One local recovery instruction for Ahmed's EXISTING visible setup chat

Continue the already authorized Ahmed Messenger repair using this extracted
package. Do not create another chat, grant, receiver, or backend, and do not ask
for or paste a password. Keep %USERPROFILE%\.ews-conduit\Ahmed in place and keep
the staged nonblocking Receive-Automatically.ps1 and Receiver-Nonblocking.ps1.

1. Verify package manifest hashes. Checkpoint the ONE existing
Ahmed-EWS-Conduit-Receiver task, its actual CLI child and durable pending/inflight
state before replacing loaded client code. Locate its actual imported device
client directory from its existing source/config; do not guess a path. Back up
only these code files. Apply the included Ews-Conduit.ps1, Ews-Device.ps1,
Ews-Mailbox-Conflicts.ps1 and Repair-EwsMailboxConflict.ps1 to that existing client
location as a single quiesced update, preserving the exact POLICY.v2.json. Pass
-StateDirectory "$env:USERPROFILE\.ews-conduit\Ahmed" for every invocation.
Do not copy or decrypt/export the grant. Do not clear protocolStopped manually.

2. In that canonical directory run the repaired DeviceStatus and Receive once.
The legacy MAILBOX_ID_CONFLICT must still be present and hold ALL mailbox writes;
the repaired client permits these read operations. Preserve all pending IDs and
receipts. Inspect the actual staged receiver source before resuming: it must save
validated intake/dedup, hold failed receipts, and dispatch existing pending input
independently of successful Ack. If Ack currently gates dispatch, fix that same
receiver without discarding state or replaying ambiguous inflight work. Keep one
receiver/task and IgnoreNew instances policy. Do not launch a substitute CLI.

3. Run Get-ConflictEvidence.ps1 -StateDirectory "$env:USERPROFILE\.ews-conduit\Ahmed".
It compares only saved pending/receipt hashes against the owner-side mailbox
digest snapshot shipped here. It outputs identifiers/digests, never message bodies
or credentials. Correlate candidates with the original failed operation/log.
A single differing candidate alone is not historical proof. If local evidence is
insufficient, retain the legacy write hold and return only these safe candidate
IDs/kinds/digests through the restored supported original visible route for exact
owner-side verification. The device grant cannot read outgoing/receipt server
files; do not attempt a forbidden GET or broaden its scope.

Only once the exact UUID/kind/local digest/server digest are verified, run:
  powershell.exe -NoProfile -File .\device-client\Repair-EwsMailboxConflict.ps1 -StateDirectory "$env:USERPROFILE\.ews-conduit\Ahmed" -MessageId <proven-uuid> -Kind <message-or-receipt> -ExpectedLocalSha256 <exact-local-sha256> -ExpectedServerSha256 <owner-exact-path-sha256>
This commits that write's quarantine before clearing only the legacy mailbox
hold. Different content under that UUID stays rejected; no immutable files or
IDs change. If the UUID is still unknown, stop this recovery step, keep reads
available, and report that exact blocker. Do not use a guessed candidate.

4. Install the UI using Install-Messenger.ps1. It reuses the canonical state and
existing receiver; it never restarts it or installs a competing receiver. Verify
original sidebar history using the running app's supported APIs and filters as
described in DESKTOP-CONTEXT.txt. Keep original chat IDs/account/projects/history
and verify original CODEX_HOME separately from native/Electron profile isolation.
Do not launch the older custom backend against the newer store without verified
compatibility; no DB/rollout edits, resets or token/profile copying.

5. Bind only original visible coordinator 01a1959d-01db-7771-b4c7-517dc40606f5
after verifying its actual owning app/API. If that owner exposes the supported
Codex Control local task/rpc API, use Bind-VisibleCoordinator.ps1 with its actual
token-file reference, owning PID, original CODEX_HOME and visible-thread proof.
If it does not, report that missing API contract; do not invent an endpoint or
start a second backend. Verify held-conflict/task failure health appears in that
same visible chat. Reuse pending handoffs f098178e-c0e9-451d-b094-2b38bbcaddfe and
da193840-c2d6-489d-9d82-d4f1dad71301; do not publish duplicate handoffs/tests.
Record receipt, API acceptance, actual visible rendering and returned reply as
separate facts. Verify polling/receipts while the model is busy and after one
controlled existing-owner restart, preserving pending work and durable dedup.

Report source hashes, exact conflict evidence, original-history IDs/counts through
supported APIs, actual owner/version/endpoint and visible round-trip evidence.
Remote EWS server launch and the old rejected custom-startup capture remain out
of this recovery. No repeat consent/login loop is needed for the scoped repair.
