The first split decomposed instead of deadlocking. The question was whether an abrupt change in a character's behaviour needed setup on the page, and four options went out [9]. One leg picked A, the other picked C, and C is A plus B, so the only thing actually in dispute was B [10]. B required inventing a fact the project's canon did not have, and the project runs an explicit rule against speculative canon, the constraint that keeps a long series from contradicting itself twenty chapters later [11]. The author's formulation there is worth borrowing: an option that can only be executed by breaking a standing invariant is less a candidate than a bug report about the brief [25]. The leg that picked C had also handed over the condition that sank it, naming as its own falsifier the case where the coincidence is later revealed as a deliberate setup by the adults, which is the central motif of the work [12]. Three independent reasons converged on A, none of them the tally [13].
The second ruling is the countable version. It asked whether a transgression at the climax leaves a visible price on the page, again over four options [16]. One leg picked A, the other D [17]. Both rejected B and C on the same structural ground, that each reframes the event as an incomplete repayment while an earlier ruling in the project had fixed the opposite invariant, that handing the object back is a registration [18]. Four options minus two distinct picks leaves two options rejected by both legs, so a tied panel still cut the option space in half [22]. The author reads that convergence as a stronger signal than either leg's preference among the survivors [19]. What separated A from D was ground truth: D asked for a beat of the character noticing that a grep of the chapter shows is already present, twice [20].
Agreement is also only as wide as the axes you commissioned. Executing A meant opening the manuscript at the clause to be strengthened, and the clause named one location while the scene it referenced happens somewhere else [14]. The flow reviewer had read eight chapters end to end without catching that, because that axis reads who, when and why, and does not check where [15]. Eight chapters of continuous reading passed without anyone checking the address.
The technique moves to someone else's review step only if three conditions hold, and the evidence behind them is thin enough to state plainly: one practitioner, four rulings, one manuscript project [24]. The overlap arithmetic works because the brief numbers its options, so two picks can be compared as sets [c7a]; free-form review comments give you nothing to intersect. The invariant kill works because the project's standing rules are written down where a reader can check them against a proposal [11]. The tiebreak works because ground truth was a grep away [20]. Hold all three and the schema change is the cheap half of the repair [6]; hold none and a 1-1 split is just a coin flip counted as a vote [26].
