docs(memory-bank): record the last checks of round 2 and the final head of rc9

Record that the branch ai/rc9 ends at 696ac38 (36 commits over d0cb57c, 2,162
cases per configuration, 0 failed), that the last two checks of round 2 found
four Nits in the message formatter and in the reading of a sequence (all fixed
test-first, the last one at its root), that both reviewers closed with no
finding, and that a collection that holds itself still ends the host through
the three flags parameters, as in rc8, which stays a known limit. The lesson
for the next agent: a failing Dispose must never change what was read.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: AI Assistant <ai@example.com>

.memory-bank/activeContext.md

@@ -17,7 +17,7 @@ passed; lab acceptance of the published bytes 10:38 to 11:47Z, record
 #129 (08:24Z), and the Memory Bank #130 (08:47Z).
 
 The last prerelease before the stable 5.0.0 is rc9 (Decision 27). The branch
-`ai/rc9` (base `d0cb57c`, 32 commits, head `de67883`) is complete and
+`ai/rc9` (base `d0cb57c`, 36 commits, head `696ac38`) is complete and
 validated; it is local until the maintainer pushes it, opens the pull request
 (body: `C:\Temp\ntfs-rc9\final-msgs\pr-rc9-body.md`, temporary), merges it, and
 tags `5.0.0-rc9`. It contains: the five piped parameters of `Add-` and
@@ -31,7 +31,7 @@ the Windows PowerShell 5.1 rules; the library fixes 3.1 to 3.3 and 3.10;
 `Set-NTFSSecurityDescriptor` at position 1; tests that bind four parameters by
 property name; label `rc9`.
 
-Validation of `de67883`: the full suite in the four CI configurations, 2,116
+Validation of `696ac38`: the full suite in the four CI configurations, 2,162
 cases each, 0 failed (rc8: 1,615; skipped 27, 359, 58, and 390, so every new
 case runs where it can); the pages and the help file are a fixpoint (0 drift, 0
 broken links; the help file is identical after a second generation);
@@ -42,13 +42,16 @@ matrix of valid shapes of the five parameters, run on the published rc8 and on
 this build in both editions, differs in exactly the 9 list rows of
 `Add-NTFSAccess`. Two review rounds (Decision 27, "Built in rc9"): round 1 on the
 first integration found a Blocker (the last-dot rule of the filter) and two
-Major findings; round 2 on the final tree and three checks of its fixes found
-only Minor findings (wrong statements in the pages, completers that were not
-culture invariant, the list for `-AccessType` and `-AppliesTo`, ways around
-that rejection, a collection that holds itself); all are fixed with a test
-that is red without the fix, or decided. A blank, `$null`, or `DBNull` value of
-the other parameter group is `AmbiguousParameterSet` (documented, not
-restructured).
+Major findings; round 2 on the final tree and five checks of its fixes found
+only Minor findings and Nits (wrong statements in the pages, completers that
+were not culture invariant, the list for `-AccessType` and `-AppliesTo`, ways
+around that rejection, a collection that holds itself, a message formatter that
+threw, an enumerator that fails in `Dispose`); all are fixed with a test that
+is red without the fix, or decided, and both reviewers closed their last
+findings with "no finding" on `696ac38`. A collection that holds itself still
+ends the host through the three flags parameters, as in rc8 (known limit). A
+blank, `$null`, or `DBNull` value of the other parameter group is
+`AmbiguousParameterSet` (documented, not restructured).
 
 Open before the stable pull request: the pull request, merge, and tag of rc9
 (maintainer); the lab acceptance of the published rc9 (agent, about 70
@@ -71,7 +74,8 @@ Decisions 22 and 24 on 2026-10-19 and decides 23 on 2026-11-02.
 - Evidence of rc9 is outside git, in the session folder
   `816cf47d-2f80-4eba-8ac9-41702c49560b\files\rc9`: the specs and review briefs,
   `suite-full1` (first integration), `suite-final`, `suite-final2`,
-  `suite-final3`, and `suite-final4` (the head `de67883`), the probes (`p-*.ps1`,
+  `suite-final3`, `suite-final4`, `suite-final5`, `suite-final6-b436423`, and
+  `suite-final7-696ac38` (the head), the probes (`p-*.ps1`,
   `fixb-probe`, `shapes-r2`, `Run-Oracle.ps1`, `Run-Pathological.ps1`), the
   published rc8 (`rc8-published`), and the analyzer and surface results. The
   review probes of the reviewers are in `C:\Temp\ntfs-rc9\review*-probes-*`

.memory-bank/debugging-insights.md

@@ -163,7 +163,31 @@ fails without it (Decision 16).
   showed that exactly the 9 list rows changed. Code that opens a one-element
   collection needs a depth limit: an array that holds itself hung the first
   version (the binder of rc8 ends the host with a stack overflow on it), so
-  the conversion opens at most 32 levels and reads at most two elements.
+  the conversion opens at most 32 levels and reads at most two elements. The
+  three flags parameters still hand such a collection to PowerShell's own
+  conversion, which ends the host with a stack overflow, as in rc8; not fixed
+  (no real input builds one, and a test would end the Pester process too).
+- The message of a bad value is built inside a catch block, and a formatter
+  there must not throw: `Describe` called `ToString` of the one element outside
+  a guard, so an element whose `ToString` throws (a script method) ended a
+  piped command with an `ExtendedTypeSystemException` instead of one error for
+  that item (found by the last delta check of the second review; the version
+  before named a collection by its type and never formatted an element). Wrap
+  the whole formatter and fall back to the type name, in the short form
+  (`Type.ToString`; `FullName` spells out the assembly of every generic
+  argument). A test that pins a limit (ten elements named, two elements read)
+  is green on the code that it guards: prove it with a mutant of the limit (no
+  limit: 4 rows red; a limit of nine: 6 rows red).
+- A loop that stops early runs `Dispose` of the enumerator, and `Dispose` can
+  fail: the catch of `TakeOneValue` returned the collection to PowerShell's
+  conversion, which combines a list bit by bit, so a sequence whose enumerator
+  throws in `Dispose` let a list through. The first patch (`isList = count >
+  1` in the catch) fixed the reported shape (two elements) and left its
+  sibling (one element that holds a comma list); the closing check found it.
+  Fix the root: read with an explicit enumerator and release it in its own
+  guard, so that a failing release never changes what was read. The test needs
+  a compiled stand-in (`Add-Type`, as `PipelineControl.Tests.ps1` does); no
+  built-in enumerator fails in `Dispose`.
 - PowerShell calls the getter of a by-property-name parameter once per
   command, when a piped object first binds it (`p-getter-calls.ps1`). A test
   that must visit it pipes an object that has the property. A property getter

.memory-bank/decisions/0027-open-items-before-5.0.0.md

@@ -183,11 +183,23 @@ decided; the rest is to be chosen. Recommended defaults are the agent's.
   string), that a collection that holds itself never returned (the binder of
   rc8 ends the host with a stack overflow on it), and that the message named
   the type, not the typo: fixed in `e39aeed` and `da41a8e`, each with tests
-  that were red first (32 and 28 rows). Final suite on `de67883`: 2,116 cases
-  per configuration, 0 failed in four configurations (rc8: 1,615); the pages
-  and the help file are a fixpoint; PSScriptAnalyzer finds nothing in the 11
-  changed scripts; the public surface equals the validated candidate; a matrix
-  of valid shapes differs from the published rc8 in exactly 9 list rows.
+  that were red first (32 and 28 rows). The last two checks found four Nits in
+  code that those fixes added or kept: the formatter of the message threw when
+  the text of an element throws (a terminating error that stopped a piped
+  command), its fallback named a generic type with the assembly of every type
+  argument, and an enumerator that throws in `Dispose` turned a list into a
+  value that PowerShell converts (first for two elements, then for one element
+  that holds a list): fixed in `ec58a18`, `ce34395`, `b436423`, and `696ac38`,
+  with 24, 2, 4, and 8 rows that were red first, and the pins of the limits
+  proved with two mutants. The last commit fixes the root,
+  not the shape: the collection is read with an explicit enumerator that is
+  released in its own guard, so a failing release changes nothing of what was
+  read. Final suite on `696ac38`: 2,162 cases per configuration, 0 failed in
+  four configurations (rc8: 1,615); the pages and the help file are a
+  fixpoint; PSScriptAnalyzer finds nothing in the 11 changed scripts; the
+  public surface equals the validated candidate; a matrix of valid shapes
+  differs from the published rc8 in exactly 9 list rows; both reviewers closed
+  their findings with "no finding".
 
 What the review found:
 
@@ -219,3 +231,12 @@ Known limits, kept and to be stated to the maintainer:
   stays undisposed, a known limit of the .NET API.
 - In PowerShell 7, the case folding of `-Filter` differs from NTFS for a few
   exotic letters (finding F7 of the review; not fixed).
+- A collection that holds itself, given to `-InheritanceFlags`,
+  `-PropagationFlags`, or `-AuditFlags`, ends the host with a stack overflow
+  inside PowerShell's own conversion, exactly as in rc8 (both reviewers
+  reproduced it in both editions, without the module too: any typed parameter
+  or cast does it). `-AccessType` and `-AppliesTo` reject it. Not fixed: no
+  real input builds such an object, a guard for the flags parameters would
+  need a visited-set walk of every argument, and a regression test would end
+  the Pester process before the fix. Named in the pull request text, not in
+  the CHANGELOG (user-visible changes only).

.memory-bank/progress.md

@@ -119,19 +119,20 @@ archive in favor of WindowsAccessControl (Decision 18).
   creates an Allow entry); Decision 27 sets the scope of rc9, the designs were
   chosen on 2026-10-21, and four work items started (branch
   `ai/rc9-before-stable`).
-- 2026-10-21: rc9 built and reviewed (Decision 27; branch `ai/rc9`, 32 commits
-  over `d0cb57c`, head `de67883`, local until he pushes). Four implementers
+- 2026-10-21: rc9 built and reviewed (Decision 27; branch `ai/rc9`, 36 commits
+  over `d0cb57c`, head `696ac38`, local until he pushes). Four implementers
   worked test-first in worktrees (the five piped parameters, the 17 handlers
   and three inconsistencies, the library items, the filter); the chains were
   cherry-picked in order, so every fix sits in the commit of its item. Round 1
   (three read-only reviewers) found a Blocker (the last-dot rule of the
   filter) and two Major findings (a named `$null`; a blank value of the other
-  parameter group, documented); round 2 (two reviewers, three checks) found
-  only Minor findings, among them completers that were not culture invariant
-  (`tr-TR`) and a list for `-AccessType` or `-AppliesTo`, which PowerShell
-  combines bit by bit; he chose to reject it. Final suite on `de67883`: 2,116
-  cases per configuration, 0 failed in four configurations (rc8: 1,615); docs
-  fixpoint; analyzer clean; valid shapes differ from rc8 in 9 list rows only.
+  parameter group, documented); round 2 (two reviewers, five checks) found
+  only Minor findings and Nits, among them completers that were not culture
+  invariant (`tr-TR`) and a list for `-AccessType` or `-AppliesTo`, which
+  PowerShell combines bit by bit; he chose to reject it. Final suite on
+  `696ac38`: 2,162 cases per configuration, 0 failed in four configurations
+  (rc8: 1,615); docs fixpoint; analyzer clean; valid shapes differ from rc8 in
+  9 list rows only; both reviewers: no finding.
 
 ## Stable capabilities