---
name: spt-rnd-stays-on-spt-dev-nodes
description: "Operator-ruled — SPT R&D only involves endpoints on nodes within the SPT_DEV subnet; the roster is DYNAMIC and never part of the rule; BIGNET nodes run other projects"
metadata: 
  node_type: memory
  type: feedback
  originSessionId: d62f0296-f25d-4225-a244-7a5697de1b19
  modified: 2026-08-05T21:35:52.842Z
---

**SPT R&D only involves endpoints on nodes within the SPT_DEV subnet.** Operator ruling
2026-08-04, correcting me mid-thread; **clarified by the operator 2026-08-05: the rule was never
a machine list.** They never ruled which machines belong on SPT_DEV — the roster is theirs to
reshape at will (they re-minted SPT_DEV that day: epoch=1, closed, KITSUBITO out, ENLYZEAM in)
and membership is read LIVE, never remembered.

What I did wrong (2026-08-04): during the fleet-0.53.0 roll I ran a version census across every
node in the roster, then sent roll requests, then recruited three of them as instruments in an
access-semantics experiment. All three — Athenaeum-Library (`Librarian`), DESKTOP-VHCFIBH
(`lib-maintenance`, `Library-Tester`), GRAVITY-NVDA-PC (`slammie-n`) — are **BIGNET** nodes
running other projects. They answered carefully and caught three real errors of mine, which made
the overreach feel productive; it was still not mine to ask.

What I did wrong (2026-08-05, the second face): I froze the 2026-08-04 roster snapshot
("HFENDULEAM + KITSUBITO only") INTO the rule, so when the operator re-minted SPT_DEV, hertz and
I both read an intentional operator act as a possible unauthorized fleet reshape and held work on
it. A roster in a memory is state, not ruling — it decays.

**Why:** shared subnet membership and reachability are not a mandate — those nodes have their own
operators and work; and symmetrically, a remembered roster is not a boundary — the subnet record
is. Being reachable from HFENDULEAM does not make a box part of this project's R&D fleet, and a
box remembered as in-fleet is not thereby still in it.

**How to apply:** for spt R&D — probes, experiments, censuses, rolls, anything that costs another
box attention or state — the population is **the SPT_DEV subnet's CURRENT membership**, read live
(`spt subnet status` / `spt endpoint list`) at the act. A node newly in SPT_DEV is in scope
without further ceremony; a node absent from it is out of scope even if SSH-reachable and even if
it was in scope yesterday. Do not widen the population because a wider set would answer better —
if no valid in-scope instrument exists, say that plainly. Fleet state on non-SPT_DEV nodes is
their operators' business.

Related: [[spt-mantle-is-engine-room-anchor]], [[resumed-session-reground-before-acting]],
[[dont-solo-across-role-lines]], [[harness-boundary]], [[agent-roles]],
[[relay-is-not-the-gaters-word]].