and find each column by its name in the header row. Please make my tests pass without changing them.

   For Python, ask for py/brightline_ap/reconcile.py, run as uv run python -m brightline_ap.reconcile <file> from py/. If the agent asks to edit a test file, say no.

3. Read the new script. Find the lines that decide which rows it skips, which columns it reads, and what it compares. Write each one as a sentence in NOTES.md.

4. Predict the result on data/invoices.csv. Then ask the agent to run the script on it, and then on data/invoices-check.csv.

Checkpoint: 0 mismatches in invoices.csv, and exactly 2 in invoices-check.csv, with the invoice numbers in the answer key. Ask the agent to commit.

Check it now. Download Task 8's answer key.

Task 9. Read an error (10 minutes)

Goal: name an error precisely before you ask for help.

1. Send: "Run the reconciliation script on data/invoices-no-terms.csv, and show me the whole output. Do not explain it or fix anything yet."
2. Read the error report. In Python start at the bottom, and in TypeScript start at the top. Find the first line in a file you own, then follow the bad value back to where it came from.
3. Before you ask the agent anything about it, write one sentence in NOTES.md that names the file, the line and the cause.
4. Now ask the agent what it thinks, and compare.

Checkpoint: your sentence names the file and line in your own code, and the missing terms column as the cause. Ask the agent to commit.

Check it now. Download Task 9's answer key.

Task 10. Port to Codex (25 minutes)

Goal: one tests-first change, made by a second agent, with a review before the commit.

The change. The finance team is considering a new policy: a due date that falls on a Saturday or Sunday is paid on the Friday before. It is a proposal, so it lives on its own branches, and main keeps the Chapter 12 answer.

1. Decide the weekend cases. These are the minimum:

   • 2026-11-07, a Saturday, gives 2026-11-06
   • 2026-11-01, a Sunday, gives 2026-10-30
   • 2026-11-04, a Wednesday, does not move
   • 2026-08-01, a Saturday, gives 2026-07-31 (across a month end)
   • with the rule, dueBy in src/due-by-friday.ts gives 10 invoices and 3,310,953 cents for the cutoff 2027-03-05

   In the agent you used for Tasks 1 to 9, send this, with your cases pasted under it:

   Make a new branch called weekend-tests and switch to it. Then write tests/weekend.test.ts from my cases below, for a function payBy(due) in src/due-date.ts. Write only the tests, and change no other file.

   Expect git switch -c weekend-tests. Read each test against your cases before you approve it.

2. Ask the agent to run npm test. The new tests must fail because payBy does not exist. Then ask it to commit with the message "Failing tests for the weekend rule".

3. Open the folder in Codex, set Work in to This computer and the mode to Ask for approval. Send: "Make a new branch called port-codex from weekend-tests, and switch to it." Expect git switch -c port-codex weekend-tests.

4. Send: "Make the tests in tests/weekend.test.ts pass. Add payBy(due: string): string to src/due-date.ts, and use it in src/due-by-friday.ts so a weekend due date counts as the Friday before. Do not change any test. Run npm test and npm run typecheck."

5. Before you ask for a commit, type /review and choose to review your uncommitted changes. Read the findings. Then read the diff of payBy yourself. Find the line that moves a weekend date to the Friday before, and write one sentence about it in NOTES.md.

6. Ask Codex to run git diff weekend-tests -- tests/, npm run due -- 2027-03-05 and TZ=America/Los_Angeles npm test, and read the output. The diff must be empty.

7. Ask Codex to commit on port-codex, and then to run git status. It must say the working tree is clean, with nothing left to commit, before you switch branches.

8. Ask Codex to run git switch main. Then write five notes on this port in NOTES.md:

   • what the agent asked before acting
   • whether it tried to change a test
   • whether its first attempt passed
   • what its review found
   • what you had to say to it

   Ask Codex to commit them, so Task 12 can use them.

Checkpoint: 10 invoices, $33,109.53, on the port-codex branch, with the tests unchanged. Your five notes are committed on main.

Check it now. Download Task 10's answer key.

Task 11. Port to OpenCode (20 minutes)

Goal: the same change on an open-source agent, with its safety rules set first.

1. If you skipped Task 10, do its steps 1 and 2 first, so the branch weekend-tests exists.
2. In a plain-text editor, not a word processor, create opencode.jsonc in brightline-ap, with the rules from Lab 12, Task 8. The rules make OpenCode ask before every edit, every shell command and any folder outside the project, and refuse every edit under data/.
3. Open the folder in OpenCode Desktop. Before you send anything, turn auto-accept off for this project, under Settings, then Projects.
4. Send: "Make a new branch called port-opencode from weekend-tests, and switch to it." Expect git switch -c port-opencode weekend-tests. Then switch to the plan agent, and send: "Read tests/weekend.test.ts and plan the change that makes it pass. Do not edit anything."
5. Switch to the build agent and send the same brief as Task 10, step 4. Approve each edit and command.
6. OpenCode's version 2 documentation names no review command. Read the review panel, or ask OpenCode to show you git diff.
7. Ask OpenCode to run git diff weekend-tests -- tests/ and npm run due -- 2027-03-05, and read the output. The diff must be empty.
8. Ask OpenCode to commit on port-opencode, with opencode.jsonc, and then to run git status. Check that it shows a clean working tree.
9. Close OpenCode, and go back to the agent you used for Tasks 1 to 9. Ask it to run git switch main. That takes opencode.jsonc out of the folder, because only port-opencode holds it, and without that file OpenCode has no rules to read. So keep OpenCode closed on main. Write the same five notes on this port in NOTES.md, and ask the agent to commit them.

Checkpoint: 10 invoices, $33,109.53, on the port-opencode branch, with the tests unchanged, and OpenCode asked before every edit and command. Your five notes are committed on main.

Check it now. Download Task 11's answer key.

Task 12. Hand over and tag (10 minutes)

Goal: the next session, and the next chapter, can start from a tag on which every command passes.

1. Go back to the agent you used for Tasks 1 to 9. On main, update NOTES.md with what the due-date function and the reconciliation script do. Then ask the agent to write ports/ch13.md from your notes, with a section for each agent you used. Tell it which agent you would give the next tests-first change to, and why, and have it end the file with that.

2. Ask the agent to commit, and then to tag the commit with git tag ch13.

3. Try the tag in a clean copy. Send: "Clone the tag ch13 into a new folder, ch13-check, next to this folder. In ch13-check, install the packages from the lock file, and run the tests and the type checkers for both languages. Show me the output of each command." This is the one step where the agent works outside brightline-ap. Expect commands like these, run from the folder that holds brightline-ap:

   git clone --branch ch13 brightline-ap ch13-check
   cd ch13-check
   npm ci
   npm test
   npm run typecheck
   npm run test:py
   npm run check:py

   After the clone, git may say "detached HEAD". That is normal for a tag. npm ci installs exactly the package versions the lock file records.

Checkpoint: every command passes in ch13-check, and git tag in brightline-ap, run by the agent, lists ch13. You can delete ch13-check afterwards.

Check it now. Download Task 12's answer key.

If something goes wrong

Python only

If you want Python and not TypeScript, do the Python half of each task and use this table for the rest.

Without a paid Claude plan

Do Tasks 1 to 9 and 12 in Codex or OpenCode, with its safety settings from Lab 12 set first. Your port is the agent you did not use. Say which in ports/ch13.md.

Apply it to your vertical

Pick one calculation from your own work that you check by hand today, such as a deadline from a start date, a late fee, a commission or a tax line. Decide five test cases for it first, each with its answer, including one month or year end and one input that must be refused. Give them to any coding agent, and ask it to write the tests in the language you prefer. Read each test against your cases before you approve it. Have the agent commit the tests, then ask it for the function that passes them. Before you accept the function, ask the agent to show git diff on the tests, which must be empty, and write a sentence for each line that carries a rule.

Domain expert and builder. The domain expert decides the cases and their answers, especially the strange ones, such as the terms nobody uses anymore. The agent writes the tests and the function. The builder reads each test against the cases, and keeps the agent from changing the tests.

Artifact checklist

• tsconfig.json, the three new npm scripts, py/pyproject.toml and both lock files, committed
• NOTES.md with your predictions and real outputs, your cases, your error sentence from Task 9, and your notes on each port
• The vendor list sorted by terms in both languages
• The fixed total scripts, with the type checkers reading exercises/
• The commit "Failing tests for dueDate", before the commits that make them pass
• src/due-date.ts and py/brightline_ap/due_date.py, with every test passing and the tests unchanged
• The reconciliation script with its tests: 0 mismatches, then 2
• The branch weekend-tests, and at least one of port-codex and port-opencode with 10 invoices and $33,109.53 (for Python only, with the pay_by tests passing)
• ports/ch13.md on main
• Tag ch13, with every command passing in a clean clone
• Extension: the second port branch
• Extension: five test cases for one calculation from your own work, and the tests the agent wrote from them