More detail: Make the fix, rerun the golden set, and keep or undo it

This page goes with the Week 4 lesson Make the fix, rerun the golden set, and keep or undo it (course/workforce/week4/lessons/anw-w4-l06.md). You don't need it to finish the lesson. Read it when you want the longer version.

Which tests rerun

Every case on your team's saved list. For your email team, that's the practice emails you saved as tests. For a designed team, it's your design's test cases, every fresh job you approved as a golden case in Week 3 (the golden: lines in your team's record). Later this week, you'll add the mistakes you save. The case your team must refuse runs three times, because refusing twice in three tries isn't a pass. Your Chief runs them with its suite runner, the same one you met in "Run five cases and fix what breaks".

Why Evy grades the results

The suite runner on its own checks only whether each job finished or refused. In the Hattie B's example, a draft that finished but said the fries were vegetarian would pass that. So Evy grades what each rerun produced against your newest rule sheet, the same way Evy graded your cases, and the before and after table shows Evy's grades. You ask Evy for that in Evy's own thread: Evy takes grading requests from you, never from your Chief.

The keep rule

Keep the fix only if:

1. the case that failed now passes, and
2. nothing that passed before now fails.

A case that failed before and still fails, and isn't the one this fix was for, isn't something the fix broke: it's a different problem, which you fix later this week. If the case this fix was for still fails, the fix didn't work: undo it and ask Gigawatt for a different plan.

Splitting a fix in half

If something that passed before now fails, Gigawatt splits the changes into two halves and reruns your golden set with each half on its own. Whichever half breaks the test is split again, until the one change that broke it is found. You see what it found and decide on a smaller fix.

The change record

Two places, in your team's folder:

• The change list, CHANGELOG.md: one line per change, with where its write-up is and whether it was kept, undone, or not decided yet.
• The write-up, in changes/, named with the date and a few words: what changed, why in your words, what drove it, what else was considered, who approved it and when, the before and after for every golden case, and kept or undone.

The before and after are copied from the rerun by a script. No agent types them. Old entries are never edited: if you undo a change later, that's a new entry. Gigawatt writes changes to agents, tools and skills; your Chief writes how much a team may do on its own, and kept or undone; the Charter Steward writes roster changes. Evy, Kansa, the weekly review and the dashboard read it.

Kansa's watch list

One folder per team, work/course/staff/watch/<team>/:

• WATCH.md: the list your Chief hands Kansa with every review for that team, one line per active item.
• One file per item, numbered: what to look for, how to tell (a pass-or-fail check, like a rule), where it came from (Evy's finding with its job, or your words), an example job, when and by whom it was added, whether it's active or retired, and how often Kansa has caught it.

Only the watch-list method (aibl-kansa-watch) writes these files. Evy's findings and your corrections go on straight away, and each new item appears in your brief with a one-word undo. Undo retires the item and keeps its history. Kansa lists which items it checked in every review.