2026-05-29 · Docs
Search page rewritten — honest about being ILIKE, not Algolia
- Audit caught the third code-vs-copy gap. Search marketing claimed Algolia-style typo tolerance + synonyms + ranking-tuning UI. Code is Postgres `ILIKE '%term%'` across 4 fixed columns (`profiles.email/name`, `support_tickets.subject`, `kb_articles.title`, `events.event_type`) with hardcoded `score=1`.
- Rewrote the product page to claim only what code does: 4-table federation, type-filter facet, zero-sync (data already in tenant), case-insensitive substring match.
- Explicit non-features called out: no typo tolerance, no synonyms, no ranking tuning, no Custom Ranking, no Personalization, no A/B testing, no edge-eval latency.
- Algolia comparison rewritten honestly — Algolia wins on every feature row; Sendora wins on the zero-sync + bundled-cost angle. 5 FAQ entries explicitly enumerate the missing features.
- Dashboard tagline updated to match: "Substring match across profiles, support tickets, KB articles, and events — one query, zero sync code."
- The win remains real for the right use case: ops command palette for in-tenant data where Postgres ILIKE is sufficient and per-record Algolia bills are wasted.