diff --git a/microsoft/knowledge/upgrade/no-changecompany-in-upgrade.md b/microsoft/knowledge/upgrade/no-changecompany-in-upgrade.md
index d8c496f..3295bba 100644
--- a/microsoft/knowledge/upgrade/no-changecompany-in-upgrade.md
+++ b/microsoft/knowledge/upgrade/no-changecompany-in-upgrade.md
@@ -11,7 +11,7 @@ application-area: [all]
 
 ## Description
 
-The platform upgrades each company in its own context: `OnUpgradePerCompany` (like the other `PerCompany` upgrade and install triggers) runs once per company, in a separate system session opened for that company, while `PerDatabase` triggers run in a session that opens no company. Calling `ChangeCompany(<name>)` in upgrade code reaches from the company being upgraded into another company. That company has its own upgrade session, so the work is either duplicated or racing it. The platform does not document an order between companies, so a read can see data in the pre-upgrade or post-upgrade shape. Per-company upgrade tags are set for the current company only, so a tag set after writing to company B is recorded for company A, and B's own session runs the migration again. Execution context does not move with `ChangeCompany`: table triggers and trigger-event subscribers still run in the calling company (see `changecompany-runs-triggers-in-the-calling-company`). Data and side effects then land in different companies, and an error in another company's data aborts the current company's upgrade. The same applies to the migration path of a feature-switch data update: the `UpdateData` and `AfterUpdate` methods of a `Feature Data Update` implementation. Feature Management runs them for one company's status row, either as a task scheduled in that company or in the current session. The interface's preflight methods, `IsDataUpdateRequired` and `ReviewData`, are a separate phase that reports scope before any update is scheduled.
+The platform upgrades each company in its own context: `OnUpgradePerCompany` (like the other `PerCompany` upgrade and install triggers) runs once per company, in a separate system session opened for that company, while `PerDatabase` triggers run in a session that opens no company. Calling `ChangeCompany(<name>)` in upgrade code reaches from the company being upgraded into another company. Cross-company work bypasses the platform's per-company execution boundary, making execution order and migration ownership difficult to reason about and potentially repeating work. A cross-company read cannot assume that the other company's migration has already run, so it can observe pre-upgrade or post-upgrade data. Per-company upgrade tags are set for the current company only, so a tag set after writing to company B is recorded for company A, which can cause B's own session to repeat the migration. Execution context does not move with `ChangeCompany`: table triggers and trigger-event subscribers still run in the calling company (see `changecompany-runs-triggers-in-the-calling-company`). Data and side effects then land in different companies, and an error in another company's data aborts the current company's upgrade. The same applies to the migration path of a feature-switch data update: the `UpdateData` and `AfterUpdate` methods of a `Feature Data Update` implementation. Feature Management runs them for one company's status row, either as a task scheduled in that company or in the current session. The interface's preflight methods, `IsDataUpdateRequired` and `ReviewData`, are a separate phase that reports scope before any update is scheduled.
 
 ## Best Practice