Part 6 Quiz
Two questions ahead, Step 14 first and Step 15 second. Settle on your own answer before opening either reveal — both of these reward a specific answer over a general one.
1. The book club's reading log
An eight-member book club has met monthly for three years and wants to start tracking, properly this time, which member has actually read which of the group's picks — right now it lives in someone's memory and the occasional "wait, did you finish it?" in the group chat. The list grows by exactly one book a month, membership hasn't changed in over a year, and nobody has ever needed to prove when a member finished a book, only whether they did. Run this through the six pre-build questions. Which one ends it, and what should the club build instead of a graph?
Reveal the answer
It ends on the question about whether the relationship set is small, fixed, and table-shaped, or genuinely growing in a way a table can't hold. Eight members by however many books the club has read is exactly a grid — one row per member, one column per book, a checkmark in each cell for "read it" — and adding one column a month for years doesn't change that shape, it just makes the grid longer. Nothing here needs a schema, an extraction step, or a pre-build checklist entry about provenance, because nobody has ever needed to trace a "read" mark back to a specific message — a shared spreadsheet with one tab already does everything this group actually wants. The other five questions might have looked fine too; a single "no" on the table question is enough to close the case.
2. Two configurations, one silent disagreement
Two teammates each build a version of a small lending-tracker pipeline from the same five source messages, independently, without comparing notes first. One resolves two different names in the messages into a single person, on the reasoning that they clearly refer to the same borrower; the other, moving faster, leaves the two names as two separate people because nothing in its instructions explicitly required a resolution step. Neither configuration errors out — both run cleanly and both look, read side by side, like reasonable ways to build the same pipeline. What does comparing the two final graphs reveal that comparing the two instructions would not?
Reveal the answer
Reading the two prompts side by side wouldn't catch this, because neither one is wrong on its face — omitting an explicit resolution step isn't an error, it's a gap that only shows up in what gets produced. Diff the final graphs and it surfaces immediately: one version has a single person-node with both a borrow and a return edge attached, so the item correctly reads as available; the other has two separate person-nodes, one holding the borrow edge and the other holding nothing, so the item incorrectly reads as still out, because the edge that would close the loan is sitting on a node the checker never associates with the borrower. Dual-tool parity is a claim about output, and the only way to check a claim about output is to compare the output — two configurations that read as equally reasonable can still diverge the moment either one makes a silent judgment call the other didn't make the same way.