feat: validate podcast cost estimation accuracy, document per-token costs, and fix subscription/plan enforcement
Issue #543 — Validate Estimated Cost Accuracy (UI vs Backend)
Backend:
- cost_estimator.py uses pricing catalog (APIProviderPricing) as single source of truth
- All 7 cost components: analysis, research (search+LLM), script, TTS, voice clone, avatar, video
- initialize_default_pricing() runs on every app startup for auto-sync
Frontend cost estimation fixes:
- Added missing analysisCost, scriptCost, voiceCloneCost to PodcastEstimate type
- toPodcastEstimate() now extracts all 7 backend fields (was dropping 3)
- headerCostEst maps analysisCost->Analyze, scriptCost->Write, voiceCloneCost->Produce
- EstimateCard shows 5 chips: Analysis, Research, Script, Voice(TTS+clone), Visuals(avatar+video)
- Chip sum now equals backend total for all configurations
Subscription & plan fixes:
- Removed Stripe re-verification from checkSubscription() (downgrade regression fix #539)
- Added verifyCheckoutRef pattern for reliable mount-time checkout polling
- One-time Stripe sync effect with pending_subscription_change flag for Customer Portal returns
- Free plan limits: stability_calls 3->10, audio_calls 5->10 (supports 2 podcasts)
- Image enforcement uses actual provider (GPT_PROVIDER), not hardcoded Stability
- Billing/pricing pages bypass onboarding check in ProtectedRoute
- Gradient buttons + loading spinner on plan chip in UserBadge
- Added metadata-based Stripe lookup fallback (Issue #538)
Documentation:
- TESTING_GUIDE.md: comprehensive testing instructions for non-technical testers
- Free plan limits, usage tracking, cost estimation formulas
- 10 test cases for UI verification
- Troubleshooting guide
- Quick-reference cost formulas with all default rates
Cleanup: removed legacy ToBeMigrated directory (70+ files, ~22K LOC)
GSC Brainstorm: service, hook, modal, and UI components for blog topic brainstorming
aaf94049da