The verification gate

Read this before running step 3. Every re-engagement workflow assumes the CRM is right. It usually is not, and the failure is asymmetric: a missed recovery forfeits one opportunity, while messaging someone who already booked, already replied, or is already being worked costs the relationship.

Failures this gate exists to prevent

Each of these has produced a false drop in a live pipeline.

• A CRM last-contacted field reported a prospect silent for three weeks. The mailbox showed an email sent five days earlier. The field lags, misses sends from other addresses, and misses non-email channels entirely.
• A follow-up was flagged as never sent because the search used the company name as a keyword. It had been sent as a reply inside an older scheduling thread whose subject never contained that name.
• A mailbox search returned only its first page. Later pages were silently absent, and valid sends were flagged as missing.
• A thread search returned only the first few messages per thread. The most recent message, a submitted proposal, sat further down and was invisible to every search made against it.
• A CRM auto-flipped past meetings to a completed status regardless of whether anyone attended, making that status worthless as evidence a call happened.

The rule that follows: a single system is never sufficient evidence, in either direction. Not evidence that someone was dropped, and not evidence that they were not.

Verification rules

• Only a forward-dated event counts as a live booking. A past-dated meeting is not a booking to protect, and an auto-applied completed status is not proof anyone attended.
• When systems disagree, the system closest to the event wins. The calendar beats the CRM on bookings. The mailbox beats the CRM on last contact. The recorder beats the calendar on attendance.
• An empty result means the query was wrong, not that the person is cold. Resolve the actual contact record and email domain first, then search by address and by domain. Never conclude "no activity" from a keyword search that returned nothing.
• Paginate to exhaustion. If a response carries a continuation token, follow it until there is none. A partial pull manufactures false drops.
• Never conclude a thread's last message from search snippets. Search commonly truncates threads to their first few messages. If any other system suggests activity more recent than the last visible message, fetch the full thread before deciding anything.
• Search by recipient address, not by keyword. A follow-up sent as a reply inside an older thread will not match a company-name search.
• Two checks are required before flagging an owed follow-up as unsent: the complete sent history, and a targeted search restricted to that recipient's domain after the event date. Either alone has produced false positives.
• Note channel blind spots rather than assuming silence. Messages sent over social, SMS, or from a personal address will not appear. Absence of evidence in the mailbox is not evidence of absence, and the gap belongs in the output as a stated limitation.

Recording the verdict

State the evidence checked for every candidate, whether it passed or failed, so the call is auditable by someone who was not there. A verdict without its evidence is an assertion, and assertions are what this gate exists to replace.