Release 1.14.0 · Blog AI generation, LAP lead capture, and landing-page Bayesian analytics

Three new product surfaces ship together:

• Blog AI: A Cognis-native blog system (posts, categories, tags, authors, translations) backed by the same AI run orchestrator that powers creatives. Per-post AI runs walk plan → outline → research → draft → final copy → image meta → finalize, then drive parallel translation runs per language. A new "Blog AI config" page at the org and brand level lets you tune voice/tone, banned phrases, reserved variables, and per-step prompt overrides; the post editor surfaces the full AI run timeline alongside the editor in a tabbed view that deep-links via ?tab=.
• LAP (Lead Aggregation Platform): New lap-edge Cloud Run service ingests partial-form / submit / cart-abandon events behind the ecom_domains HTTPS LB (now service_target='lap'). Leads are persisted as lap_leads with deterministic visitor + form linkage and a per-lead status machine (started / partial / completed / recovered). Org admins can drop a folder of HTML/CSS/assets into /org/{id}/lap/sites and the upload action hardens it (MIME allowlist, CSP for HTML, Cache-Control by file shape, 25 MB cap, path normalisation).
• Landing-page analytics + Bayesian winner: Live storefront tracking for landing pages publishes batched events to BigQuery; the new marts_lp.sql model powers a Bayesian winner panel on the variant detail page (probability-of-best per variant + lift CIs). The legacy landing_page_versions table is fully retired in favour of the variant-first model.

Other notable shipping changes: a Cloudflare integration revamp with optional Account ID and clearer upstream error surfacing; a WordPress integration that can sync the new blog system to a connected WP site; ecom guest order-tracking session hardening; storefront cart-attribute / checkout-redirect tightening; and a sweep of new EN/SV translations.