TV Shows & Streaming Database — Deep Research Brief, Part 11: Technical Architecture, Costs, and Business Model
104 preserved extracts. Research planning document; requested work is not evidence that the work was completed.
- NOTE 4356
TV Shows & Streaming Database — Deep Research Brief, Part 11: Technical Architecture, Costs, and Business Model
Research window: January 1, 2019 – September 15, 2026 (with emphasis on the last 24 months). State the research date; record the date of every price quote and program term.
- NOTE 4357
TV Shows & Streaming Database — Deep Research Brief, Part 11: Technical Architecture, Costs, and Business Model
Conduct exhaustive research on the technology stack, operating costs, and revenue options for a TV-only database, tracker, and "where to watch" site, from a solo-founder launch to a large-scale service. The goal is to give the website a concrete architecture, cost models at four traffic levels, a verified list of monetization programs that actually accept this kind of site, budget and break-even scenarios, and an MVP scope with 90-day and 12-month roadmaps.
- NOTE 4358
1. How to research this part
Treat every vendor price, free-tier limit, RPM figure, affiliate commission, and company financial as a lead to verify on the vendor's current pricing page, program terms, or filings. Prices change often; record the access date.
- NOTE 4359
1. How to research this part
Minimum record counts: at least 60 infrastructure vendor records (hosting, database, search, CDN, image, email, push, analytics, monitoring, AI APIs); at least 40 monetization program records (ad networks, affiliate programs, sponsorship marketplaces, data buyers); at least 15 comparable-company economics records; cost models at 4 traffic tiers; 3 budget scenarios.
- NOTE 4360
1. How to research this part
Prefer: official pricing pages and docs, program terms, SEC filings and investor letters, founder interviews and public revenue posts (label as claimed), benchmark reports from ad networks (label as vendor-reported).
- NOTE 4361
1. How to research this part
Show all assumptions (pages per visit, cache hit rate, API calls per page, conversion rates) so figures can be recalculated.
- NOTE 4362
1. How to research this part
Do not recommend scraping data providers, violating API terms, or ad-network policy workarounds (such as arbitrage or incentivized clicks).
- NOTE 4363
2. Ingestion pipelines
Sources to connect (see master brief section 4): TMDB API, TVmaze API, Wikidata dumps and SPARQL, licensed IMDb datasets if licensed, TheTVDB (licensed), AniList GraphQL, Jikan (MyAnimeList, check terms), availability APIs (JustWatch partner, Watchmode, Streaming Availability API, Reelgood B2B), Netflix Top 10 TSV files, Nielsen published charts (text only, attributed).
- NOTE 4364
2. Ingestion pipelines
Pipeline design: scheduled fetchers per source with rate-limit budgets; raw snapshot storage (object storage, immutable, dated); normalization to a staging schema; entity resolution; merge to canonical store with provenance per field; change events emitted to queues.
- NOTE 4365
2. Ingestion pipelines
Orchestration options: cron plus a job queue (BullMQ, Sidekiq, Celery), Temporal, Dagster, Prefect, Airflow. Compare operational burden for one person.
- NOTE 4366
2. Ingestion pipelines
Worked example: nightly TVmaze full-show updates using its "updates" endpoint (verify), hourly TMDB changes endpoint (verify), daily availability delta per country for top 20,000 shows, weekly full refresh for the long tail.
- NOTE 4367
2. Ingestion pipelines
Produce: source-by-source ingestion spec (endpoint, frequency, quota, fields, license flags, attribution string).
- NOTE 4368
3. Entity resolution
Matching keys: external IDs (IMDb tt, TMDB, TVDB, TVmaze, Wikidata QID, EIDR), normalized titles (transliteration, articles, punctuation), first air year, country, network, episode count, and top-billed cast overlap.
- NOTE 4369
3. Entity resolution
Methods: deterministic ID crosswalks first; then probabilistic matching (Splink, Dedupe, Zingg (verify maintenance status)); then human review queue for scores in a gray band.
- NOTE 4370
3. Entity resolution
Episode alignment: map aired order vs. streaming order vs. absolute (anime) using date sequences and title similarity; store mappings rather than overwrite.
- NOTE 4371
3. Entity resolution
Worked example: La Casa de Papel / Money Heist where Netflix re-cut episodes into different parts and counts than the Antena 3 broadcast (verify counts). Show the crosswalk table.
- NOTE 4372
3. Entity resolution
Metrics: precision/recall on a hand-labeled set of 2,000 pairs.
- NOTE 4373
4. Data store: Postgres vs. graph
Postgres with JSONB, recursive CTEs, and extensions (pgtrgm, pgvector, Apache AGE (verify status)) vs. Neo4j, Amazon Neptune, Memgraph, or ArangoDB.
- NOTE 4374
4. Data store: Postgres vs. graph
Evaluate for queries: franchise trees, "actors who worked together," character played by multiple actors, cross-country availability joins, change history.
- NOTE 4375
4. Data store: Postgres vs. graph
Managed Postgres options with pricing tiers: Neon, Supabase, AWS RDS/Aurora, Google Cloud SQL/AlloyDB, Crunchy Bridge, PlanetScale Postgres (verify launch), Railway, Render.
- NOTE 4376
4. Data store: Postgres vs. graph
Audit trail: append-only field-change table with source, editor, and timestamp; temporal tables; rollback tooling.
- NOTE 4377
4. Data store: Postgres vs. graph
Recommendation to research: Postgres as canonical store plus derived read models (search index, cache, precomputed graph edges).
- NOTE 4378
5. Search engines
Compare Typesense (open source and Typesense Cloud), Meilisearch (open source and Cloud), OpenSearch (self-hosted and AWS OpenSearch Service/Serverless), Elasticsearch (Elastic Cloud), Algolia, and Postgres full-text as baseline.
- NOTE 4379
5. Search engines
Criteria: typo tolerance, multilingual tokenization (Japanese, Korean, Chinese, Hindi, Arabic, Turkish), synonyms and alternate titles, faceting, geo/country filtering, vector/hybrid search for "that show where…" queries, ranking by popularity, index size for ~5 million documents, pricing at 1 million and 50 million searches per month (verify each vendor's page).
- NOTE 4380
5. Search engines
Test set: 200 queries including misspellings ("brekaing bad"), alternate titles ("La Casa de Papel"), romanized titles ("Kimetsu no Yaiba"), character names ("Walter White"), and natural language ("show about a hospital night shift in Pittsburgh").
- NOTE 4381
6. Recommendation stack
Stages: content-based similarity (genre, tone tags, cast/crew, embeddings of original synopses); collaborative filtering once tracking data exists (implicit ALS, LightFM, two-tower models); editorial overrides; filters by user's services and country.
- NOTE 4382
6. Recommendation stack
Tools: pgvector, Qdrant, Weaviate, Pinecone, Vespa; open-source libraries (implicit, RecBole (verify)); managed options (AWS Personalize, Google Vertex AI Search for media (verify)).
- NOTE 4383
6. Recommendation stack
Cold start: onboarding picks, import from other trackers (important after TV Time's shutdown July 15, 2026 (verify) and Trakt free-tier limits).
- NOTE 4384
6. Recommendation stack
Privacy: watch history is sensitive; process under explicit settings (see Part 09).
- NOTE 4385
7. Availability refresh, calendars, and notifications
Refresh jobs: priority tiers by popularity and volatility; change detection diffs; "last verified" timestamps; alerts when a top title disappears from all services.
- NOTE 4386
7. Availability refresh, calendars, and notifications
Deep-link validation: periodic HTTP checks where permitted by terms; user "link broken" reports.
- NOTE 4387
7. Availability refresh, calendars, and notifications
Release calendar: store release instant in UTC plus source time zone and precision; render in user time zone.
- NOTE 4388
7. Availability refresh, calendars, and notifications
Notifications: email (Amazon SES, Postmark, Resend, SendGrid, Mailgun), push (Firebase Cloud Messaging, APNs, OneSignal, Expo), web push. Fan-out design: a premiere for a show with 500,000 trackers; batching, deduplication, quiet hours, spoiler-safe copy.
- NOTE 4389
7. Availability refresh, calendars, and notifications
Produce: notification cost per million messages by vendor (verify).
- NOTE 4390
8. Image pipeline and license-aware caching
Image sources carry license records (source, license, attribution, allowed cache duration, hotlink-only flag). TMDB image terms and attribution (verify); press-site editorial restrictions.
- NOTE 4391
8. Image pipeline and license-aware caching
Transform and CDN: Cloudflare Images, imgix, Cloudinary, Bunny Optimizer, AWS CloudFront plus Lambda, self-hosted imgproxy or Thumbor.
- NOTE 4392
8. Image pipeline and license-aware caching
Formats: AVIF/WebP, responsive srcset, LCP optimization for key art.
- NOTE 4393
8. Image pipeline and license-aware caching
Enforcement: pipeline refuses to cache images whose license forbids it; purge on license expiry or takedown.
- NOTE 4394
9. Rendering, apps, API, and analytics
Rendering: Next.js (App Router, ISR, partial prerendering (verify)), Astro, Remix/React Router 7, Nuxt, SvelteKit. Hosting on Vercel, Netlify, Cloudflare Pages/Workers, Fly.io, or containers. Compare cost behavior for millions of pages (ISR revalidation, function invocations, bandwidth).
- NOTE 4395
9. Rendering, apps, API, and analytics
Strategy: SSG for top pages, ISR with on-demand revalidation on data change, edge caching with stale-while-revalidate, long tail rendered on first request.
- NOTE 4396
9. Rendering, apps, API, and analytics
Public API: REST or GraphQL; keys, quotas, attribution rules; license of site's own contributed data (for example, CC BY-NC vs. commercial tier); cannot relicense third-party data.
- NOTE 4397
9. Rendering, apps, API, and analytics
Mobile and TV apps: PWA first; React Native/Expo or Flutter; tvOS, Android TV/Google TV, Roku (BrightScript/SceneGraph), Fire TV, Samsung Tizen, LG webOS. App store fees (Apple and Google small business rates (verify)) and 2024–2026 anti-steering rule changes (verify).
- NOTE 4398
9. Rendering, apps, API, and analytics
Analytics: Plausible, Fathom, PostHog, Umami, GA4 (consent implications), server logs; product analytics for tracker funnels.
- NOTE 4399
9. Rendering, apps, API, and analytics
Monitoring: Sentry, Grafana Cloud, Better Stack, uptime checks.
- NOTE 4400
10. AI enrichment with human review
Uses: mood/tone tagging, trigger-warning suggestions, original synopsis drafts, episode summary drafts from licensed or public data, entity-resolution assistance, moderation triage, natural-language search.
- NOTE 4401
10. AI enrichment with human review
Vendors and cost per million tokens (verify current pricing): Anthropic, OpenAI, Google, Mistral, open models on Together, Fireworks, Groq, or self-hosted.
- NOTE 4402
10. AI enrichment with human review
Guardrails: never generate facts without source; human approval for published text; label AI-assisted content; hallucination audit sample of 500 outputs per release; do not feed copyrighted synopses in ways that violate terms.
- NOTE 4403
10. AI enrichment with human review
Worked example: cost to draft original 120-word synopses for 60,000 shows and 400,000 seasons, plus editor review time at 2 minutes each.
- NOTE 4404
11. Cost models by traffic
Build four models with explicit assumptions (pages per visit 2.5, cache hit rate 90–98%, 5% of visitors logged in, 1% of visitors trigger search, image bytes per page):
- NOTE 4405
11. Cost models by traffic
Tier — Monthly visits — Items to price
- NOTE 4406
11. Cost models by traffic
A — 10,000 — Hosting free/hobby tiers, managed Postgres small, search self-host or free tier, email under 10K/month, domain, TMDB free API (check commercial terms)
- NOTE 4407
11. Cost models by traffic
B — 100,000 — Paid hosting plan, Postgres 2–4 GB RAM, search cloud small, availability API paid tier, image CDN, email 50K
- NOTE 4408
11. Cost models by traffic
C — 1,000,000 — Multi-region CDN, Postgres with replica, search cluster, availability data license, push notifications, monitoring, part-time moderation
- NOTE 4409
11. Cost models by traffic
D — 10,000,000 — Dedicated infrastructure or committed-use pricing, commercial metadata license (Gracenote or TiVo quotes, verify availability to small companies), staff
- NOTE 4410
11. Cost models by traffic
For each tier: low/base/high monthly cost; data licensing is likely the largest line at tiers C and D (research Watchmode, Streaming Availability API, JustWatch partner pricing (verify; many are quote-based)).
- NOTE 4411
11. Cost models by traffic
Include people costs separately: moderation, editorial, engineering.
- NOTE 4412
12. Revenue: advertising
Entertainment display RPM ranges by country and by network: Google AdSense, Ezoic, Mediavine (Journey for smaller sites (verify)), Raptive (formerly AdThrive; traffic minimum (verify)), Freestar, Playwire, Monumetric, Snigel, SHE Media. Record minimum traffic requirements, revenue share, and content policy fit (spoilers, mature shows).
- NOTE 4413
12. Revenue: advertising
Research seasonality (Q4 peak, Q1 drop), U.S. vs. UK vs. India RPM gaps, and consent-rate effects in EEA.
- NOTE 4414
12. Revenue: advertising
Video ads: outstream and instream, trailer-adjacent inventory rules.
- NOTE 4415
12. Revenue: advertising
Direct sold: studio and streamer FYC and launch campaigns; research whether small TV sites get Emmy FYC budgets (verify).
- NOTE 4416
13. Revenue: affiliate and partner programs (verify each)
For every program record: network (Impact, CJ, Rakuten Advertising, Partnerize, Awin, FlexOffers), eligibility, countries, commission type (CPA per trial or paid subscriber, revenue share), cookie window, restrictions (no brand bidding, no coupon use, no incentivized traffic), and whether content sites are accepted.
- NOTE 4417
13. Revenue: affiliate and partner programs (verify each)
Streaming services to check: Netflix (historically no public affiliate program (verify)), Prime Video via Amazon Associates (Prime trial bounties, Prime Video Channels subscriptions (verify)), Disney+/Hulu/ESPN bundle, HBO Max, Paramount+, Peacock, Apple TV via Apple Services Performance Partners (verify current program scope and commissions), YouTube TV, Sling, Fubo, Philo, DirecTV Stream, Starz, MGM+, AMC+, BritBox, Acorn TV, Crunchyroll, HIDIVE, Viki, Kocowa, Hallmark+, Great American Pure Flix, CuriosityStream, Shudder, Dropout, Nebula, NordVPN-type offers (exclude any geo-evasion framing).
- NOTE 4418
13. Revenue: affiliate and partner programs (verify each)
Devices and adjacent: Roku, Fire TV, Chromecast/Google TV Streamer, TVs, soundbars, physical media (Blu-ray box sets via Amazon), books that inspired shows, merchandise.
- NOTE 4419
13. Revenue: affiliate and partner programs (verify each)
Aggregator partnerships: JustWatch or Reelgood referral arrangements (verify).
- NOTE 4420
13. Revenue: affiliate and partner programs (verify each)
Worked example: 1 million monthly visits, 3% click to a provider, 5% of clicks start a trial, $5 average CPA (all assumptions to verify) = $1,500/month; compare to display.
- NOTE 4421
14. Other revenue
Sponsorships: newsletter sponsors (streamers, device makers), sponsored lists with disclosure, awards-season FYC.
- NOTE 4422
14. Other revenue
Premium membership: ad-free, advanced stats, unlimited lists, import tools, calendar feeds, early features. Benchmarks: Letterboxd Pro and Patron, Trakt VIP, Simkl VIP, Serializd premium, MyAnimeList Supporter, AniList donations (verify prices). Conversion rates 0.5–3% of active users (verify benchmarks).
- NOTE 4423
14. Other revenue
B2B data licensing: renewal/cancellation feeds, episode-level data, availability change logs, community tags. Buyers: smart TV makers, ad-tech, research firms, AI companies (check rights to license derived data).
- NOTE 4424
14. Other revenue
API tiers: free non-commercial, paid commercial.
- NOTE 4425
15. Comparable company economics (verify all)
IMDb inside Amazon (no standalone financials; IMDbPro subscription pricing).
- NOTE 4426
15. Comparable company economics (verify all)
JustWatch: B2B and advertising for studios; funding and employee counts.
- NOTE 4427
15. Comparable company economics (verify all)
Reelgood: pivot to B2B data (verify).
- NOTE 4428
15. Comparable company economics (verify all)
Letterboxd: acquisition by Tiny (2023) and reported user and revenue figures.
- NOTE 4429
15. Comparable company economics (verify all)
TMDB: community model, commercial API licensing.
- NOTE 4430
15. Comparable company economics (verify all)
TheTVDB: subscription model and licensing to media-center apps.
- NOTE 4431
15. Comparable company economics (verify all)
Trakt: VIP pricing changes and ownership history.
- NOTE 4432
15. Comparable company economics (verify all)
TV Time: Whip Media ownership and shutdown (July 2026).
- NOTE 4433
15. Comparable company economics (verify all)
Simkl, Serializd, Likewise (funding), Plex (funding and layoffs), Common Sense Media (nonprofit funding), What's on Netflix (independent publisher), MyAnimeList (Media Do ownership), Fandom (ownership and layoffs).
- NOTE 4434
16. Budget scenarios and break-even
Bootstrapped solo founder: $0–$500/month infrastructure, open/low-cost data, founder time unpaid; 12-month target: 250,000 monthly visits.
- NOTE 4435
16. Budget scenarios and break-even
Seed scenario: $1.5–3 million raised (verify typical 2025–2026 consumer seed sizes); team of 5–8; commercial availability license; apps.
- NOTE 4436
16. Budget scenarios and break-even
12-month path to sustainability: month-by-month revenue ramp and cost lines.
- NOTE 4437
16. Budget scenarios and break-even
Break-even analysis: traffic needed under (a) ads only at base RPM, (b) ads plus affiliate, (c) ads plus affiliate plus 1% premium conversion, (d) B2B data contract. Show sensitivity to RPM and data license cost.
- NOTE 4438
17. MVP scope
Must have: show/season/episode pages for top 20,000 shows (U.S., UK, Canada, Australia, India); where to watch with last-verified dates in 5 countries; search; watchlist and episode tracking; imports (Trakt, TV Time export, CSV); release calendar; renewal tracker; account privacy settings.
- NOTE 4439
17. MVP scope
Later: ratings heatmaps, social, apps, API, recommendations beyond content-based, more countries.
- NOTE 4440
18. Roadmap
90 days: weeks 1–3 data pipelines and schema; weeks 4–6 page templates and search; weeks 7–9 tracker and imports; weeks 10–12 calendar, renewal tracker, launch, Search Console monitoring.
- NOTE 4441
18. Roadmap
12 months: Q1 launch and indexation; Q2 newsletter and PWA; Q3 ratings and recommendations; Q4 apps, API beta, B2B pilot, premium.
- NOTE 4442
19. Timeline 2019 – 2026
Log: TheTVDB licensing changes; Letterboxd acquisition (2023); Reddit and X API pricing changes (2023) as warnings; Heroku free tier end (Nov 2022); Vercel and Netlify pricing changes (verify); Elastic license change and OpenSearch fork (2021) and Elastic's AGPL option (2024) (verify); Algolia pricing changes; LLM API price declines 2023–2026; Mediavine and Raptive threshold changes; Apple and Google app store rule changes (2024–2026); Crunchyroll free tier end (Dec 2025); TV Time shutdown (July 2026); Trakt free-tier restriction (2025–2026).
- NOTE 4443
20. Deliverables for Part 11
Architecture diagram and component decision records.
- NOTE 4444
20. Deliverables for Part 11
Ingestion spec for 15+ sources.
- NOTE 4445
20. Deliverables for Part 11
Entity-resolution method and evaluation set (2,000 labeled pairs).
- NOTE 4446
20. Deliverables for Part 11
Search engine benchmark (6 engines, 200 queries).
- NOTE 4447
20. Deliverables for Part 11
Vendor database (60+) with pricing and access dates.
- NOTE 4448
20. Deliverables for Part 11
Cost models at 10K/100K/1M/10M visits with assumptions.
- NOTE 4449
20. Deliverables for Part 11
Monetization program database (40+) with terms verified.
- NOTE 4450
20. Deliverables for Part 11
Comparable company economics table (15+).
- NOTE 4451
20. Deliverables for Part 11
Budget scenarios (bootstrapped, seed, sustainability) and break-even sensitivity tables.
- NOTE 4452
20. Deliverables for Part 11
MVP spec, 90-day plan, 12-month roadmap.
- NOTE 4453
20. Deliverables for Part 11
Glossary of 70+ technical and business terms; source list of 120+ sources.
- NOTE 4454
Research standards
State research date; every price, limit, commission, and RPM needs a source and access date.
- NOTE 4455
Research standards
Label vendor-reported benchmarks, founder claims, and third-party estimates as such.
- NOTE 4456
Research standards
Show formulas and assumptions for all cost and revenue models.
- NOTE 4457
Research standards
Verify each affiliate program's current existence and acceptance of content sites; many streaming programs open, close, or change networks.
- NOTE 4458
Research standards
Respect all data-provider terms; note license constraints on caching, images, and commercial use.
- NOTE 4459
Research standards
Flag contracts, data licenses, and app-store terms for lawyer review.