docs: drop the duplicated SubstrateKind bullet and correct substrate comments

crates/libs/api/src/routes/admin/node_maintenance.rs

@@ -1,4 +1,4 @@
 use pplx_space_common_core::sql;
 use std::{collections::HashSet, time::Duration};
 
 use axum::{
@@ -189,9 +189,12 @@
     }
 }
 
-/// The substrate whose nodes this registry owns. Every registry statement
-/// scopes to it, and an update's `verify` reads only a node of it, so the
-/// registry never touches a node another substrate manages.
+/// The substrate whose nodes this registry owns. Every registry write to
+/// `nodes`/`node_direct` binds it, and an update's `verify` proceeds only for
+/// a live node of it, so registration never changes a node another substrate
+/// manages. Its reads (`fetch_node`, `fetch_nodes`) return nodes of every
+/// substrate, and the template pull and delete routes (`node_templates.rs`)
+/// act on any node `fetch_node` returns.
 const REGISTRY_SUBSTRATE: SubstrateKind = SubstrateKind::Space;
 
 fn require_admin_node_registry(

crates/libs/api/src/routes/admin/node_maintenance/tests/substrate_scope.rs

@@ -1,4 +1,5 @@
-//! The Direct node registry reads and writes only Direct (`'space'`) nodes.
+//! The Direct node registry writes, and `verify` proceeds for, only Direct
+//! (`'space'`) nodes.
 //!
 //! Each case pairs a Direct node with an otherwise identical node whose
 //! `managed_by` is `'k8s'`. Both keep a `node_direct` row, so only the

crates/libs/common/core/src/types/node.rs

@@ -804,9 +804,9 @@ impl SubstrateKind {
     ///
     /// - Kubernetes: yes. A dormant sandbox (suspended, or created and never
     ///   started) has no pod until a wake creates one.
-    /// - Space: no. Direct wakes a dormant sandbox on the access path before
-    ///   connecting (`DirectSubstrate::access_runtime_client` in the api
-    ///   crate), not after a failed control connect.
+    /// - Space: no. Direct wakes a suspended or restoring sandbox on the
+    ///   access path before connecting (`DirectSubstrate::access_runtime_client`
+    ///   in the api crate), not after a failed control connect.
     pub const fn wakes_after_failed_control_connect(self) -> bool {
         match self {
             Self::Kubernetes => true,

docs/rust/space-conventions.md

@@ -201,14 +201,6 @@ sandbox store's public `SandboxRecord.managed_by` and
 as text (`proves_no_execution` parses the latter). Convert them when touched;
 they are not a pattern to copy.
 
-- A decision about one row's substrate, its sandbox or node `managed_by`,
-  calls a `SubstrateKind` method named for the question; add one when no
-  method asks it. Rust that repeats a statement's fixed scope compares
-  against the named `SubstrateKind` const the statement binds, and mapping a
-  variant to a value (wire labels, route identity, construction) may match
-  it. A fact about the running substrate reads the api crate's
-  `capabilities()` (`SubstrateCapabilities`). Never compare the label string.
-
 - A decision about one row's substrate, its sandbox or node `managed_by`,
   calls a `SubstrateKind` method named for the question (for example
   `proves_execution_absence` or `restores_through_pods`); add one when no