fix: validate resolved ancestry inside atomic projection dispatch

Resolve one connected lineage and retain its root-to-parent capabilities. Backends validate ancestor identity/membership and refresh all policy metadata inside the same boundary as target validators and typed dispatch; hidden/replaced ancestors reject as NotFound without invocation or effects. Preserve pinned leaf and configured-root capabilities. Covers MR ANCESTOR_VISIBILITY_TOCTOU with HTTP memory and Linux replacement/budget/guard regressions; all T-19 workspace acceptance commands pass (Clippy via Nix shell).

crates/wip-fs-server/README.md

@@ -23,8 +23,9 @@ For an external backend/composition:
 
 1. Implement `BackendEntry` on a local, cloneable, root-confined stable capability.
    Its `WipObject` metadata is public-only; its validator binds identity and the
-   protected state. Implement `FilesystemBackend` to resolve capabilities,
-   enumerate bounded children, and execute against an already-resolved entry.
+   protected state. Implement `FilesystemBackend` to resolve one connected
+   `ResolvedEntry<Entry>` (root-to-parent ancestry plus pinned target), enumerate
+   bounded children, and execute against that already-resolved lineage.
 2. Register generated `#[wip_interface::interface]` definitions using the existing
    `wip_host::InterfaceRegistry`. `DefaultInterfaces::register` optionally supplies
    reusable built-in tokens; using those tokens does not force their publication.
@@ -55,10 +56,26 @@ listener, a global object registry, or dynamic plugin loading.
 ### Extension security contract
 
 Backend implementations and Operation authors are trusted code, not sandboxed
-plugins. `FilesystemBackend::execute` must refresh state/validators **inside** its
-transaction or lock before calling the engine callback, retain the same resolved
-target through validation and dispatch, and return the committed post-operation
-validator. Confirm rollback only after rollback succeeds. Pre-dispatch failures
+plugins. `FilesystemBackend::resolve` returns `ResolvedEntry::new(ancestors, target)`
+from one connected backend traversal/snapshot, not independent path-prefix
+resolutions. A root target has no ancestors; other targets retain the root first,
+then intermediate parent containers. This is backend-private context, never wire
+metadata. Backend `resolve` and `execute` signatures both carry this type;
+compositions still receive only the exact scoped target `Entry`.
+
+`FilesystemBackend::execute` must validate ancestor membership/identity and
+refresh **all** lineage state/validators inside its transaction or lock before
+calling the engine callback, retaining the same target and connected ancestry
+through validation and dispatch. Replaced/detached ancestors must return
+`NotFound` without callback invocation, even when the replacement is visible.
+The engine evaluates Object visibility for every refreshed ancestor and target
+inside that same boundary; a hidden ancestor rejects the entire subtree before
+publication/typed dispatch. Never fix a changed lineage by resolving the raw path
+again and substituting capabilities. Participating policy-relevant backend state
+changes must be serialized by this boundary, not just target-content writes.
+Return the committed post-operation validator. Retained ancestry is subject to
+handle/work ceilings; resource exhaustion is an error, not missing ancestry.
+Confirm rollback only after rollback succeeds. Pre-dispatch failures
 must never commit; failed started operations without a non-commit guarantee remain
 `OperationOutcomeUnknown`. Object implementations must delegate validators to
 that entry and declare requirements per `(InterfaceRef, operation)`, rather than
@@ -68,13 +85,16 @@ cancellation and resource budgets throughout backend work and report effects
 truthfully. POSIX symlinks, identity, atomic mutations and locking cannot be assumed
 for other backends.
 
-The default Linux backend uses advisory exclusive locks on regular files,
-refreshes validators after acquiring the lock and holds it through post-dispatch
-state observation. This serializes cooperating library calls; it does **not**
-promise transactional isolation from unrelated processes ignoring `flock`.
-Descriptor-relative pinned handles still prevent replacement from redirecting an
-operation to a different file. Linux capabilities only expose bounded methods,
-not raw descriptors or backend paths.
+The default Linux backend acquires advisory exclusive locks in root-to-parent
+order and then on the lockable target (directory or regular file). It compares
+retained ancestor identity to no-follow descriptor-relative lookup results,
+refreshes lineage snapshots after acquiring the locks, and holds the guards
+through post-dispatch state observation. The configured root remains its pinned
+capability even if its textual path is renamed; replaced intermediate ancestors
+are rejected. Leaf path replacement never redirects execution to a new inode.
+This serializes cooperating library calls; it does **not** promise transactional
+isolation from unrelated processes ignoring `flock`. Linux capabilities only
+expose bounded methods, not raw descriptors or backend paths.
 
 A policy must be stable/deterministic within one effective Worldspace view.
 Interface/operation visibility is uniform per registered reference within that

crates/wip-fs-server/src/engine.rs

@@ -32,6 +32,39 @@ pub trait BackendEntry: WipObject + Clone {
     fn kind(&self) -> EntryKind;
 }
 
+/// A single resolved target together with its retained root-to-parent lineage.
+///
+/// The backend must resolve a connected lineage, never independent raw-path
+/// prefixes. At execution it validates ancestor membership/identity, refreshes
+/// their policy metadata, and pins that snapshot through dispatch. Target
+/// replacement must not redirect the original capability. This value is private
+/// backend context, not protocol metadata.
+#[derive(Clone)]
+pub struct ResolvedEntry<E> {
+    ancestors: Vec<E>,
+    target: E,
+}
+
+impl<E> ResolvedEntry<E> {
+    /// Create backend resolution context. `/` has no ancestors; a non-root target
+    /// has the pinned root first, followed by its intermediate parent containers.
+    pub fn new(ancestors: Vec<E>, target: E) -> Self {
+        Self { ancestors, target }
+    }
+    /// Pinned ancestor capabilities, ordered from the root to the direct parent.
+    pub fn ancestors(&self) -> &[E] {
+        &self.ancestors
+    }
+    /// The original resolved target capability, never a redirected replacement.
+    pub fn target(&self) -> &E {
+        &self.target
+    }
+    /// Consume this snapshot for refresh or request-time publication.
+    pub fn into_parts(self) -> (Vec<E>, E) {
+        (self.ancestors, self.target)
+    }
+}
+
 /// Trusted backend boundary. Implementations need not use POSIX semantics.
 ///
 /// Resolve and enumerate pinned capabilities within the configured root. Enforce
@@ -39,17 +72,28 @@ pub trait BackendEntry: WipObject + Clone {
 /// advertise capabilities the backend cannot guarantee. Errors must not include
 /// backend paths, credentials or hidden metadata.
 ///
-/// `execute` must retain the same target and serialize validator observation,
-/// publication, validation, typed dispatch and commit/rollback. Refresh entry
-/// validators *inside* the boundary before invoking the callback, and return the
-/// committed post-operation validator on success. Never upgrade an unknown
-/// outcome except after confirmed rollback. Locks only protect against actors
-/// participating in that locking protocol; document any external-writer limits.
+/// `execute` must retain the same target and complete lineage and serialize
+/// ancestor identity/membership checks, policy metadata, validator observation,
+/// publication, validation, typed dispatch and commit/rollback. Changed/detached
+/// ancestors must return `NotFound` before the callback; never substitute a new
+/// ancestor or redirect the target by re-resolving its raw path. Refresh *all*
+/// lineage metadata/validators inside the boundary and retain that snapshot
+/// through dispatch, then return the committed post-operation validator. Never
+/// upgrade an unknown outcome except after confirmed rollback. Locks only protect
+/// actors participating in the protocol; document any external-writer limits.
 pub trait FilesystemBackend {
     /// Stable capability supplied to the composition.
     type Entry: BackendEntry;
-    /// Resolve a canonical WIP path without escaping the configured root.
-    fn resolve(&self, context: &ExecutionContext, path: &str) -> ProviderResult<Self::Entry>;
+    /// Resolve a canonical WIP path and retain its connected ancestry.
+    ///
+    /// Obtain the target and lineage in one backend traversal/snapshot. Enforce
+    /// configured handle/work budgets while retaining ancestors; do not silently
+    /// omit ancestry on resource exhaustion.
+    fn resolve(
+        &self,
+        context: &ExecutionContext,
+        path: &str,
+    ) -> ProviderResult<ResolvedEntry<Self::Entry>>;
     /// Enumerate direct children, at most `maximum`; excess must fail, not truncate.
     fn children(
         &self,
@@ -61,8 +105,8 @@ pub trait FilesystemBackend {
     fn execute(
         &self,
         context: &ExecutionContext,
-        entry: Self::Entry,
-        operation: impl FnOnce(Self::Entry) -> OperationAttempt,
+        entry: ResolvedEntry<Self::Entry>,
+        operation: impl FnOnce(ResolvedEntry<Self::Entry>) -> OperationAttempt,
     ) -> ProviderResult<OperationAttempt>;
 }
 
@@ -205,39 +249,40 @@ where
     C: Composition<B::Entry>,
     P: ProjectionPolicy<B::Entry, C::Object>,
 {
-    fn resolve(&self, context: &ExecutionContext, path: &str) -> ProviderResult<B::Entry> {
+    fn resolve(
+        &self,
+        context: &ExecutionContext,
+        path: &str,
+    ) -> ProviderResult<ResolvedEntry<B::Entry>> {
         if let Err(error) = preflight(context) {
             return ProviderResult::Failure(error);
         }
         if wip_protocol::validate_path(path).is_err() {
             return ProviderResult::NotFound;
         }
-        // Check ancestors as well: guessing a descendant cannot bypass hidden parents.
-        let mut end = 1;
-        loop {
-            let prefix = if end == 1 { "/" } else { &path[..end] };
-            let entry = match self.projection.backend.resolve(context, prefix) {
-                ProviderResult::Found(entry) => entry,
-                ProviderResult::NotFound => return ProviderResult::NotFound,
-                ProviderResult::Failure(error) => return ProviderResult::Failure(error),
-            };
-            if !self
-                .projection
-                .policy
-                .object_visible(self.principal, &entry)
-            {
+        match self.projection.backend.resolve(context, path) {
+            ProviderResult::Found(resolved) => {
+                if !lineage_matches_path(&resolved, path) {
+                    return ProviderResult::Failure(internal());
+                }
+                if !self.lineage_visible(&resolved) {
                     return ProviderResult::NotFound;
                 }
-            if prefix == path {
-                return ProviderResult::Found(entry);
+                ProviderResult::Found(resolved)
             }
-            let start = if end == 1 { 1 } else { end + 1 };
-            end = path[start..]
-                .find('/')
-                .map_or(path.len(), |index| start + index);
+            ProviderResult::NotFound => ProviderResult::NotFound,
+            ProviderResult::Failure(error) => ProviderResult::Failure(error),
         }
     }
 
+    fn lineage_visible(&self, resolved: &ResolvedEntry<B::Entry>) -> bool {
+        resolved
+            .ancestors()
+            .iter()
+            .chain(std::iter::once(resolved.target()))
+            .all(|entry| self.projection.policy.object_visible(self.principal, entry))
+    }
+
     fn publication(
         &self,
         entry: B::Entry,
@@ -342,6 +387,7 @@ where
             }
             ProviderResult::Failure(error) => return Err(read_error(error)),
         };
+        let (_ancestors, entry) = entry.into_parts();
         let mut remaining = u64::from(self.projection.limits.max_observation_nodes)
             .min(context.resource_budget().unwrap_or(u64::MAX));
         let response = self.observe_entry(context, entry, request.depth, &mut remaining)?;
@@ -426,10 +472,13 @@ where
         operation: impl FnOnce(&PublishedObject<Self::Object>) -> R,
     ) -> ProviderResult<R> {
         match self.resolve(context, path) {
-            ProviderResult::Found(entry) => match self.publication(entry, context) {
+            ProviderResult::Found(resolved) => {
+                let (_ancestors, entry) = resolved.into_parts();
+                match self.publication(entry, context) {
                     Ok(publication) => ProviderResult::Found(operation(&publication)),
                     Err(error) => ProviderResult::Failure(error),
-            },
+                }
+            }
             ProviderResult::NotFound => ProviderResult::NotFound,
             ProviderResult::Failure(error) => ProviderResult::Failure(error),
         }
@@ -445,16 +494,16 @@ where
             ProviderResult::NotFound => return ProviderResult::NotFound,
             ProviderResult::Failure(error) => return ProviderResult::Failure(error),
         };
-        self.projection.backend.execute(context, entry, |entry| {
-            // Recheck visibility using the atomic-boundary snapshot.
-            // Failure here is still pre-dispatch, with no operation effects.
-            if !self
-                .projection
-                .policy
-                .object_visible(self.principal, &entry)
-            {
+        self.projection.backend.execute(context, entry, |resolved| {
+            // Backend validates retained lineage membership/identity and refreshes
+            // all metadata before this callback under the same boundary as dispatch.
+            if !lineage_matches_path(&resolved, path) {
+                return OperationAttempt::reject(wip_host::CallErrorKind::Internal);
+            }
+            if !self.lineage_visible(&resolved) {
                 return OperationAttempt::reject(wip_host::CallErrorKind::NotFound);
             }
+            let (_ancestors, entry) = resolved.into_parts();
             if let Err(error) = preflight(context) {
                 return OperationAttempt::reject(match error.kind() {
                     wip_host::HostErrorKind::Internal => wip_host::CallErrorKind::Internal,
@@ -544,6 +593,24 @@ where
     }
 }
 
+fn lineage_matches_path<E: BackendEntry>(resolved: &ResolvedEntry<E>, path: &str) -> bool {
+    let names: Vec<_> = if path == "/" {
+        vec![""]
+    } else {
+        std::iter::once("").chain(path[1..].split('/')).collect()
+    };
+    if resolved.ancestors().len() + 1 != names.len()
+        || resolved.target().name() != names[names.len() - 1]
+    {
+        return false;
+    }
+    resolved
+        .ancestors()
+        .iter()
+        .zip(names)
+        .all(|(entry, name)| entry.kind() == EntryKind::Directory && entry.name() == name)
+}
+
 fn read_boundary<T>(
     operation: impl FnOnce() -> Result<T, ProtocolError>,
 ) -> Result<T, ProtocolError> {

crates/wip-fs-server/src/lib.rs

@@ -58,7 +58,7 @@ use std::time::Duration;
 pub use defaults::{DefaultComposition, DefaultInterfaces, DefaultObject, default_projection};
 pub use engine::{
     BackendEntry, Composition, DefaultPolicy, EntryKind, FilesystemBackend, Projection,
-    ProjectionPolicy, ProjectionView,
+    ProjectionPolicy, ProjectionView, ResolvedEntry,
 };
 pub use http_adapter::HttpAdapter;
 pub use interfaces::{

crates/wip-fs-server/src/projection.rs

@@ -7,12 +7,12 @@ use std::sync::{Arc, Mutex, MutexGuard, TryLockError};
 
 use crate::engine::{BackendEntry, EntryKind, FilesystemBackend};
 use crate::interfaces::{FileRead, FileWrite, SymlinkInspect};
-use crate::{AccessMode, ResourceLimits};
+use crate::{AccessMode, ResolvedEntry, ResourceLimits};
 use rustix::fs::{
     CWD, FileType, FlockOperation, Mode as FsMode, OFlags, RawDir, flock, fstat, ftruncate, openat,
     readlinkat,
 };
-use rustix::io::{dup, pread, pwrite};
+use rustix::io::{pread, pwrite};
 use wip_host::{
     EffectGuarantee, ExecutionContext, HostError, OperationAttempt, ProviderResult, Validator,
     WipObject,
@@ -130,6 +130,7 @@ impl LinuxEntry {
 
     fn refresh_validator(&mut self) -> Result<(), ProjectionError> {
         let stat = fstat(self.fd()).map_err(|_| ProjectionError::Internal)?;
+        self.kind = entry_kind(stat.st_mode);
         let validator = matches!(self.kind, EntryKind::Regular | EntryKind::Symlink)
             .then(|| validator_from_stat(&stat));
         self.initial_validator = validator.clone();
@@ -180,10 +181,12 @@ enum FailureInjection {
 
 /// Linux filesystem backend using pinned, no-follow handles.
 ///
-/// Regular-file execution holds an advisory exclusive `flock` across validator
-/// refresh, dispatch, and post-operation observation. Serialization only covers
-/// actors that participate in this locking protocol; external writers can still
-/// modify the inode without taking the lock. Symlinks are inspected, not followed.
+/// Execution holds advisory exclusive `flock` guards on directories in
+/// root-to-parent order and on directory or regular-file targets across lineage
+/// validation, validator refresh, dispatch, and post-operation observation.
+/// Serialization only covers actors that participate in this locking protocol;
+/// external writers can still mutate entries without taking the locks. Symlinks
+/// are inspected, not followed.
 pub struct LinuxBackend {
     root: OwnedFd,
     access: AccessMode,
@@ -220,22 +223,28 @@ impl LinuxBackend {
         &self,
         context: &ExecutionContext,
         path: &str,
-    ) -> Result<LinuxEntry, ProjectionError> {
+    ) -> Result<ResolvedEntry<LinuxEntry>, ProjectionError> {
         check_context(context)?;
         wip_protocol::validate_path(path).map_err(|_| ProjectionError::NotFound)?;
+        let root = self.entry(
+            context,
+            String::new(),
+            reopen(&self.root, OFlags::RDONLY | OFlags::DIRECTORY)?,
+        )?;
         if path == "/" {
-            let fd = reopen(&self.root, OFlags::RDONLY | OFlags::DIRECTORY)?;
-            return self.entry(context, String::new(), fd);
+            return Ok(ResolvedEntry::new(Vec::new(), root));
         }
-        let mut parent = dup(&self.root).map_err(|_| ProjectionError::Internal)?;
+        // Retain the exact directories traversed, rather than resolving prefixes again.
+        let mut ancestors = vec![root];
         let mut segments = path[1..].split('/').peekable();
         while let Some(segment) = segments.next() {
             check_context(context)?;
             if segment.as_bytes().contains(&0) {
                 return Err(ProjectionError::NotFound);
             }
+            let parent = ancestors.last().ok_or(ProjectionError::Internal)?;
             let probe = openat(
-                &parent,
+                parent.fd(),
                 segment,
                 OFlags::PATH | OFlags::NOFOLLOW | OFlags::CLOEXEC,
                 FsMode::empty(),
@@ -243,14 +252,15 @@ impl LinuxBackend {
             .map_err(classify_lookup)?;
             let stat = fstat(&probe).map_err(|_| ProjectionError::Internal)?;
             let kind = entry_kind(stat.st_mode);
-            if segments.peek().is_some() {
-                if kind != EntryKind::Directory {
+            if segments.peek().is_some() && kind != EntryKind::Directory {
                 return Err(ProjectionError::NotFound);
             }
-                parent = reopen(&probe, OFlags::RDONLY | OFlags::DIRECTORY)?;
-            } else {
             let fd = self.reopen_final(probe, kind)?;
-                return self.entry(context, segment.to_owned(), fd);
+            let entry = self.entry(context, segment.to_owned(), fd)?;
+            if segments.peek().is_some() {
+                ancestors.push(entry);
+            } else {
+                return Ok(ResolvedEntry::new(ancestors, entry));
             }
         }
         Err(ProjectionError::NotFound)
@@ -323,6 +333,49 @@ impl LinuxBackend {
         })
     }
 
+    fn validate_ancestry(
+        &self,
+        context: &ExecutionContext,
+        ancestors: &[LinuxEntry],
+        target: &LinuxEntry,
+    ) -> Result<(), ProjectionError> {
+        check_context(context)?;
+        let root = ancestors.first().unwrap_or(target);
+        let configured = fstat(&self.root).map_err(|_| ProjectionError::Internal)?;
+        let retained = fstat(root.fd()).map_err(|_| ProjectionError::Internal)?;
+        // The configured root is a capability, not its possibly renamed textual path.
+        if entry_kind(retained.st_mode) != EntryKind::Directory
+            || retained.st_dev != configured.st_dev
+            || retained.st_ino != configured.st_ino
+        {
+            return Err(ProjectionError::NotFound);
+        }
+        for pair in ancestors.windows(2) {
+            check_context(context)?;
+            let parent = &pair[0];
+            let child = &pair[1];
+            let visible = openat(
+                parent.fd(),
+                child.name.as_str(),
+                OFlags::PATH | OFlags::NOFOLLOW | OFlags::CLOEXEC,
+                FsMode::empty(),
+            )
+            .map_err(classify_lookup)?;
+            let visible = fstat(&visible).map_err(|_| ProjectionError::Internal)?;
+            let retained = fstat(child.fd()).map_err(|_| ProjectionError::Internal)?;
+            if entry_kind(visible.st_mode) != EntryKind::Directory
+                || entry_kind(retained.st_mode) != EntryKind::Directory
+                || visible.st_dev != retained.st_dev
+                || visible.st_ino != retained.st_ino
+            {
+                return Err(ProjectionError::NotFound);
+            }
+        }
+        // Deliberately do not look up the target name: leaf replacement must not
+        // redirect execution away from its original pinned inode.
+        check_context(context)
+    }
+
     #[cfg(test)]
     fn set_injection(&self, injection: FailureInjection) {
         *self
@@ -341,7 +394,11 @@ impl BackendEntry for LinuxEntry {
 impl FilesystemBackend for LinuxBackend {
     type Entry = LinuxEntry;
 
-    fn resolve(&self, context: &ExecutionContext, path: &str) -> ProviderResult<LinuxEntry> {
+    fn resolve(
+        &self,
+        context: &ExecutionContext,
+        path: &str,
+    ) -> ProviderResult<ResolvedEntry<LinuxEntry>> {
         match self.resolve_entry(context, path) {
             Ok(entry) => ProviderResult::Found(entry),
             Err(error) => error.host(),
@@ -393,16 +450,32 @@ impl FilesystemBackend for LinuxBackend {
     fn execute(
         &self,
         context: &ExecutionContext,
-        mut entry: LinuxEntry,
-        callback: impl FnOnce(LinuxEntry) -> OperationAttempt,
+        resolved: ResolvedEntry<LinuxEntry>,
+        callback: impl FnOnce(ResolvedEntry<LinuxEntry>) -> OperationAttempt,
     ) -> ProviderResult<OperationAttempt> {
         if let Err(error) = check_context(context) {
             return error.host();
         }
-        let lock_entry = Arc::clone(&entry.inner);
-        let _lock = if entry.kind == EntryKind::Regular {
-            match lock_with_context(&lock_entry, context, self.limits.request_timeout) {
-                Ok(lock) => Some(lock),
+        let (mut ancestors, mut entry) = resolved.into_parts();
+        // Independent resolutions have independent open-file descriptions; clones
+        // also need the per-entry mutex because flock alone treats them as one owner.
+        // Always acquire directories root-to-parent, then the target (if lockable).
+        let mut lock_entries: Vec<_> = ancestors
+            .iter()
+            .map(|ancestor| Arc::clone(&ancestor.inner))
+            .collect();
+        if matches!(entry.kind, EntryKind::Directory | EntryKind::Regular) {
+            lock_entries.push(Arc::clone(&entry.inner));
+        }
+        let started = std::time::Instant::now();
+        let mut locks = Vec::with_capacity(lock_entries.len());
+        for inner in &lock_entries {
+            let remaining = self
+                .limits
+                .request_timeout
+                .saturating_sub(started.elapsed());
+            match lock_with_context(inner, context, remaining) {
+                Ok(lock) => locks.push(lock),
                 Err(ProjectionError::Resource) => {
                     return ProviderResult::Failure(HostError::resource_limit(
                         "filesystem lock deadline exceeded",
@@ -416,15 +489,22 @@ impl FilesystemBackend for LinuxBackend {
                     ));
                 }
             }
-        } else {
-            None
-        };
-        // Both snapshots must describe the state under the lock, not at resolve time.
+        }
+        if let Err(error) = self.validate_ancestry(context, &ancestors, &entry) {
+            return error.host();
+        }
+        for ancestor in &mut ancestors {
+            if ancestor.refresh_validator().is_err() {
+                return ProviderResult::Failure(internal_not_committed());
+            }
+            ancestor.context = context.clone();
+        }
+        // Both snapshots must describe the state under the locks, not at resolve time.
         if entry.refresh_validator().is_err() {
             return ProviderResult::Failure(internal_not_committed());
         }
         entry.context = context.clone();
-        let mut attempt = callback(entry.clone());
+        let mut attempt = callback(ResolvedEntry::new(ancestors, entry.clone()));
         if let OperationAttempt::Started(Ok(result)) = &mut attempt {
             if entry.refresh_validator().is_err() {
                 return ProviderResult::Failure(HostError::internal(
@@ -735,7 +815,7 @@ mod tests {
     use super::{FailureInjection, FlockOperation, LinuxBackend, LinuxEntry};
     use crate::engine::{DefaultPolicy, FilesystemBackend, Projection};
     use crate::interfaces::{FileRead, FileWrite};
-    use crate::{AccessMode, DefaultComposition, ResourceLimits};
+    use crate::{AccessMode, DefaultComposition, ResolvedEntry, ResourceLimits};
 
     type DefaultProjection = Projection<LinuxBackend, DefaultComposition, DefaultPolicy>;
 
@@ -765,7 +845,11 @@ mod tests {
         })
     }
 
-    fn resolved(backend: &LinuxBackend, context: &ExecutionContext, path: &str) -> LinuxEntry {
+    fn resolved(
+        backend: &LinuxBackend,
+        context: &ExecutionContext,
+        path: &str,
+    ) -> ResolvedEntry<LinuxEntry> {
         match backend.resolve(context, path) {
             ProviderResult::Found(entry) => entry,
             ProviderResult::NotFound => panic!("entry should resolve"),
@@ -786,8 +870,8 @@ mod tests {
         .expect("open backend");
         let context = ExecutionContext::new();
         let entry = resolved(&backend, &context, "/file");
-        let original = entry.initial_validator.clone();
-        let pinned = entry.clone();
+        let original = entry.target().initial_validator.clone();
+        let pinned = entry.target().clone();
         // Simulate a cooperating writer committing between resolution and execution.
         fs::write(&path, b"changed before lock acquisition").expect("external mutation");
         let lock_probe = fs::OpenOptions::new()
@@ -796,7 +880,8 @@ mod tests {
             .open(&path)
             .expect("open independent lock probe");
         let mut callback_validator = None;
-        let attempt = backend.execute(&context, entry, |entry| {
+        let attempt = backend.execute(&context, entry, |resolved| {
+            let entry = resolved.target();
             assert!(std::sync::Arc::ptr_eq(&entry.inner, &pinned.inner));
             assert_ne!(entry.initial_validator, original);
             assert_eq!(
@@ -824,7 +909,9 @@ mod tests {
         assert_ne!(result.validator, callback_validator);
         assert_eq!(
             result.validator,
-            resolved(&backend, &context, "/file").initial_validator,
+            resolved(&backend, &context, "/file")
+                .target()
+                .initial_validator,
         );
         assert_eq!(fs::read(&path).expect("read written file"), b"new");
         // The retained clone must not keep the operation's flock held.
@@ -846,10 +933,11 @@ mod tests {
         .expect("open backend");
         let context = ExecutionContext::new();
         let entry = resolved(&backend, &context, "/file");
-        let pinned = entry.clone();
+        let pinned = entry.target().clone();
         fs::remove_file(&path).expect("unlink pinned entry");
         fs::write(&path, b"replacement").expect("replace path");
-        let attempt = backend.execute(&context, entry, |entry| {
+        let attempt = backend.execute(&context, entry, |resolved| {
+            let entry = resolved.target();
             assert!(std::sync::Arc::ptr_eq(&entry.inner, &pinned.inner));
             entry
                 .replace(wip_interface::Bytes(b"pinned write".to_vec()))
@@ -870,6 +958,239 @@ mod tests {
         );
     }
 
+    #[test]
+    fn detached_or_replaced_ancestors_reject_before_callback_without_effects() {
+        // Exercise both the immediate parent and a higher ancestor, with a
+        // visible directory replacement, no replacement, and a symlink replacement.
+        for ancestor in ["a", "a/b"] {
+            for replacement in ["directory", "missing", "symlink"] {
+                let temporary = tempdir().expect("temporary root");
+                fs::create_dir_all(temporary.path().join("a/b")).expect("create lineage");
+                fs::write(temporary.path().join("a/b/file"), b"original").expect("write original");
+                let backend = LinuxBackend::open(
+                    temporary.path(),
+                    AccessMode::ReadWrite,
+                    ResourceLimits::default(),
+                )
+                .expect("open backend");
+                let context = ExecutionContext::new();
+                let entry = resolved(&backend, &context, "/a/b/file");
+                let pinned = entry.target().clone();
+                let detached = temporary.path().join("detached");
+                let ancestor_path = temporary.path().join(ancestor);
+                fs::rename(&ancestor_path, &detached).expect("detach ancestor");
+                if replacement == "directory" {
+                    fs::create_dir(&ancestor_path).expect("replace directory");
+                } else if replacement == "symlink" {
+                    std::os::unix::fs::symlink(&detached, &ancestor_path)
+                        .expect("replace ancestor with symlink to original");
+                }
+                let visible_leaf =
+                    ancestor_path.join(if ancestor == "a" { "b/file" } else { "file" });
+                if replacement == "directory" {
+                    fs::create_dir_all(visible_leaf.parent().expect("leaf parent"))
+                        .expect("create replacement lineage");
+                    fs::write(&visible_leaf, b"replacement").expect("write replacement");
+                }
+                let mut called = false;
+                let attempt = backend.execute(&context, entry, |entry| {
+                    called = true;
+                    entry
+                        .target()
+                        .replace(wip_interface::Bytes(b"bad".to_vec()))
+                        .expect("would write pinned inode");
+                    OperationAttempt::reject(CallErrorKind::ValidatorMismatch)
+                });
+                assert!(
+                    matches!(attempt, ProviderResult::NotFound),
+                    "{ancestor}: {replacement}"
+                );
+                assert!(!called);
+                assert_eq!(
+                    pinned.read(None, None).expect("read original inode").0,
+                    b"original"
+                );
+                if replacement != "missing" {
+                    assert_eq!(
+                        fs::read(&visible_leaf).expect("read visible leaf"),
+                        if replacement == "directory" {
+                            b"replacement".as_slice()
+                        } else {
+                            b"original".as_slice()
+                        }
+                    );
+                }
+                // Rejection releases every guard, even while the target is retained.
+                let root_probe = fs::File::open(temporary.path()).expect("open root probe");
+                rustix::fs::flock(&root_probe, FlockOperation::NonBlockingLockExclusive)
+                    .expect("rejection released root guard");
+            }
+        }
+    }
+
+    #[test]
+    fn execution_retains_and_refreshes_ancestors_under_ordered_directory_guards() {
+        let temporary = tempdir().expect("temporary root");
+        fs::create_dir_all(temporary.path().join("a/b")).expect("create lineage");
+        fs::write(temporary.path().join("a/b/file"), b"original").expect("write file");
+        let backend = LinuxBackend::open(
+            temporary.path(),
+            AccessMode::ReadOnly,
+            ResourceLimits::default(),
+        )
+        .expect("open backend");
+        let context = ExecutionContext::new();
+        let entry = resolved(&backend, &context, "/a/b/file");
+        assert_eq!(
+            entry
+                .ancestors()
+                .iter()
+                .map(WipObject::name)
+                .collect::<Vec<_>>(),
+            ["", "a", "b"]
+        );
+        let pinned_ancestors = entry.ancestors().to_vec();
+        let (mut ancestors, target) = entry.into_parts();
+        // Stale snapshots must not escape the execution boundary.
+        for ancestor in &mut ancestors {
+            ancestor.validator = Some(wip_host::Validator::new(vec![1]));
+            ancestor.initial_validator = Some(vec![1]);
+        }
+        let probes: Vec<_> = ["", "a", "a/b", "a/b/file"]
+            .into_iter()
+            .map(|path| fs::File::open(temporary.path().join(path)).expect("open lock probe"))
+            .collect();
+        let attempt = backend.execute(&context, ResolvedEntry::new(ancestors, target), |entry| {
+            for (ancestor, pinned) in entry.ancestors().iter().zip(&pinned_ancestors) {
+                assert!(std::sync::Arc::ptr_eq(&ancestor.inner, &pinned.inner));
+                assert!(ancestor.validator().is_none());
+                assert!(ancestor.initial_validator.is_none());
+            }
+            for probe in &probes {
+                assert_eq!(
+                    rustix::fs::flock(probe, FlockOperation::NonBlockingLockExclusive),
+                    Err(rustix::io::Errno::WOULDBLOCK)
+                );
+            }
+            OperationAttempt::reject(CallErrorKind::ValidatorMismatch)
+        });
+        assert!(matches!(attempt, ProviderResult::Found(_)));
+        for probe in &probes {
+            rustix::fs::flock(probe, FlockOperation::NonBlockingLockExclusive)
+                .expect("guard released after callback");
+            rustix::fs::flock(probe, FlockOperation::Unlock).expect("unlock probe");
+        }
+    }
+
+    #[test]
+    fn configured_root_stays_pinned_after_textual_rename_and_replacement() {
+        let temporary = tempdir().expect("temporary parent");
+        let configured = temporary.path().join("root");
+        fs::create_dir_all(configured.join("dir")).expect("create original root");
+        fs::write(configured.join("dir/file"), b"original").expect("write original file");
+        let backend = LinuxBackend::open(
+            &configured,
+            AccessMode::ReadWrite,
+            ResourceLimits::default(),
+        )
+        .expect("open backend");
+        let context = ExecutionContext::new();
+        let root = resolved(&backend, &context, "/");
+        assert!(root.ancestors().is_empty());
+        let file = resolved(&backend, &context, "/dir/file");
+        let moved = temporary.path().join("moved");
+        fs::rename(&configured, &moved).expect("rename configured root");
+        fs::create_dir_all(configured.join("dir")).expect("replace textual root");
+        fs::write(configured.join("dir/file"), b"replacement").expect("write replacement file");
+        for entry in [
+            root,
+            resolved(&backend, &context, "/"),
+            resolved(&backend, &context, "/dir"),
+        ] {
+            let attempt = backend.execute(&context, entry, |entry| {
+                assert_eq!(entry.target().kind, crate::EntryKind::Directory);
+                OperationAttempt::reject(CallErrorKind::ValidatorMismatch)
+            });
+            assert!(
+                matches!(attempt, ProviderResult::Found(_)),
+                "root/directory locks once"
+            );
+        }
+        let attempt = backend.execute(&context, file, |entry| {
+            entry
+                .target()
+                .replace(wip_interface::Bytes(b"pinned write".to_vec()))
+                .expect("write beneath original root");
+            OperationAttempt::Started(Ok(wip_host::CallResult {
+                result: Value::Bytes(Vec::new()),
+                validator: None,
+            }))
+        });
+        assert!(matches!(
+            attempt,
+            ProviderResult::Found(OperationAttempt::Started(Ok(_)))
+        ));
+        assert_eq!(
+            fs::read(moved.join("dir/file")).expect("read original root"),
+            b"pinned write"
+        );
+        assert_eq!(
+            fs::read(configured.join("dir/file")).expect("read textual replacement"),
+            b"replacement"
+        );
+        assert_eq!(
+            resolved(&backend, &context, "/dir/file")
+                .target()
+                .read(None, None)
+                .expect("resolution still uses pinned root")
+                .0,
+            b"pinned write"
+        );
+    }
+
+    #[test]
+    fn deep_resolution_counts_all_retained_ancestors_against_handle_budget() {
+        let temporary = tempdir().expect("temporary root");
+        fs::create_dir_all(temporary.path().join("a/b")).expect("create lineage");
+        fs::write(temporary.path().join("a/b/file"), b"original").expect("write file");
+        let context = ExecutionContext::new();
+        for maximum in [3, 4] {
+            let backend = LinuxBackend::open(
+                temporary.path(),
+                AccessMode::ReadOnly,
+                ResourceLimits {
+                    max_open_handles: maximum,
+                    ..ResourceLimits::default()
+                },
+            )
+            .expect("open backend");
+            match backend.resolve(&context, "/a/b/file") {
+                ProviderResult::Failure(error) if maximum == 3 => {
+                    assert_eq!(error.kind(), wip_host::HostErrorKind::ResourceLimitExceeded);
+                    assert_eq!(
+                        error.effect_guarantee(),
+                        wip_host::EffectGuarantee::NotCommitted
+                    );
+                }
+                ProviderResult::Found(entry) if maximum == 4 => {
+                    assert_eq!(entry.ancestors().len(), 3);
+                    assert_eq!(backend.handles.current.load(super::Ordering::Acquire), 4);
+                    let cloned =
+                        ResolvedEntry::new(entry.ancestors().to_vec(), entry.target().clone());
+                    drop(entry);
+                    assert_eq!(backend.handles.current.load(super::Ordering::Acquire), 4);
+                    drop(cloned);
+                }
+                _ => panic!("unexpected resolution outcome for budget {maximum}"),
+            }
+            assert_eq!(backend.handles.current.load(super::Ordering::Acquire), 0);
+            let root = resolved(&backend, &context, "/");
+            assert_eq!(backend.handles.current.load(super::Ordering::Acquire), 1);
+            drop(root);
+            assert_eq!(backend.handles.current.load(super::Ordering::Acquire), 0);
+        }
+    }
+
     #[test]
     fn rejected_execution_releases_lock_even_when_callback_retains_entry() {
         let temporary = tempdir().expect("temporary root");
@@ -914,7 +1235,7 @@ mod tests {
         .expect("open backend");
         let context = ExecutionContext::new();
         let entry = resolved(&backend, &context, "/file");
-        let cloned = entry.clone();
+        let cloned = ResolvedEntry::new(entry.ancestors().to_vec(), entry.target().clone());
         let attempt = backend.execute(&context, entry, |entry| {
             let started = std::time::Instant::now();
             let nested = backend.execute(&context, entry, |_| {
@@ -956,10 +1277,10 @@ mod tests {
         .expect("open backend");
         let context = ExecutionContext::new();
         let root = resolved(&backend, &context, "/");
-        assert!(backend.children(&context, &root, 1).is_err());
+        assert!(backend.children(&context, root.target(), 1).is_err());
         assert_eq!(backend.handles.current.load(super::Ordering::Acquire), 1);
         let children = backend
-            .children(&context, &root, 2)
+            .children(&context, root.target(), 2)
             .expect("complete children");
         assert_eq!(children.len(), 2);
         let cloned = children[0].clone();
@@ -986,13 +1307,14 @@ mod tests {
         .expect("open backend");
         let context = ExecutionContext::new();
         let entry = resolved(&backend, &context, "/file");
-        assert_eq!(entry.access_mode(), AccessMode::ReadOnly);
-        assert!(entry.supports_file_read());
-        assert!(!entry.supports_file_write());
-        assert!(!entry.supports_symlink_inspect());
-        assert!(entry.read(None, None).is_err());
-        assert_eq!(entry.read(None, Some(2)).expect("bounded read").0, b"ol");
-        assert!(entry.replace(wip_interface::Bytes(Vec::new())).is_err());
+        let target = entry.target();
+        assert_eq!(target.access_mode(), AccessMode::ReadOnly);
+        assert!(target.supports_file_read());
+        assert!(!target.supports_file_write());
+        assert!(!target.supports_symlink_inspect());
+        assert!(target.read(None, None).is_err());
+        assert_eq!(target.read(None, Some(2)).expect("bounded read").0, b"ol");
+        assert!(target.replace(wip_interface::Bytes(Vec::new())).is_err());
         assert_eq!(fs::read(&path).expect("read unchanged file"), b"old");
         let cancellation = wip_host::CancellationToken::new();
         cancellation.cancel();

crates/wip-fs-server/tests/extensions.rs

@@ -5,7 +5,7 @@ use std::time::{Duration, Instant};
 
 use wip_fs_server::{
     BackendEntry, Composition, DefaultInterfaces, EntryKind, FileRead, FilesystemBackend,
-    HttpAdapter, Projection, ProjectionPolicy, ResourceLimits,
+    HttpAdapter, Projection, ProjectionPolicy, ResolvedEntry, ResourceLimits,
 };
 use wip_host::{
     CancellationToken, EffectGuarantee, ExecutionContext, HostError, InterfaceRef,
@@ -77,15 +77,20 @@ struct Entry {
     path: String,
     node: Arc<Node>,
     validator: Validator,
+    hidden: bool,
 }
 impl Entry {
     fn snapshot(path: String, node: Arc<Node>) -> Self {
-        let revision = node.data.lock().unwrap().revision;
+        let data = node.data.lock().unwrap();
+        let revision = data.revision;
+        let hidden = data.hidden;
+        drop(data);
         let validator = Validator::new(format!("{}:{revision}", node.id).into_bytes());
         Self {
             path,
             node,
             validator,
+            hidden,
         }
     }
 }
@@ -118,6 +123,8 @@ struct Seen {
 struct Controls {
     bump_before_execute: bool,
     hide_before_execute: bool,
+    hide_ancestor: Option<String>,
+    replace_ancestor: Option<String>,
     replace_mapping: bool,
     rollback: bool,
     cancel_on_execute: bool,
@@ -168,7 +175,11 @@ fn check_context(context: &ExecutionContext) -> Result<(), HostError> {
 }
 impl FilesystemBackend for MemoryBackend {
     type Entry = Entry;
-    fn resolve(&self, context: &ExecutionContext, path: &str) -> ProviderResult<Entry> {
+    fn resolve(
+        &self,
+        context: &ExecutionContext,
+        path: &str,
+    ) -> ProviderResult<ResolvedEntry<Entry>> {
         self.0.record("resolve", path, context);
         if let Err(error) = check_context(context) {
             return ProviderResult::Failure(error);
@@ -179,11 +190,33 @@ impl FilesystemBackend for MemoryBackend {
                 EffectGuarantee::NotCommitted,
             ));
         }
-        // Map lookup only: there is no host path to reopen or escape.
-        match self.0.nodes.lock().unwrap().get(path).cloned() {
-            Some(node) => ProviderResult::Found(Entry::snapshot(path.into(), node)),
-            None => ProviderResult::NotFound,
+        // One connected map snapshot, serialized with participating mutations.
+        let _transaction = self.0.transaction.lock().unwrap();
+        let nodes = self.0.nodes.lock().unwrap();
+        let Some(node) = nodes.get(path) else {
+            return ProviderResult::NotFound;
+        };
+        let mut ancestors = Vec::new();
+        if path != "/" {
+            let prefixes = std::iter::once("/").chain(
+                path.match_indices('/')
+                    .skip(1)
+                    .map(|(index, _)| &path[..index]),
+            );
+            for prefix in prefixes {
+                let Some(parent) = nodes.get(prefix) else {
+                    return ProviderResult::NotFound;
+                };
+                if parent.kind != EntryKind::Directory {
+                    return ProviderResult::NotFound;
                 }
+                ancestors.push(Entry::snapshot(prefix.into(), parent.clone()));
+            }
+        }
+        ProviderResult::Found(ResolvedEntry::new(
+            ancestors,
+            Entry::snapshot(path.into(), node.clone()),
+        ))
     }
     fn children(
         &self,
@@ -224,10 +257,11 @@ impl FilesystemBackend for MemoryBackend {
     fn execute(
         &self,
         context: &ExecutionContext,
-        entry: Entry,
-        operation: impl FnOnce(Entry) -> OperationAttempt,
+        resolved: ResolvedEntry<Entry>,
+        operation: impl FnOnce(ResolvedEntry<Entry>) -> OperationAttempt,
     ) -> ProviderResult<OperationAttempt> {
         let _lock = self.0.transaction.lock().unwrap();
+        let (ancestors, entry) = resolved.into_parts();
         self.0.record("execute", &entry.path, context);
         if let Err(error) = check_context(context) {
             return ProviderResult::Failure(error);
@@ -243,6 +277,24 @@ impl FilesystemBackend for MemoryBackend {
         if controls.hide_before_execute {
             entry.node.data.lock().unwrap().hidden = true;
         }
+        if let Some(path) = controls.hide_ancestor.take() {
+            self.0.node(&path).data.lock().unwrap().hidden = true;
+        }
+        if let Some(path) = controls.replace_ancestor.take() {
+            self.0.nodes.lock().unwrap().insert(
+                path,
+                Arc::new(Node {
+                    id: 1000,
+                    kind: EntryKind::Directory,
+                    data: Mutex::new(Data {
+                        bytes: b"replacement ancestor".to_vec(),
+                        revision: 1,
+                        hidden: false,
+                    }),
+                    calls: Mutex::new(Vec::new()),
+                }),
+            );
+        }
         if controls.replace_mapping {
             // The callback must still receive the original capability, not this decoy.
             self.0.nodes.lock().unwrap().insert(
@@ -265,9 +317,25 @@ impl FilesystemBackend for MemoryBackend {
         }
         let rollback = controls.rollback;
         drop(controls);
+        // Verify lineage identity/membership within the same transaction. The
+        // callback cannot see a substituted ancestor, nor a redirected target.
+        let nodes = self.0.nodes.lock().unwrap();
+        for ancestor in &ancestors {
+            if !nodes
+                .get(&ancestor.path)
+                .is_some_and(|node| Arc::ptr_eq(node, &ancestor.node))
+            {
+                return ProviderResult::NotFound;
+            }
+        }
+        drop(nodes);
+        let refreshed_ancestors = ancestors
+            .into_iter()
+            .map(|ancestor| Entry::snapshot(ancestor.path, ancestor.node))
+            .collect();
         let before = entry.node.data.lock().unwrap().clone();
         let refreshed = Entry::snapshot(entry.path.clone(), entry.node.clone());
-        let attempt = operation(refreshed);
+        let attempt = operation(ResolvedEntry::new(refreshed_ancestors, refreshed));
         let attempt = match attempt {
             OperationAttempt::Started(Ok(mut result)) => {
                 result.validator = Some(
@@ -437,8 +505,7 @@ struct Policy {
 impl ProjectionPolicy<Entry, Object> for Policy {
     type Principal = Principal;
     fn object_visible(&self, principal: &Principal, entry: &Entry) -> bool {
-        !entry.node.data.lock().unwrap().hidden
-            && (!matches!(principal, Principal::Limited) || entry.path != "/secret")
+        !entry.hidden && (!matches!(principal, Principal::Limited) || entry.path != "/secret")
     }
     fn interface_visible(&self, principal: &Principal, interface: &InterfaceRef) -> bool {
         !(matches!(principal, Principal::Limited) && self.hide_read && interface == &self.read)
@@ -690,6 +757,109 @@ impl Fixture {
         (response, decoded)
     }
 }
+#[test]
+fn ancestor_visibility_refresh_inside_boundary_prevents_descendant_invocation_and_effects() {
+    for (target, ancestor) in [
+        ("/added", "/"),
+        ("/secret/nested/data", "/"),
+        ("/secret/nested/data", "/secret"),
+        ("/secret/nested/data", "/secret/nested"),
+    ] {
+        for operation in ["denied", "read", "change"] {
+            let fixture = Fixture::new();
+            let observation = fixture.observe(&Principal::Admin, target, 0);
+            let descriptor = fixture.descriptor(&Principal::Admin, &fixture.custom);
+            let mut request = fixture.request(
+                target,
+                &fixture.custom,
+                operation,
+                observation.object.validator,
+            );
+            if operation == "change" {
+                request
+                    .arguments
+                    .insert("contents".into(), Value::Bytes(b"must not commit".to_vec()));
+            }
+            let leaf = fixture.shared.node(target);
+            let before = leaf.data.lock().unwrap().clone();
+            fixture.shared.controls.lock().unwrap().hide_ancestor = Some(ancestor.into());
+            let (wire, response) = fixture.http_call(
+                &Principal::Admin,
+                &request,
+                &descriptor,
+                &ExecutionContext::new(),
+            );
+            wire_error(response, ProtocolErrorCode::NotFound);
+            assert!(leaf.calls.lock().unwrap().is_empty());
+            assert_eq!(*leaf.data.lock().unwrap(), before);
+            assert!(!String::from_utf8_lossy(wire.body()).contains(BACKEND_SECRET));
+            let error = fixture
+                .engine
+                .view(&Principal::Admin)
+                .observe(&ExecutionContext::new(), &observe_request(target, 0))
+                .unwrap_err();
+            assert_eq!(error.code, ProtocolErrorCode::NotFound);
+            // The entire target resolves once, not as unrelated path prefixes.
+            let seen = fixture.shared.seen.lock().unwrap();
+            assert!(
+                seen.iter()
+                    .filter(|event| event.stage == "resolve")
+                    .all(|event| event.path == target)
+            );
+        }
+    }
+}
+
+#[test]
+fn replaced_visible_ancestors_reject_without_redirecting_or_dispatching_the_pinned_target() {
+    for (target, ancestor) in [
+        ("/added", "/"),
+        ("/secret/nested/data", "/secret"),
+        ("/secret/nested/data", "/secret/nested"),
+    ] {
+        let fixture = Fixture::new();
+        let observation = fixture.observe(&Principal::Admin, target, 0);
+        let descriptor = fixture.descriptor(&Principal::Admin, &fixture.custom);
+        let mut request = fixture.request(
+            target,
+            &fixture.custom,
+            "change",
+            observation.object.validator,
+        );
+        request
+            .arguments
+            .insert("contents".into(), Value::Bytes(b"must not commit".to_vec()));
+        let leaf = fixture.shared.node(target);
+        let original_ancestor = fixture.shared.node(ancestor);
+        let before = leaf.data.lock().unwrap().clone();
+        fixture.shared.controls.lock().unwrap().replace_ancestor = Some(ancestor.into());
+        let (_, response) = fixture.http_call(
+            &Principal::Admin,
+            &request,
+            &descriptor,
+            &ExecutionContext::new(),
+        );
+        wire_error(response, ProtocolErrorCode::NotFound);
+        assert!(leaf.calls.lock().unwrap().is_empty());
+        assert_eq!(*leaf.data.lock().unwrap(), before);
+        assert!(!Arc::ptr_eq(
+            &original_ancestor,
+            &fixture.shared.node(ancestor)
+        ));
+        assert!(
+            fixture
+                .shared
+                .node(ancestor)
+                .calls
+                .lock()
+                .unwrap()
+                .is_empty()
+        );
+        // A replacement is visible but cannot authorize a previously resolved lineage.
+        assert!(!fixture.shared.node(ancestor).data.lock().unwrap().hidden);
+    }
+}
+
 fn observe_request(path: &str, depth: u32) -> ObserveRequest {
     ObserveRequest {
         path: path.into(),
@@ -868,8 +1038,8 @@ fn principal_views_hide_children_and_guessed_descendants_for_observation_and_cal
             .lock()
             .unwrap()
             .iter()
-            .all(|seen| !seen.path.starts_with("/secret/")),
-        "hidden descendants were resolved"
+            .all(|seen| seen.stage == "resolve"),
+        "hidden lineage was published, enumerated, or dispatched"
     );
     assert!(
         fixture