The archive decision appeared as a row in a board on your laptop and as a single tappable card on your phone. Those are two arrangements of the same recorded card. When you answered on the phone, the answer went back to the laptop's record, not into a separate phone history.
A projection is one of those arrangements: a read-only view built from the on-device record. It can group, filter and lay out what happened however a screen needs to. It never owns a copy of it.
The views can differ; the source cannot
A laptop can show every open thread at once because it has room. A phone opens on the one thing that needs you. Both are showing the archive card that the coding agent raised, so neither has to be reconciled with the other.
This is why answering from the phone is unremarkable rather than a synchronization problem. Your choice is recorded against the card in the origin machine's record. Writes go online to that machine; if it is unreachable, the surface stays read-only instead of building a local queue that would have to be merged later.
A fair manual comparison
The equivalent without ROOK is a shared issue tracker open in a browser tab on the laptop and in an app on the phone. That works, and it is a normal way to answer a question while away from your desk.
The friction is that a general tracker is a system of record in its own right, so you end up maintaining it: moving cards, closing duplicates, keeping it honest against what the code actually does. A projection has no such obligation, because it holds nothing. Delete a view, rebuild it, or write a different one, and the archive thread is untouched.
What a view can still get wrong
Sharing one source rules out one failure and leaves others open. Two screens cannot disagree about what was recorded. They can absolutely disagree about what it means.
- A view can select the wrong field and show a confident, wrong summary.
- A view can render stale data if it has not read again.
- A view can label something in a way that invites the wrong conclusion.
The same applies to the message shapes the surfaces exchange. Generated decoders make wire-schema mismatches fail loud at decoded boundaries, so a surface running against the wrong shape breaks immediately rather than silently. That covers the shape of a message. It does not cover semantic, behavioral or presentation agreement between screens.
| The record guarantees | It does not guarantee |
|---|---|
| No projection can become the authority | That a projection is complete |
| Every view derives from the same entries | That every view explains them well |
| Your answer returns to the originating card | That two screens must look or behave alike |
So a projection is replaceable by design. That is the point of it, and it is also the reason the next section can talk about work continuing on its own: the screens are not part of the loop.
Section 4 picks the story back up at the moment the archive card was raised, and asks what the rest of the project was doing while that thread waited.
Next
With the record established, continue to autonomy.