What changed in Tahanic.

Why it mattered: an alert asks someone to act, so a false alert is worse than an incorrect passive number. A manager could be told to respond to low reviews that the page itself said did not exist.

What changed: the low-review alert now evaluates the same 7-day window against the same scope the page displays, and figures are labelled sample, mixed or real so the source of every number is visible. Regression tests cover the scope agreement so the contradiction cannot silently return.

Why it mattered: owners judge whether table cards and QR placements work by these numbers. Miscounted origins make a working channel look dead — or a dead one look alive.

What we deliberately did not do: historical records that were already lost were not reconstructed — fabricating history the system cannot verify would be worse than an honest gap. From the fix onward, both events persist.

Why it mattered: zero reviews does not mean a zero-star average. A fabricated 0.0 tells an owner their reputation is catastrophic when the truth is simply "no data yet" — the difference between panic and patience.

Why it mattered: a silent failure erodes every promise the product makes — reminders that never fired and alerts that never went out looked identical to quiet days. Distinguishing "failed" from "nothing to do" is what makes the reliability of everything else observable.

Assignment makes ownership explicit; notes keep the discussion next to the review instead of in a chat app; manager flags mark what needs senior attention; version history preserves what was drafted, edited and approved — by whom and when.

Why it mattered: publishing to the review platform currently happens outside Tahanic, so a "published" label inside Tahanic would be a claim the product cannot verify. Approved records who accepted the text and when — nothing more, nothing invented.

Why server-side: a draft grounded in client-supplied data could be manipulated or stale; resolving the profile on the server makes every draft traceable to what the business actually configured. Voice learning from approved replies was also improved — it learns from what you accepted, not from what it generated.

Terminology was corrected too: the metric formerly implying publishing speed is now "time to approve" — it measures the approval, which is what actually happens inside Tahanic. Edit counts distinguish drafts used as-is from drafts you rewrote, which is the honest measure of how well the AI knows your voice.

Why gating was removed: filtering who gets asked based on sentiment violates review-platform policies and misrepresents a business's real reputation. Every customer now gets the same ask; unhappy customers can additionally choose a private route that reaches the owner directly.