Server move (connect mode) always fails at "Stop running work" on 0.45.0: freeze blocks the bb-account RPC needed for the source host's grant

[RichieGP]

Summary

On bb 0.45.0, bb server move --to <machine> in connect mode fails within ~100 ms at step stop-work with "Server access provider failed". It reproduces every time (5 of 5 attempts, including after removing and re-enrolling the target from scratch).

Cause (from source at main)

runStopWork in apps/server/src/services/server-move/coordinator.ts:

1. sets the write freeze (setServerMoveFrozen(deps.db, true)), then
2. calls resolveServerHostGrant(move.sourceServerHost.id).

The source host here is a desktop-app server (primary host, connectMachineId: null), so it has no stored connect grant. The connect plugin's acquire() therefore calls createMachineCode(). Since #4253 (2026-09-29), that goes through the bb account plugin's RPC, POST /api/v1/plugins/bb-account/rpc/bb-account.v1.fetch. serverMoveFreezeMiddleware rejects it with 503 server_moving. hosted.ts maps the resulting AccountUnavailableError to MachineCodeError("not_paired"), and the move fails.

Server log, same second:

ERROR [server] Server error response {"method":"POST","path":"/api/v1/plugins/bb-account/rpc/bb-account.v1.fetch","status":503,"body":{"code":"server_moving",...}}
WARN  [server] [plugin:connect] server access failed: not_paired
WARN  [server] Server move failed {"step":"stop-work"}  Error: Server access provider failed

#4394 (bb 0.44.0, also a desktop-app source) got past stop-work, so this looks like a regression from moving connect onto the bb account plugin.

Side effect

resolve() writes the error to the source host's statusMessage, so the server machine then shows "Server access provider failed" in Settings → Machines.

Environment

• bb desktop 0.45.0 on macOS (source server, primary host). Target: Linux VM, manual provider, enrolled with install.sh, bb 0.45.0.
• bb Connect paired and healthy (bb connect status: paired, connected; bb account status: signed-in). bb connect machine-code works when the server isn't frozen.
• --check passes (canMove: true).

Suggested fixes

1. Resolve the source host's grant before setting the freeze, or
2. exempt the bb account / connect plugin RPC routes from the freeze (they're already in CONNECT_HOLD_SOURCES and stay running during the move), and
3. report the underlying error instead of "not_paired".

[bb-slop-cop]

🚨 SLOP COP 🚨 · new-issue-autopilot

• Classification: Bug · Priority: High · Effort: Low. Labels: remote, connect, confirmed-repro.
• Report: https://get-bb.github.io/reports/issues/4837.html — REPRODUCED, high root-cause confidence. The coordinator/HTTP-boundary regression fails in two clean trusted-main checkouts; no live cloud or macOS desktop test was performed.
• Root cause: The coordinator freezes public writes before resolving the source grant, blocking the account RPC POST that grant acquisition can require.
• Fix: Ready PR Resolve server-move source access before freezing writes #4838 resolves the grant before freezing writes, without changing freeze exemptions or authorization. Two files, 55 changed text lines.
• Verification: The regression returns 503 before the fix and 200 afterward; all 97 server-move tests, server typecheck, formatting, and git diff --check pass.