Software Updates

Katura Deck quality overhaul — editorial design system, language separation, full-bleed charts, editor v2

Katura Deck session 2 — PDF→deck generation via Claude vision (pdfjs client processing, no-hallucination contract, verbatim chart embedding, cue-point notes)

Katura Deck session 1 — model, APIs, editor (sorter/canvas/notes), present mode, share, nav

Yen searchDocs/readDoc/createDoc tools + unify messenger @yen onto main runtime (tools, guardrails, failover, budget)

paragraph-by-paragraph doc translation, Yen daily briefing (9AM ET cron + dashboard card + 11labs listen), voice env wiring, Yen deep-dive doc

page-break indicators, persisted translation view, ElevenLabs read-aloud

make split-view language bars exactly symmetrical (fixed h-8 headers)

Katura Sheets — Univer-powered spreadsheets

• Full Excel-grade engine via @univerjs/presets (Apache-2.0): ~400 formula functions, toolbar, context menus, number formats
• /admin/sheets list page (search, create, delete) mirroring Docs
• /admin/sheets/[id] full-screen editor, client-only (next/dynamic)
• Autosave: Univer workbook snapshot persisted 2s after last command
• Sharing: viewer/editor roles per teammate email (owner-managed)
• Viewers get a read-only workbook (setEditable(false))
• New models KaturaSheet + KaturaSheetShare with hand-written migration
• Nav entry under Content with en/pt/es labels

side-by-side translation view + 24 languages; chore: clean root

• Translation mode is now a full-screen split view (admin sidebar hidden): original doc stays editable on the left, AI translation renders on the right — no more content swapping
• Language list expanded from 8 to all 24 storefront locales, with RTL support (Arabic, Hebrew) on the translation pane
• Refresh button re-translates after edits (saves first, then re-runs)
• Back arrow exits split view; exports still follow the viewed language
• Root cleanup: moved 11 loose files into docs/ (sessions, audits, marketing, reference/copilot); gitignored tsconfig.tsbuildinfo

anchor .vercelignore patterns to repo root

• CRITICAL BUG: 'docs/' (unanchored) was matching ANY directory named
• 'docs' at any depth, not just the root markdown docs/ folder. This
• silently stripped src/app/admin/docs/ and src/app/api/admin/docs/
• (the entire Katura Docs feature) out of every Vercel build — hence
• the 404s, even though the code was committed and 'Compiled successfully'.
• Anchored all patterns with a leading slash (/docs/, /logs/, /tmp/,
• /e2e/, /coverage/, /scripts/...) so they only match at the repo root.

Word import/export, page sizes, print & PDF

• Import .docx/.doc → mammoth converts to HTML (images inlined as base64), creates a new doc — 'Import Word' button on /admin/docs
• Export menu: Word (.docx via html-to-docx with real page dimensions), PDF (browser print engine, pixel-perfect @page CSS), raw HTML
• Exports respect the currently-viewed translation language
• Page size selector (A4 / Letter / Legal): editor canvas resizes to true page dimensions, persisted per doc, drives print + docx output
• Print button with proper @page margins and typography

trim NEXT_PUBLIC_APP_URL to remove trailing newline in reset URLs

• /api/auth/forgot-password/route.ts
• /api/auth/register/route.ts
• /api/admin/team/reset-password/route.ts
• /api/admin/wholesale/leads/[id]/convert/route.ts

Enhanced marketing analytics dashboard with geographic + device breakdowns

• Interactive pie chart + bar chart toggle for country distribution
• Country details expandable list with percentages
• Device type breakdown (desktop/mobile/tablet/other) with visual bars
• Top traffic sources (referer domains) analysis
• Peak hours bar chart showing hourly traffic distribution
• Device detection from User-Agent string
• Referer domain extraction and ranking
• Peak hours calculation (UTC time-based)
• Improved response structure for UI consumption
• Added MARKETING_ANALYTICS_ENHANCEMENTS_2026_06_08.md
• 24 feature ideas for future implementation
• Prioritization roadmap (3 phases)
• Success metrics and technical notes

Google tracking params + pricing update

• Extract and forward gbraid + wbraid (Google iOS app/web remarketing IDs)
• Extract and forward utm_matchtype + utm_device (Google ValueTrack params)
• Incoming UTM params on the /go/ request now WIN over link config defaults (previously link config would override incoming campaign params)
• All ad-platform IDs (gclid, gbraid, wbraid, fbclid, ttclid) forwarded
• Lab grown price: $9,000 → $7,000 (both hero strip and value proof card)
• Added note: 'stone only · casting & setting not included'
• Applied to both the mobile hero section and the desktop value proof cards

add gclid (Google Ads click ID) end-to-end attribution

• Click counter stuck at 94: Google Ads 'Final URL' was pointing directly to /engagement/consultation/777?... instead of /go/google. Direct traffic bypasses the /go/ route, so MarketingSmartLinkClick rows are never created. FIX: Change the Final URL in Google Ads to https://katura1999.com/go/google
• gclid not captured anywhere: use-attribution.ts hook read utm_*, fbclid, smart_link, sid from URL params, but completely ignored gclid. All Google Ads click IDs were silently dropped — no CRM attribution, no conversion tracking closure.
• use-attribution.ts: Added gclid to AttributionFields interface, initial state, fromUrl reading (params.get('gclid')), and merged result — now persisted in localStorage and included in every form submission
• offer-consultation-page.tsx: Pass attribution.gclid in all 3 form submit bodies (inquiry form, full booking, Saturday request)
• availability-request/route.ts: Added gclid to Zod schema; stores in EngagementLead notes as 'GCLID: xxx' line pending DB migration
• consultation/route.ts: Added gclid to Zod schema; passes to bookConsultation()
• booking.ts: Accept gclid in BookConsultationInput type; writes gclid to Consultation.gclid column (after migration)
• prisma/schema/engagement.prisma: Added gclid column to both Consultation and EngagementLead models
• prisma/migrations/add_gclid_attribution.sql: Migration to add gclid column to Consultation and EngagementLead tables in production DB

split inquiry form into Name + Email (required) + Phone (optional)

• Heading: 'GET CURRENT 7.77 AVAILABILITY' → 'INQUIRE ABOUT AVAILABILITY'
• Subtitle: 'Concierge will send current shapes, certification, and pricing.'
• Field 'First name' → 'Name' (accepts any name format)
• Replaced single 'Phone or email' field with:
• Email address * (required)
• Phone (optional)
• Applied to both the sticky sidebar copy and the mobile/hero copy
• Both forms now share the same leadName/leadEmail/leadPhone state
• Section border changed amber → blue (matching pricing strip)
• email is now required (z.string().email()) - no more optional fallback
• Removed .refine() either-or check
• CRM lookup now always uses email
• CRM create/update uses v.email directly (no phantom placeholder)

simplify attribution storage to avoid Prisma schema issues

• Store attribution data as plain text in notes field
• Build attribution string without creating objects with schema fields
• This avoids Prisma trying to validate column names that don't exist in db
• Still captures all attribution data until db migration is applied

remove debug log - availability-request now working

add debug log to verify deployment

availability-request API database schema mismatch

• Removed attribution fields (offerVariant, landingPageVariant, utm*, meta*) from EngagementLead.create() as these columns don't exist in production db
• Attribution data is now stored in notes field as JSON reference for now
• TODO: Apply database migration to add these columns
• Changed 'Inquire about availability' section border from yellow to blue to match the LAB GROWN DIAMONDS pricing section

improve error handling in availability-request API

• Added try-catch blocks for CRM operations (continues on failure)
• Added try-catch for EngagementLead creation with detailed logging
• Return error details in 500 response for easier debugging
• Better console logging for troubleshooting production issues

CRM submission + notification settings + heart shape + blue pricing

• availability-request API now uses plain prisma directly instead of tenantPrisma (eliminates RLS/extension layer that was causing silent failures); tenantId is still explicitly set in all create() calls
• Added leadFirstError state + error display in inquiry form — users now see a message if the API call fails instead of silent no-op
• Built full notification recipients settings page replacing the placeholder stub
• Add/remove email addresses per trigger type via live UI
• Two trigger types: 'Availability Request Form' and 'Full Consultation Scheduling'
• Stored in SaasTenant.brandSettings.notificationRecipients (no migration)
• New API: GET/PATCH /api/admin/crm/notification-settings
• availability-request route now reads dynamic recipient list
• consultation/route.ts now fires team notification email post-booking using the same dynamic recipient list
• Added Heart shape (heart.png) to the shape picker on /777 page
• Changed LAB GROWN price box from amber/yellow to light blue (bg-blue-50 border-blue-200 text-blue-800)

rebrand currency selector to 'Select Country' with native language display names

• Label: 'Currency' → 'Select Country'
• Format: 'US Dollar' → 'United States - USD'
• Format: 'Brazilian Real' → 'Brasil - BRL'
• Format: 'Japanese Yen' → '日本 (Japan) - JPY'
• Format: 'Israeli Shekel' → 'ישראל (Israel) - ILS'
• Format: 'Vietnamese Đồng' → 'Việt Nam (Vietnam) - VND'
• Format: 'Chinese Yuan' → '中国 (China) - CNY'

remove cpus:4 cap + reduce NODE_OPTIONS heap to prevent OOM on Standard

• Standard (8GB, 4 vCPU): cold build ~6-8 min, no OOM, warm builds 3-4 min
• Turbo (60GB, 30 vCPU): cold build ~3-5 min, warm builds 2-3 min

delete .npmrc throttle (was limiting Turbo to 1/30 cores) + remove redundant pnpm install from vercel-build

• [aff0ee0a] Route slug conflict [invoiceId] vs [id] — FIXED (was causing 208 runtime errors)
• [f9ed212a] Smart-links sequential timeouts — FIXED (parallelized via Promise.allSettled)
• [bb68d13b→this] Build throttle — FIXED (removed .npmrc, restored defaults for Turbo)

optimize for memory-constrained Vercel builds to fix OOM errors

• Add .npmrc with pnpm parallelization limits: fetch-parallel-limit=3, build-parallel-limit=2, workers=1 Reduces concurrent processes during install to prevent OOM on standard Vercel containers (~3GB RAM)
• Update vercel-build script to explicitly use --frozen-lockfile --no-optional --prefer-offline Ensures faster, leaner install without re-resolving or fetching optional deps

resolve dynamic route slug conflict ([invoiceId] vs [id]) causing INTERNAL_FUNCTION_INVOCATION_TIMEOUT

• Move check-stripe-payment + sync-stripe from [invoiceId] into [id]
• Delete orphaned [invoiceId] directory (sibling conflict crashed Next.js router)
• Delete orphaned literal \[locale\] directory (backslash-bracket name != dynamic route)
• Update params type from { invoiceId } to { id: invoiceId } in both moved routes