1.6.0
Runningc2396adFixed
Approve now moves a task to Staged, not straight to Done — the same place the
[APPROVED]marker has always sent it. Until now the two ways of accepting work disagreed: posting the marker reached Staged, clicking the button skipped it, and anything approved by button lost the distinction between accepted and actually released. The button predates Staged existing, and PROEN-46 updated only the marker path, so this shipped alongside Staged rather than breaking afterwards. Done is still reachable from the status dropdown (PROEN-61).The e2e suite no longer asserts before a page has its data. Every surface that loads client-side now publishes a
data-page-loadingattribute from the same flag that gates its skeleton, and the tests wait on that instead of on hydration — which only ever proved the framework was interactive, not that the content had arrived. This was CI going red about half the time on a different test each run, with no application defect behind it (PROEN-55, absorbing PROEN-36).
Nothing in this release changes the /api/v1 contract, so no client needs to do anything.
One known issue is deliberately not fixed here and is worth knowing about if CI goes red. This entry originally described it as the e2e dev server accepting a request and never answering it. That was wrong, and PROEN-62 has since established what actually happens: in development Nuxt compiles a route on its first request, inside that request, which costs 8-13 seconds. Nothing hangs — under load, 2400 requests produced four slow ones and every one returned a 200. The "never answered" reading came from Playwright recording a request it gave up on and a request the browser cancelled in exactly the same way, which is indistinguishable from one that was never answered. The harness now compiles those routes before any test is timed.
The two things this entry said that were true remain true: PROEN-55 closed with four consecutive green CI runs rather than the five its own acceptance criteria asked for, and this affects the test harness rather than production traffic. The second is now measured rather than assumed — the same load against a production build peaked at 1.4 seconds against development's 12.7, with nothing left outstanding, because a production build compiles its routes ahead of time.