Hold the first words until the news is checked

When you ask Mu about current events, the answer now holds its tongue until the freshness guard has done its work. Previously, streaming could begin before the tool had finished checking whether the news results were recent enough to share confidently. A user might see half a sentence about a story before the system realized the data was from last week and added a caveat. That gap eroded trust.

The fix is small by design. As soon as a news or headline tool is invoked, the stream pauses. The freshness guard runs its check — comparing timestamps, flagging results older than the threshold — and only then does Mu start speaking. If the only available information is dated, the first words warn of that. If the news is fresh, the answer flows without a later correction.

This does not make news perfect. The guard still depends on what the underlying search index returns; it can only flag stale data, not invent current data where none exists. And it adds a brief delay — a few hundred milliseconds — to the start of the response. That trade-off is intentional: a waiting cursor is less harmful than a misleading first impression.

The goal is straightforward: when Mu answers a time-sensitive question, the answer should be honest from the first word. No bait-and-switch, no retroactive “actually, this is old.” The tool protects the person’s trust, not the system’s speed.