Build: E-commerce Storefront
Toggle features and choose options to customize your spec
Preset
Authentication Methods*
Access Control*
Multi-factor Authentication
Tradeoffs
Each provider requires an OAuth app registration and key rotation policy
Requires IdP partnership and XML-based protocol handling; significant integration work
Permission checks must be applied consistently across every data access path
Billing Model*
Payment Processor*
Tradeoffs
Requires handling trial periods, dunning, proration, and cancellation flows
Must instrument every billable action and send metered events to billing provider
Less customizable checkout; Paddle acts as legal seller so you avoid VAT registration
Search Approach*
Search Scope*
Tradeoffs
Requires FTS index maintenance; adds write-time overhead
Embedding generation adds latency and API cost per indexed document
Results must be unified and ranked across disparate data models
Delivery Method*
User Control
Tradeoffs
Requires persistent connection infrastructure (e.g. Redis pub/sub, WebSocket server)
Higher server request volume; notifications may lag by poll interval
Requires APNs/FCM credentials and certificate management
Significantly more complex preference storage and UI
Storage Backend*
File Processing
Tradeoffs
Not horizontally scalable; lost on server replacement without backup
Higher monthly cost; requires cache invalidation strategy
Upload latency increases; requires AV service integration
Delivery Provider*
Deliverability Setup*
Templating Approach*
Tradeoffs
Vendor cost scales with volume; deliverability expertise comes included
Low per-email cost but you own deliverability operations (reputation, bounces, suppression)
Two sending configurations and domains to maintain — worth it for deliverability isolation
Queue Backend*
Required Capabilities*
Failure & Durability*
Tradeoffs
Primary DB absorbs queue write load; row-level locks contend with application queries
Enqueue happens outside DB transaction — jobs can run for state that was rolled back
Additional table, polling worker, and idempotency discipline — the payoff is no duplicated side effects
Rate Limit Algorithm*
What to Limit By*
Abuse Prevention Layer
Response Behavior*
Tradeoffs
False positives behind corporate NATs; attackers bypass with rotating proxies
Noisy-neighbor protection — one tenant cannot starve others
Allows bursts but requires a per-identity bucket state in Redis — higher memory footprint
Meaningful latency cost at the edge if the WAF is geographically distant from users
Summary
8 of 9 features enabled
Effort Estimate
10+ weeks
8 enabled features
Key Decisions
User System & Auth
Will this product be sold to businesses (B2B)?
If yes
Add SAML/SSO and RBAC. Enterprise procurement often requires both.
If no
Email + password plus one OAuth option covers 95% of consumer use cases.
User System & Auth
Is this a security-sensitive application?
If yes
Enable TOTP MFA. Consider making it mandatory for privileged users.
If no
MFA is optional — offer it but do not require it to reduce friction.
User System & Auth
Email+password, passwordless, or SSO-only?
If yes
Passwordless (magic links or passkeys) eliminates password reset tickets and credential stuffing risk.
If no
Keep email+password as a universal fallback — OAuth outages should not lock users out.
User System & Auth
Do you need social providers (Google, GitHub, Apple)?
If yes
Add Google for B2C breadth; add GitHub for developer tools; add Apple only if you ship iOS (App Store requires it when you offer other social login).
If no
Skip social OAuth and avoid the app registration / key rotation overhead.
User System & Auth
Do you need SCIM provisioning?
If yes
Add SCIM alongside SAML — enterprise IT uses it to auto-provision/deprovision employees and map group membership to roles.
If no
Manual invite flows are fine until your first enterprise customer asks for SCIM in a security review.
User System & Auth
Should MFA be required, optional, or risk-based?
If yes
Risk-based (step up MFA on new device, new IP, or sensitive actions) gives security without friction on every login.
If no
Offer MFA as optional first; require it only for admins or on privileged actions.
User System & Auth
Which MFA factors will you support (TOTP, SMS, WebAuthn/passkeys, hardware keys)?
If yes
Prefer WebAuthn/passkeys and TOTP. Avoid SMS as a primary factor — SIM swapping is a real threat.
If no
TOTP alone (Google Authenticator, Authy) covers the vast majority of users with minimal implementation cost.
User System & Auth
Do you need device fingerprinting or trusted-device flows?
If yes
Remember trusted devices for 30 days to skip MFA; challenge on new device or changed fingerprint.
If no
Re-prompt MFA on every login — simpler and safer for low-volume or highly sensitive apps.
User System & Auth
Offer passkey-only sign-in?
If yes
Passkeys eliminate passwords entirely — use WebAuthn with platform authenticators. Still keep an email recovery path for lost devices.
If no
Offer passkeys as an optional second factor; users without compatible devices keep using passwords.
User System & Auth
Support staff impersonation of user accounts?
If yes
Add an impersonation flow that logs both the staff identity and the target user, with a visible banner in the impersonated session.
If no
Skip impersonation — instead build admin-side read views and support tooling that do not require acting as the user.
User System & Auth
Captcha or bot detection on signup?
If yes
Add hCaptcha or Cloudflare Turnstile on signup and password reset — invisible challenges avoid user friction.
If no
Skip captcha for internal tools or invite-only products where bot signups are not a realistic threat.
User System & Auth
Use lockout or rate-limit throttling for credential stuffing?
If yes
Exponential rate limits per IP and per account — lockouts create support tickets and denial-of-service vectors via targeted lockout.
If no
If account takeover risk is low, a simple fixed rate limit (e.g., 10 attempts per 15 min) is sufficient.
User System & Auth
Allow multiple concurrent sessions per user?
If yes
Show active sessions in account settings with a revoke button — expected behavior for any multi-device product.
If no
Single-session apps (banking, compliance) should terminate old sessions on new login.
Payments & Billing
Is the revenue model recurring (SaaS)?
If yes
Choose subscription billing. Evaluate usage-based if pricing scales with consumption.
If no
One-time purchase is far simpler. Consider Stripe Checkout for a no-code option.
Payments & Billing
Is global VAT/sales tax compliance a concern?
If yes
Use Paddle as merchant of record — they handle tax across jurisdictions.
If no
Stripe gives more control; integrate TaxJar or Stripe Tax if needed later.
Payments & Billing
Do you want to offer a free trial without requiring a card upfront?
If yes
Use reverse trials — free-tier access with a prompt to add a card at the end. Higher signup conversion but lower trial-to-paid conversion.
If no
Card-required trials filter out tire-kickers and produce 2–3x higher trial-to-paid rates. Stripe supports both via checkout.
Payments & Billing
Do you sell to enterprise customers with procurement processes (POs, net-30 terms)?
If yes
Support invoicing workflows (Stripe Invoicing or manual PDF invoices via finance). Self-serve credit-card checkout is insufficient at that ACV.
If no
Credit-card-only is simpler and covers all SMB/prosumer use cases.
Payments & Billing
Do you sell in markets where customers transact in non-USD currencies?
If yes
Enable multi-currency pricing in Stripe or Paddle. Price in local currency — EU/UK customers strongly prefer EUR/GBP over USD conversions.
If no
USD-only is simpler; add currencies only when a market demands it.
Payments & Billing
Will you close large B2B deals that need ACH or wire transfer (>$5k)?
If yes
Enable Stripe ACH Credit Transfer or wire instructions on invoices. Credit-card fees on large invoices are prohibitive.
If no
Card-only is fine for SMB and prosumer ticket sizes.
Payments & Billing
Do you have EU customers (SCA/3DS compliance required)?
If yes
Use Stripe Payment Intents (handles 3DS authentication automatically) or Paddle. Do not use raw Charges API — it predates SCA.
If no
Still use modern Payment Intents API — SCA will apply to US processors eventually.
Payments & Billing
Is self-serve cancellation acceptable, or do you need "contact us to cancel"?
If yes
Self-serve cancellation via Stripe Customer Portal — legally required in California (FTC Click-to-Cancel) for many subscriptions.
If no
Contact-us friction increases short-term retention but damages NPS and is increasingly regulated. Think twice.
Payments & Billing
Do marketing or sales teams need to issue coupons and discounts?
If yes
Use Stripe Coupons / Paddle Discounts. Support percentage and fixed-amount discounts with expiry and redemption limits.
If no
Skip — discount UX adds complexity and is rarely needed outside marketing-led motions.
Payments & Billing
Do users frequently upgrade/downgrade mid-cycle?
If yes
Enable proration in Stripe (proration_behavior: create_prorations). Immediate upgrade + end-of-period downgrade is the customer-friendly pattern.
If no
Wait-until-renewal plan changes are simpler; skip proration logic.
Payments & Billing
Do you have >1000 paying customers or expect significant failed-payment volume?
If yes
Enable Stripe Smart Retries (free) plus a custom dunning email sequence (day 0, 3, 7, 14). Recovers 30–50% of failed payments.
If no
Default Stripe retries are enough; add custom dunning once failed payments become a meaningful revenue leak.
Payments & Billing
Should customers manage their own billing (payment methods, invoices, plan changes)?
If yes
Use Stripe Customer Portal or Paddle Retain — pre-built UI, handles tax/invoices/cancellation. Massive support-ticket reducer.
If no
Build a minimal billing page and route the rest to support — only viable at low customer counts.
Payments & Billing
Do you connect buyers and sellers and need to split payments (marketplace)?
If yes
Use Stripe Connect (Standard or Express accounts). Do not build split payments yourself — tax forms, KYC, and payouts are legal minefields.
If no
Standard direct charges are simpler and correct for first-party sales.
Payments & Billing
Do you need to issue refunds regularly with reason tracking and approval flows?
If yes
Build an internal refund tool that captures reason, links to audit log, and uses Stripe Refunds API. Required for support scale and compliance.
If no
Manual refunds through the Stripe dashboard are fine until volume demands tooling.
Payments & Billing
Are you tempted to store raw card numbers to avoid re-entry (PCI Level 1 scope)?
If yes
Do not. Use Stripe Payment Methods or Paddle saved cards — tokenized references keep you out of PCI DSS Level 1 scope. The compliance overhead is massive.
If no
Good — always tokenize. Stripe Elements, Paddle Checkout, or hosted checkout keeps card data off your servers entirely.
Search
Do users search by meaning, not just keywords?
If yes
Invest in semantic/vector search. Start with full-text and migrate.
If no
Full-text search covers keyword use cases at a fraction of the cost.
Search
DB full-text or a dedicated engine (Elasticsearch, Typesense, Algolia)?
If yes
Reach for a dedicated engine when DB FTS can't meet latency or ranking needs. Algolia for hosted DX, Typesense/Meilisearch for self-hosted.
If no
Start with SQLite FTS5 or Postgres tsvector — no extra service to run.
Search
Do users need faceted filtering (refine by category, tag, date range)?
If yes
Use a dedicated engine — faceting across millions of rows in Postgres FTS gets slow fast. Typesense and Algolia do this natively.
If no
Keyword-only over FTS is sufficient; add facets later.
Search
Do you need typo tolerance / fuzzy matching on short queries?
If yes
A dedicated engine (Typesense, Algolia, Meilisearch) gives this out of the box. DB FTS typo tolerance is weak.
If no
Strict matching is fine for structured queries and technical users.
Search
Will you support multiple languages with proper stemming?
If yes
Pick an engine with per-locale analyzers (Elasticsearch, Meilisearch). Postgres tsvector ships only a handful of language dictionaries.
If no
English-only tsvector or FTS5 is plenty.
Search
Do users expect personalized ranking (their clicks influence their results)?
If yes
Algolia or a custom scoring layer on top of an engine — personalization needs per-user signals, not just index weights.
If no
Use global relevance scoring (BM25) — simpler and predictable.
Search
Is autocomplete / instant search (as-you-type) part of the UX?
If yes
Pick Typesense, Algolia, or Meilisearch — all tuned for sub-50ms responses. Postgres FTS will feel sluggish here.
If no
Submit-driven search works against any backend.
Search
Do you need search analytics (popular queries, zero-result queries)?
If yes
Log queries + result counts separately; feed into your analytics pipeline. Algolia and Meilisearch expose this natively.
If no
Skip until product teams ask for it.
Search
Must results respect per-user permissions (ACL-aware)?
If yes
Index ACL identifiers alongside documents and filter at query time. Global cross-entity search is especially risky — validate before shipping.
If no
A flat index is simpler; use when all users see the same corpus.
Search
Do customers need custom synonyms and stop-words (domain vocabulary)?
If yes
A dedicated engine with synonym dictionaries (Algolia, Elasticsearch) — editable without re-indexing.
If no
Default analyzers work for general-purpose text.
Search
Do you need rule-based boosting (featured or sponsored results)?
If yes
Algolia has a dashboard for this; Elasticsearch supports function_score. Don't hand-roll on top of tsvector.
If no
Pure relevance ranking is cleaner.
Search
Is mobile bandwidth a constraint for search-as-you-type?
If yes
Debounce aggressively (~300ms), return tiny payloads, and consider a provider that supports partial-result responses.
If no
Desktop-grade instant search is fine; no special tuning needed.
Search
Must newly created content be searchable within seconds (near-real-time)?
If yes
Index on write into an engine with NRT support (Elasticsearch, Typesense). Budget for higher write amplification.
If no
Batch reindex every few minutes via background job — simpler and cheaper.
Search
Do users search across multiple indices / entity types in one query (federated)?
If yes
Global scope is required. Use Algolia multi-index search or aggregate in app code — plan ranking carefully.
If no
Per-resource search is simpler and faster.
Search
Do users need "did you mean" spell correction for empty-result queries?
If yes
Meilisearch and Algolia provide this out of the box. Ties naturally with zero-result analytics.
If no
Show filters and suggested queries instead; simpler to build.
Search
Should users be able to save searches or get alerts on new matches?
If yes
Store the query, schedule a job to re-run it, and diff results. Pair with the notifications module for delivery.
If no
Manual re-runs cover most use cases; skip the infra.
Notifications
Do users need to know about events immediately?
If yes
Use real-time delivery. Budget for WebSocket/SSE infrastructure.
If no
Polling is simpler and cheaper — pick an interval that matches your SLA.
Notifications
Is this a mobile-first product?
If yes
Add mobile push (APNs/FCM). Plan for certificate rotation.
If no
Skip mobile push; web notifications or email cover most cases.
Notifications
Do users often go offline and still need to receive notifications?
If yes
Add email fallback driven by a "last seen" timestamp. Send email only when the user has been offline >5 minutes to avoid spam.
If no
In-app real-time or polling is enough — email adds unsubscribe risk without upside.
Notifications
Will you send multiple notification types with different urgency levels?
If yes
Adopt per-type granularity so users can mute digests without muting security alerts. Define the taxonomy up front.
If no
A single all-or-nothing toggle is enough and avoids preference-UI bloat.
Notifications
Is notification volume high enough that users will complain about floods?
If yes
Batch similar notifications in a rolling window (e.g., "Alice and 12 others liked your post"). Entity + time-window grouping is the standard pattern.
If no
Deliver individually — grouping adds UI complexity that is not worth it at low volume.
Notifications
Are any notifications security-critical or financially sensitive?
If yes
Require at-least-once delivery with persistent storage and retry. Pair with email fallback for the highest-urgency classes.
If no
Best-effort (fire-and-forget) is fine for social and informational notifications.
Notifications
Do users have reasonable expectations of quiet hours / do-not-disturb?
If yes
Add a per-user quiet-hours window and a priority flag that lets critical alerts bypass it. Requires per-type granularity to know what is critical.
If no
Skip quiet hours — the added preference surface is not worth it for low-volume products.
Notifications
Do users need to mark notifications as read / track what they have seen?
If yes
Persist notifications with a read state. Consider cross-device read-state sync via a server-side timestamp or event log.
If no
Ephemeral toast-style notifications are simpler and appropriate for low-stakes alerts.
Notifications
Do you need delivery analytics (delivered, opened, clicked)?
If yes
Use a provider like Knock or Novu that tracks delivery funnel by channel. Essential if notifications drive revenue or retention KPIs.
If no
Skip the instrumentation — simple success/failure logs are enough.
Notifications
Do users have multiple devices (web + mobile + desktop)?
If yes
Sync read-state across devices via a server-side last-read timestamp. Without this, users see the same notification over and over.
If no
Device-local read state is simpler and avoids backend round-trips.
Notifications
Do you need SMS as a notification channel?
If yes
Use Twilio or Messagebird. Reserve SMS for high-priority alerts only — it is expensive per message and users churn fast on SMS spam.
If no
Skip SMS — push and email cover 99% of use cases at a fraction of the cost.
Notifications
Are some notifications purely in-app / ephemeral (toast-style)?
If yes
Deliver via SSE/WebSocket without persisting — no storage, no retry, no read-state UI. Good for "Saved", "Uploaded" feedback.
If no
All notifications go through the durable pipeline with read-state tracking.
Notifications
Do you need localized notification content (multi-language)?
If yes
Store notification templates with i18n keys and render per-recipient locale at send time. Avoid pre-rendered strings in the event.
If no
Hardcoded English strings are fine until you have non-English users.
Notifications
Do you have more than 5 notification types or expect to add more regularly?
If yes
Invest in a template engine (Knock, MJML, or Handlebars) with versioned templates rather than hardcoded message strings.
If no
Hardcoded message strings in the sender are simpler and fine for a small stable set.
Notifications
Do product/marketing teams need to preview and test-send notifications before release?
If yes
Build an internal preview tool with a test-recipient flag. Prevents production embarrassments.
If no
Skip — engineers can validate in staging until non-engineers start authoring templates.
Notifications
Are there urgent notifications that must bypass quiet hours (security alerts, outages)?
If yes
Add a priority flag on notifications and skip DND for priority=urgent. Document this behavior so users expect it.
If no
Quiet hours apply uniformly — simpler and avoids abuse of the override.
File Storage
Will the app run on more than one server?
If yes
Object storage is required — local disk breaks horizontal scaling.
If no
Local disk works for prototypes; switch to S3-compatible before scaling.
File Storage
Do users upload files directly via presigned URLs?
If yes
Issue short-lived (5–15 min) presigned PUT URLs so uploads skip your server — cheaper, faster, and sidesteps body-size limits.
If no
Proxying uploads through your API is simpler but caps throughput at your server bandwidth — fine for small files only.
File Storage
Do you need chunked/multipart uploads for large files?
If yes
Use S3 multipart upload for anything over ~100MB — single-PUT uploads fail expensively on flaky networks.
If no
For small files (photos, documents under 10MB), a single PUT is simpler and sufficient.
File Storage
Resumable uploads required?
If yes
Use tus.io protocol or S3 multipart with client-side state so users can resume after network drops — critical for mobile video uploads.
If no
Small-file workflows can safely require restart-on-failure.
File Storage
Auto image/video transcoding or thumbnail generation?
If yes
Pipe uploads to a transcoding service (Cloudinary, Mux, Lambda@Edge with sharp/ffmpeg) — async, never block upload completion.
If no
Skip transcoding for document-heavy workflows where files are not consumed as media.
File Storage
Access model: files public, private, or signed-URL access?
If yes
Private by default with short-lived signed URLs issued per authenticated request — safest for user-generated content.
If no
Fully public buckets are fine for static assets (logos, public images); never for user data.
File Storage
Do you need client-side encryption at rest?
If yes
Encrypt on the client before upload with customer-managed keys — required for strict compliance (healthcare, legal).
If no
Server-side encryption (SSE-S3, SSE-KMS) is enough for most workloads and is transparent to clients.
File Storage
Store file metadata in DB or only in object storage?
If yes
Keep a files table in your DB with upload metadata, owner, status, and a content hash — drives permissions and search.
If no
Object storage alone is sufficient only for truly anonymous, throwaway uploads.
File Storage
Do you need cold-storage tiering for old files?
If yes
Use S3 Intelligent-Tiering or lifecycle rules to move files untouched for 30/90 days to Glacier — dramatic cost savings on archival data.
If no
Skip tiering for small storage footprints or hot-access workloads where retrieval latency matters.
File Storage
Track storage usage per user/tenant for quotas/billing?
If yes
Aggregate usage in a separate table, updated on upload/delete events — do not scan object storage for totals in real time.
If no
Skip metering for internal tools or flat-rate products without storage-based billing.
File Storage
CDN fronting for downloads?
If yes
Put Cloudflare or CloudFront in front of downloads — cuts global latency 5–10x and offloads bandwidth cost. Use signed CDN URLs for private content.
If no
Direct object-storage serving is fine when users are regional and file access is infrequent.
File Storage
Deduplicate identical uploads (content-addressed)?
If yes
Hash files on upload and store by content hash — saves storage on duplicate assets but complicates deletion (reference counting).
If no
Skip dedup for user-facing products where each upload is semantically distinct regardless of bytes.
File Storage
Strip EXIF/metadata from uploaded images for privacy?
If yes
Strip GPS and camera metadata on upload — critical for public-facing photos where location leakage is a real privacy risk.
If no
Keep EXIF for creative-workflow tools (photography, mapping) where metadata is a feature, not a leak.
File Storage
Retain prior versions of files (file versioning)?
If yes
Enable S3 object versioning and track versions in your files table — required for collaborative document editing and compliance workflows.
If no
Overwrite-in-place is simpler and sufficient for user-managed assets where history is not valuable.
File Storage
Soft-delete with trash/recycle bin?
If yes
Mark files deleted in DB but retain objects for 30 days — dramatically reduces "help, I lost my file" support tickets.
If no
Hard delete is appropriate for compliance-driven retention where files must be gone when requested.
Transactional Email
Will you ever send marketing email (newsletters, promotions) from the same brand?
If yes
Plan separate transactional and marketing streams from the start.
If no
A single stream is simpler; split later if you add marketing email.
Transactional Email
Do you expect to send >100k emails/month in year one?
If yes
Evaluate SES or Postmark pricing carefully; negotiate volume discounts.
If no
Pick the best developer experience (Resend / Postmark) — the price delta is rounding at low volume.
Transactional Email
Should you use a managed provider (SendGrid, Postmark, SES) or self-host?
If yes
Use a managed provider — Postmark for deliverability, Resend for DX, SES for cost at volume. Self-hosting is never worth it for transactional.
If no
Only self-host for regulated environments with egress constraints; expect months of reputation work.
Transactional Email
Is a dedicated IP justified for your volume?
If yes
Above ~100k emails/month, request a dedicated IP and budget a 2–4 week warm-up. Below that, shared pools from Postmark/SendGrid are cleaner.
If no
Stay on the shared pool — reputation is managed for you.
Transactional Email
Should mail send from an isolated subdomain (mail.yourdomain.com)?
If yes
Standard practice — protects root-domain reputation from email mistakes and makes DNS records easier to manage.
If no
Only sending from root if you have no other choice; keep SPF/DKIM alignment tight.
Transactional Email
Is DMARC enforcement (p=reject or p=quarantine) required?
If yes
Start with p=none for reporting, then ramp to quarantine and reject once SPF+DKIM alignment is verified across all senders.
If no
Inbox providers increasingly require DMARC — plan to enforce within 6 months anyway.
Transactional Email
Do user replies to transactional email need to drive app actions (reply-to-comment)?
If yes
Use a provider with inbound reply parsing (Postmark, SendGrid Inbound Parse) and a dedicated Reply-To subdomain with MX records.
If no
Set Reply-To to a monitored support inbox or no-reply address.
Transactional Email
Does marketing/CX need to edit templates without a code deploy?
If yes
Use provider-hosted templates (SendGrid Dynamic Templates, Postmark) or a notification platform (Knock, Courier). Keep security emails in code.
If no
Code-owned templates (React Email, MJML) are reviewable and version-controlled.
Transactional Email
Do templates need per-recipient personalization beyond name / link?
If yes
Use a templating engine with merge fields (Handlebars, Liquid). Provider templates handle this well; React Email makes it trivial in code.
If no
Static templates with a few variables are fine — don't over-engineer.
Transactional Email
Do you need template versioning with rollback?
If yes
Code templates get this from git for free. For provider templates, pick one with built-in versioning (Postmark) or snapshot before edits.
If no
Direct edits are fine for low-stakes messages.
Transactional Email
Do you need to send localized email content per recipient?
If yes
Either one template per locale (simple, duplicated) or a single template with i18n key lookups. Store recipient locale on the user record.
If no
English-only ships faster; add locales when revenue justifies it.
Transactional Email
Do you need open and click tracking for product email?
If yes
All major providers offer it as a toggle. Useful for onboarding email analytics — but disclose tracking in your privacy policy.
If no
Disable trackers on security-sensitive email (password resets) regardless — tracking pixels in those emails look phishy.
Transactional Email
Are bounce and complaint webhooks processed to suppress bad addresses?
If yes
Non-negotiable at any real volume. Subscribe to provider webhooks and maintain a suppression table checked before every send.
If no
You will tank your sender reputation within weeks — this is not optional.
Transactional Email
Do you need to schedule sends for a future time?
If yes
Most managed providers support scheduled sends natively; otherwise enqueue to a delayed job queue (BullMQ, SQS with delay).
If no
Send immediately from the triggering event — simpler.
Transactional Email
Do you send high-fan-out batches (announcement to all users at once)?
If yes
Use the provider's batch send API (SendGrid v3, Postmark batch). Chunk to stay under per-call limits and spread over minutes to avoid throttling.
If no
One-at-a-time calls via your background queue are simpler.
Transactional Email
Can end-users (white-label customers) customize email content?
If yes
Sandbox the template language (no arbitrary code), validate on save, and preview before activating. Use a notification platform if this is core.
If no
Keep templates locked down — far fewer support tickets.
Transactional Email
Do you need an internal preview / test-send surface for QA?
If yes
Build an admin route that lists all templates with sample data. Pair with Mailpit/Mailhog in dev to catch rendering bugs before prod.
If no
You'll hear about broken templates from customers — not recommended.
Transactional Email
Do you need a single unsubscribe list shared across product surfaces?
If yes
Centralize in your user record or a notification platform — users unsubscribing from any email should stop all non-critical mail.
If no
Per-stream unsubscribes create support tickets; avoid if at all possible.
Background Jobs & Queues
Does any job charge money, send external messages, or call a paid API?
If yes
Use transactional outbox and make handlers idempotent. Store an idempotency key on the job.
If no
Standard retry + DLQ is sufficient.
Background Jobs & Queues
Do you already run Redis or a broker?
If yes
Use it for jobs — a second persistence dependency is rarely justified.
If no
Start with a database-backed queue; migrate only when volume demands it.
Background Jobs & Queues
Do you need scheduled/cron jobs in addition to on-demand enqueues?
If yes
Enqueue from a single scheduler process (not per-worker cron) to avoid duplicates in a horizontally-scaled deployment.
If no
Pure on-demand enqueue is simpler — add scheduled capability only when you actually have recurring jobs.
Background Jobs & Queues
Is at-least-once delivery sufficient, or do you need exactly-once?
If yes
Exactly-once requires transactional outbox plus idempotent handlers — no library gives it to you for free.
If no
At-least-once with idempotent handlers is the pragmatic production default.
Background Jobs & Queues
Do you have mixed-priority workloads (user-visible vs batch)?
If yes
Use at least two queues (default, bulk) with separate worker pools so a long batch job never starves user-triggered work.
If no
A single queue is simpler and fine for homogeneous workloads.
Background Jobs & Queues
Are your handlers idempotent by contract?
If yes
Aggressive retries are safe — store an idempotency key per job and dedupe on handler entry.
If no
Lean on transactional outbox and accept retries will sometimes double-invoke side effects unless you add keys.
Background Jobs & Queues
Do you need DAG pipelines (jobs that spawn dependent jobs)?
If yes
Use a durable workflow engine (Temporal, Inngest, BullMQ Flows) — rolling your own DAG orchestration is a year-long tarpit.
If no
Flat enqueue is simpler and covers the majority of use cases.
Background Jobs & Queues
Do you need a dead-letter queue for failures?
If yes
Any production queue needs a DLQ with alerting on depth growth — silent job failure is a common outage source.
If no
Skip only for best-effort one-off jobs where losing the job is acceptable.
Background Jobs & Queues
Do you need per-tenant queue isolation for noisy neighbors?
If yes
Shard queues by tenant or add per-tenant concurrency caps so one customer bursting to 10k jobs does not stall everyone else.
If no
A shared queue is fine in single-tenant or low-variance workloads.
Background Jobs & Queues
Do long-running jobs need to be cancellable mid-run?
If yes
Pass a cancellation token through the handler and checkpoint progress so cancellation is responsive without data loss.
If no
If jobs complete quickly, retry-on-failure is simpler than implementing graceful cancellation.
Background Jobs & Queues
Do you need per-job-type retry and backoff configuration?
If yes
Different failure modes need different backoff — network errors retry fast, rate-limit errors retry slow. Configure per job class.
If no
A single global retry policy (5 attempts, exponential backoff) is the pragmatic default.
Background Jobs & Queues
Are jobs CPU-bound (heavy compute) or IO-bound (external calls)?
If yes
CPU-bound: use a worker pool sized to core count. Avoid async in the same process — it will not help and may hurt.
If no
IO-bound: use async/concurrent workers to maximize throughput on waiting time.
Background Jobs & Queues
Do you need queue-depth and worker-lag observability?
If yes
Emit per-queue depth, processing latency, retry count, and DLQ size to Prometheus/Datadog — and alert on them.
If no
The built-in queue dashboard (Sidekiq Web, BullMQ Board) is enough for small teams.
Background Jobs & Queues
Do you need to persist job results for later retrieval?
If yes
Store results in a separate results table keyed by job ID — clients poll or receive webhook/SSE when done.
If no
Fire-and-forget jobs are simpler; only persist results when a user UI depends on them.
Background Jobs & Queues
Do job payloads contain sensitive data?
If yes
Encrypt payloads at rest (envelope encryption with KMS) — queue storage is usually less hardened than your primary DB.
If no
Plaintext payloads are fine for internal, non-PII work.
Background Jobs & Queues
Should similar jobs be batched for efficiency?
If yes
Coalesce jobs (e.g., "send digest for user X") within a short window — a single batch handler beats N individual invocations for I/O.
If no
Per-job execution is simpler to reason about and debug.
Background Jobs & Queues
Do you need an admin UI to list, retry, and cancel jobs?
If yes
Mount the queue library dashboard (Sidekiq Web, Oban Web, BullMQ Board) behind admin auth — zero-effort ops leverage.
If no
CLI tools and logs are enough for a small team; add UI when non-engineers need to investigate job failures.
Rate Limiting & Abuse Prevention
Do you have unauthenticated endpoints (signup, login, public API)?
If yes
Add per-IP limits on those endpoints plus CAPTCHA on threshold. Assume credential-stuffing is attempted on day one.
If no
Per-user limits on authenticated APIs are sufficient.
Rate Limiting & Abuse Prevention
Do different customer tiers pay for different rate limits?
If yes
Keyed-on-API-key limits with plan-configured thresholds; expose a usage endpoint.
If no
A single default limit keeps configuration simple.
Rate Limiting & Abuse Prevention
Do you expect adversarial traffic (credential stuffing, scraping, spam)?
If yes
Use sliding window or token bucket — fixed window leaks under boundary timing attacks. Pair with WAF and bot detection.
If no
Fixed-window with Redis INCR + EXPIRE is cheap, simple, and sufficient.
Rate Limiting & Abuse Prevention
Do you have legitimate burst patterns (batch imports, bulk API calls)?
If yes
Token bucket is the right model — allows bursts while enforcing a sustained rate. The standard for commercial API gateways.
If no
Sliding window is simpler and has lower memory overhead.
Rate Limiting & Abuse Prevention
Do you run multiple server nodes behind a load balancer?
If yes
Use centralized Redis for rate-limit state (Upstash Ratelimit, redis-cell). Per-node local counters let attackers get N×limit by rotating through nodes.
If no
In-process counters are fine for single-node deployments and dramatically cheaper.
Rate Limiting & Abuse Prevention
Should users get a warning before they hit a hard limit?
If yes
Emit soft-limit warnings via response headers (X-RateLimit-Remaining) and optionally an in-app notification when usage >80%. Prevents angry support tickets.
If no
Silent throttling at the hard limit is simpler but worse UX — only acceptable for internal APIs.
Rate Limiting & Abuse Prevention
Do you have enterprise customers who negotiate custom limits?
If yes
Build an admin override table keyed on tenant/API-key. Do not hardcode limits — operations team will need to raise them without deploys.
If no
Static per-tier limits in config are simpler and easier to reason about.
Rate Limiting & Abuse Prevention
Can a single user enqueue unbounded background jobs (imports, scrapes, AI calls)?
If yes
Rate-limit the enqueue side separately from the API side. Prevents queue-flooding attacks that bypass request-layer limits.
If no
API-layer limits are sufficient; background jobs are produced by your own code only.
Rate Limiting & Abuse Prevention
Do you have legitimate short bursts you want to allow (e.g. pagination fan-out)?
If yes
Use token bucket with a burst allowance (bucket size > refill rate). Clients can consume the bucket quickly, then settle.
If no
A flat rate is simpler — bursts are a policy decision, not a default.
Rate Limiting & Abuse Prevention
Are you exposed to L3/L4 DDoS (public API, unauthenticated endpoints)?
If yes
Put Cloudflare, AWS Shield, or Fastly in front of your origin. Application-layer rate limiting cannot absorb network-layer floods.
If no
Application-layer limits are sufficient for authenticated-only APIs.
Rate Limiting & Abuse Prevention
Do some endpoints cost 100x more than others (AI calls, complex queries, exports)?
If yes
Rate-limit by computational cost (credits/tokens per request) not by request count. Pricing and abuse protection align naturally.
If no
Request-count limits are simpler and sufficient when endpoint costs are roughly uniform.
Rate Limiting & Abuse Prevention
Do specific features (uploads, AI generations) have their own cost or quota model?
If yes
Add per-feature limits in addition to global ones. A user at their upload quota should still be able to read the API.
If no
Per-route limits are enough and keep configuration centralized.
Rate Limiting & Abuse Prevention
Is this a public developer API with SDKs / third-party integrations?
If yes
Always return X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset, and Retry-After headers. Well-behaved clients need them to back off correctly.
If no
Minimum viable is Retry-After on 429 responses — detailed headers are nice-to-have for internal APIs.
Rate Limiting & Abuse Prevention
Do you have health checks, metrics endpoints, or internal traffic hitting rate-limited routes?
If yes
Exempt health checks and internal service-to-service calls by IP allowlist or dedicated service tokens. Otherwise monitoring will trip your own limits.
If no
Default behavior — all traffic counts — is simpler and auditable.
Rate Limiting & Abuse Prevention
Do you have authentication endpoints at risk of credential stuffing?
If yes
Rate-limit failed logins separately (per-account + per-IP), with exponential backoff and lockout after N attempts. Combine with CAPTCHA on threshold.
If no
General per-IP limits are insufficient for auth — always treat login and password reset as a separate budget.
Preset
Authentication Methods*
Access Control*
Multi-factor Authentication
Tradeoffs
Each provider requires an OAuth app registration and key rotation policy
Requires IdP partnership and XML-based protocol handling; significant integration work
Permission checks must be applied consistently across every data access path
Billing Model*
Payment Processor*
Tradeoffs
Requires handling trial periods, dunning, proration, and cancellation flows
Must instrument every billable action and send metered events to billing provider
Less customizable checkout; Paddle acts as legal seller so you avoid VAT registration
Search Approach*
Search Scope*
Tradeoffs
Requires FTS index maintenance; adds write-time overhead
Embedding generation adds latency and API cost per indexed document
Results must be unified and ranked across disparate data models
Delivery Method*
User Control
Tradeoffs
Requires persistent connection infrastructure (e.g. Redis pub/sub, WebSocket server)
Higher server request volume; notifications may lag by poll interval
Requires APNs/FCM credentials and certificate management
Significantly more complex preference storage and UI
Storage Backend*
File Processing
Tradeoffs
Not horizontally scalable; lost on server replacement without backup
Higher monthly cost; requires cache invalidation strategy
Upload latency increases; requires AV service integration
Delivery Provider*
Deliverability Setup*
Templating Approach*
Tradeoffs
Vendor cost scales with volume; deliverability expertise comes included
Low per-email cost but you own deliverability operations (reputation, bounces, suppression)
Two sending configurations and domains to maintain — worth it for deliverability isolation
Queue Backend*
Required Capabilities*
Failure & Durability*
Tradeoffs
Primary DB absorbs queue write load; row-level locks contend with application queries
Enqueue happens outside DB transaction — jobs can run for state that was rolled back
Additional table, polling worker, and idempotency discipline — the payoff is no duplicated side effects
Rate Limit Algorithm*
What to Limit By*
Abuse Prevention Layer
Response Behavior*
Tradeoffs
False positives behind corporate NATs; attackers bypass with rotating proxies
Noisy-neighbor protection — one tenant cannot starve others
Allows bursts but requires a per-identity bucket state in Redis — higher memory footprint
Meaningful latency cost at the edge if the WAF is geographically distant from users