This is my review of the paper. 1. Is the historical trace accurate? Yes, but there are two things I would flag, one overstatement and one factual slip. The overstatement. Kohlhoff's own model treats `post` and `defer` differently. In P0113R0, the distinguishing property of `defer`, the one that carries the "continuation" semantics, is explicit: `defer` signals that this scheduling is a continuation of the current call context; `post` does not make that claim. P4286 treats `dispatch`/`post`/`defer` as a uniform "continuation-scheduling" trio, and that seems wrong. Only `defer` maps cleanly onto the continuation framing in Kohlhoff's own terminology. The factual slip. Par. 3 says "three of the four [P1525R0 authors] co-authored P2300." All four, Niebler, Shoop, Baker, Howes, appear on P2300R10 per the paper's own reference [10]. Minor, but it is evidence the paper uses (that the P1525R0 authors were participants in the executor unification), and it is wrong against the paper's own reference list. 2. Did knowing AI was used affect my read? Yes. I must say that the paper reads like a coherent argument by one author, but knowing AI was involved made me verify direct quotes at word-level. 3. Is the structural claim sound? Mostly. But there's one thing which is not clear to me: the par. 6 table row "Error channel" for P4003R3 says "None after acceptance. `post` throws before." That is a genuine deviation from the pattern. `post` throwing is exactly the "throw-before-submission" behavior that P1525R0 par. 2.2 explicitly criticized (the paper argues that generic code must be ready for `execute` to throw). It is not obvious why the same behavior is a defect in P0443 but acceptable in P4003R3. The paper does not defend the asymmetry; it presents "post throws before" as if it is a footnote. I would guess it is "the throw happens before ownership transfer, so the caller is still alive and has a stack to unwind on", but a reader has to reconstruct that themselves. 4. What is missing, and where is the argument weakest? a. The elided qualifier from P3941R4. Par. 4 quotes P3941R4 as ending "...a scheduler used with `affine_on` is infallible." The sentence in P3941R4 actually continues: "...is infallible, at least when used appropriately (i.e., when providing a receiver whose associated stop token is an `unstoppable_token`)." The quotation stops before the qualifier, and that qualifier opens a counter-reading the paper does not touch. If `set_stopped` is only forbidden when the receiver's stop-token is `unstoppable_token`, then the infallibility requirement is not a property of the continuation-scheduling role; it is a property of "the caller has opted out of cancellation for this operation." Under that reading, the symmetry with P0443R14 (where the caller had no cancellation opt-in mechanism at all) becomes weaker: P0443R14 excludes cancellation universally; P3941R4 excludes it when the caller asks for it to be excluded. Those are different design decisions. The paper should engage with this reading and either concede it or refute it. b. The diagnosis-to-return asymmetry between par. 2 and par. 4. Par. 2 tracks only "the absence of an error channel". Par. 4 tracks both `set_error` and `set_stopped`. If the symmetry is real, cancellation belongs on both sides, and P1525R0 par. 3 gives the material for cancellation-as-diagnosis. Adding cancellation to par. 2 would actually strengthen the argument (P1525R0 flagged four deficiencies; P3941R4 addresses at least two of them the same way). Right now the paper argues a one-dimensional diagnosis meets a two-dimensional return, and the mismatch is natural. -- Gennaro Prota <https://prota.dev>