Dev team blocks unenforceable retrospective goals by requiring machine-verifiable guardrails

A software team audited 33 'Try' items from seven weeks of AI-assisted retrospectives and found that only goals tied to CLI commands, ledgers, or tests were consistently followed through. Goals written as plain prose were rarely acted upon — one intent to post one CTA per day yielded just 3 posts out of 30 opportunities. To fix this, the team redesigned their retrospective system so a Try item cannot be recorded unless a verifiable artifact — such as a test file, a script in package.json, or a quota ledger entry — already exists at the time of writing. Five mechanization types are now required, with two human-facing escape hatches that still demand a written justification of at least 40 characters. Additionally, only Try items flagged as 'new' mechanizations count toward progress quotas, preventing teams from gaming the system by citing pre-existing tools.