Notes why the held-send map keeps its shape

[johnnyt]

What

A dated foot Note on ADR-0069, "why the held-send map keeps its shape". The Amendment of 2026-10-09 gives Statifier.Replay's use of the held-send map as the reason a replace leaves the map as it is. The Note adds the stronger reason: held_sends is a typed optional key of the plan context Statifier.Session.Effects.plan/2 takes (t:Statifier.Session.Effects.context/0, its value t:Statifier.Session.Effects.held_sends/0), so recording the registration in the map's values would change a type a host that plans through plan/2 codes against.

The Note decides nothing and changes no answer. It says that routing a held send's cancel to the registration recorded when the send was handed over would need only a new optional plan-context key, and that this routing is not decided here: it is not taken this round (decided by the conductor under a standing consent, 2026-10-10: the Note only).

Direction check of the record text

Every claim was read on main at c2c69ef7 (an ancestor of origin/main) and re-read at this branch's base 02e729f3, which touches none of the cited files. The Amendment sentence the Note quotes is in its paragraph on <cancel>, verbatim. Statifier.Replay's plan_context/1 passes its own held_sends and it is kept by Statifier.Session.Effects.register_held_send/3. t:Statifier.Session.Effects.context/0 declares optional(:held_sends) => held_sends(), and t:Statifier.Session.Effects.held_sends/0 is send_id => [type] with the types sorted and unique. docs/extending.md, section "The <send> half: host-registered send types", tells a host planning through plan/2 that the context carries held_sends. The cancel arm of Statifier.Session.Effects.plan/2 looks each held type's processor up in the context's send_processors when the cancel is planned. git diff origin/main -- docs/adr/ shows zero removed lines. The Note sits after the last Note (2026-10-11, "two cases the replace call's answers leave out") and has no Status line.

Gate

Docs and record only; no changelog fragment (changelog.d/README.md: none for ADRs). mix quality --profile merge ran green on the rebased head (the ADR judge skipped: no file in a judged scope).