The paned collision view is the forge's half of overlap arbitration and it is not built

When two replicas concurrently rewrite the same region, B2 arbitration now puts one author's whole version in the file and buries the other's, with a `Yielded` flag naming the group and what prevailed. That is half of what was adopted. The other half is option D: the file's contended span as one pane and the yielded version as the other, both reached from the flag, in the forge's review surface. The options note was explicit that these are not alternatives: "D is not an alternative to A or B, it is the other half of whichever of them is chosen" (`overlap_render.md` §3.4). It also gave the reason D over B is cheap where D over the old interleave was not: with B, one pane is the file, so the view costs one counterfactual render rather than two, and the reader holds two texts in mind rather than three. The engine work D needs is done. `tree::at(log, &[frontier])` renders any causally closed subset and writes nothing; `Repo::placement()` and `render::Placement` give file-relative coordinates; the yielded operation's own bytes are `Atoms::slice` and need no counterfactual render at all. The cheapest half is already shipped on the tool's side as option E, the `.ore/collisions/<op-id>` spill, which gives a reviewer with no forge at all a `diff`. The as-built section says where it stands: "Option D is not [in], and is unchanged work for the forge" (`overlap_render.md` §10), and the self-hosting trial's own as-built repeats it -- of the six gaps the trial's verdict listed, five are closed and the sixth "is not closed, and not the tool's ... the paned view remains the Oregami Repository tier's". The stake is larger than one view. The trial's verdict on retiring the one-writer-per-repository rule was that it is retirable for durability and convergence now, and unconditionally retirable once the review surface grows a collision view and a reviewed state. The reviewed state shipped. This is the last condition. Transcribed from the project record on 2026-08-14. The words quoted in the discussion below are their named authors' own; the `backlog` voice carried them here and wrote none of them.