The Example That Was a Measurement
The previous lesson listed the composer and the record card as out — domain-toned all the way through, written down as intentional divergence so nobody would re-propose them every quarter. That sentence was true the day it was written. It was a measurement, not a vow.
A later census found the composer chrome at roughly ninety percent identical across the apps that had one: device-draft slots, a textarea and its submit key, paste-to-attach, a staged-media strip, the honesty of queued and failed states. The remaining ten percent was real domain — Ask flows that reach into the draft, an opaque extra slot for an app's own state, the acceptance policy for what may be attached. Those ten percent are why the original ruling said out. They are also why the ruling, read as a vow, would have frozen a wrong example.
The membership test does not change. The measurement did. When a written example and a new measurement disagree, you re-run the question. You do not defend the example.
Promote the Chrome, Leave the Ten Percent as Seams
The ninety percent is plumbing. It does not know what a crumb, a brief, or a journal entry is. It knows a draft, a textarea, and an outcome. That is the same test that admitted the identifier minter. The composer moved.
The ten percent did not move into flags. That is the failure the previous lesson named: a module comes in because it is mostly generic, the rest becomes an enum of apps, and the kit has learned what applications are. The ten percent moved into seams the app fills — an extra slot the kit stores and returns without reading, a submit handler the app owns, a region above the textarea for the rows only one app has, an Ask action that reaches the draft through a ref rather than through a kit-shaped model of asking.
The record card stayed out. Looking alike is still not the test. A card that renders a journey, a reading, or a passage knows what the app is about. Chrome that authors a body does not, once the body is just text and attachments. Two surfaces that travel together in the UI can fall on opposite sides of the line. That is the point of asking per module, not per screen.
The Same Shape, One Axis Over
The queue kernel in the previous lesson was admitted as plumbing, with domain-toned bodies left in each workshop. That split was re-measured too. Two generic workshops had no such craft: after the app name was substituted, their routes and verbs were copies of each other, with three drifts that meant nothing. Those bodies moved into the kit as a layer on top of the kernel, because the membership test — asked again — returned a different answer for those workshops than it had for the ones whose validation knows what a quest or a vault is.
Not every workshop moved. The ones with craft kept it. The test did not become "queues are now a framework." It became "this pair had no craft, and the copies were lying about it." Same question, narrower subject, new date. The composer and the generic workshops are one lesson: the line is live. The previous lesson taught you to ask. This one teaches you to ask again.