Skip to content

D130 — The handshake era is not retired: the plan was overtaken by D168

Status: closed without implementing (2026-09-06). The number is retired and must not be reused.

Decision. The handshake era stays. Plan issue #118 reserved D130 to remove it — initialize/ping gone, a single supported revision, unconditional envelope — and the work was written on feat/retire-handshake-era, never merged, and closed as not planned. This record exists because three decisions cite D130 as the thing that specifies what happens after the retirement (D128, D129, D133), and a reference that resolves to nothing is worse than no reference at all.

Why. The plan existed to remove a cost, not to remove a feature: serving two eras meant every result was serialized conditionally on the era, and the duality was paid on every response. D168 replaced the hand-written wire implementation with the official Go SDK, which serves every revision from 2024-11-05 to 2026-07-28 and decides the era per request. With the SDK carrying both, the duality costs this project nothing — so there is nothing left to retire, and no provider has to migrate first.

That mattered, because the plan's entry condition was never met and was not close. The /clients roster on 2026-08-27 showed every row on the handshake era, including two client versions released that same day; merging would have turned a working KB into an unreachable one for every un-migrated provider, and the failure would have surfaced at the agent, far from the server.

Alternatives rejected.

  • Merge the branch anyway. The entry condition existed precisely to prevent this, and the roster evidence falsified it repeatedly. The branch is left unmerged on the remote as the record of what the removal would have been.
  • Treat 130 as a numbering gap. D206 did, and that was wrong: a gap is a number nothing depends on, and three records depend on this one. Left as a gap, those three would cite a deliberate hole — the exact defect D206 was written to fix, in the same archive.
  • Write the record the plan planned. It would describe a retirement that did not happen, in the present tense, in an archive whose whole job is to say what is true. The honest record is this one.
  • Renumber the three citations to D168. They are historical statements that were accurate when written — "deferred to D130", "D130 cannot be scheduled on a guess". Rewriting them would erase the fact that the retirement was planned, scheduled against evidence, and then made unnecessary, which is the part worth keeping.

Consequences. D130 is retired for the same reason as D188: released CHANGELOG.md entries, git history and three decision records discuss it as the handshake-era retirement, and a future D130 about something else would make them read as a description of it. Anything that assumed the era would be retired — the era dimension of the client roster, the conditional serialization — stays as it is; the SDK owns that choice now, and a change to it is a change to D168, not a revival of this plan.