Week 11 Shift: Bot audit
You audit ShiftBot against four weeks of its own decisions and find out what the rules you wrote in week 7 have actually been doing
17 May 2027
Four weeks of automated decisions, made on your behalf, by a thing that cannot be asked what it meant.
Before the Shift#
Your Contribution Log closes at noon today. Weeks 2 to 11, submitted.
Export ShiftBot’s decision history and the week 7 rationale register. The 2026 cohort’s audit and its amendment are published, including the false positive they traded away.
In the Shift#
Take each rule and put it against what it did. Volume, hit rate, and — the part that matters — the removals nobody appealed, because those are the ones the system got away with rather than got right.
Find a false positive and trace it back to the exact word or threshold in the rule that caused it. This is usually humbling, because the rule will be one somebody in the room argued hard for in week 7.
Then decide: keep, amend, retire. Every decision goes in the register with a reason, because next week you hand this to strangers.
Last half hour, the harder question: generated submissions. The sandbox has been receiving them. Assessment is your scarce resource and it always was, so the cohort’s position on machine contribution has to be a position on who does the checking.
Afterwards#
The amended config ships. Your Governance Redesign is due in week 12 and now has evidence behind it.
The spec
What you bring to the Shift. Written so a reader can tell whether it has been met without asking you.
- every ShiftBot rule is checked against its actual decisions since week 7
- you have found at least one false positive and traced it to the rule's text
- each rule is kept, amended or retired, with a reason recorded in the register
- your Governance Redesign states a position on machine contribution