Film Price Guide · physical media market intelligence

What we built

A full-stack data product for collectors: automated sold-listing ingestion, community signal, wave scoring, and a public web app — grounded in real comps, not asking prices.

Production snapshot · March 2026 · standalone portfolio (main app offline)

901,713

Price records

3,753

Titles with comps

3,770

Catalog titles

38,075

Wave analyses

Product surfaces

Each item below is a shipped feature in the RetroPrice codebase: PHP pages, APIs, cron jobs, and Python analyzers wired to PostgreSQL. This page describes them — no live app links.

Price guide

Sold-price medians and format breakdowns per title, SEO-friendly title/format landing pages, and confidence cues from listing volume.

PostgreSQL aggregates
Format-aware comps

Search & title pages

Fast lookup across the TMDB-backed catalog with enriched metadata (posters, years, cast) and deep links into comp history.

TMDB / OMDb
Redis caching

Thrift check

In-store “is this a deal?” flow: compare shelf price to recent sold medians across formats and conditions.

Mobile-first UX
Median comps

Goodwill deals

Automated poller surfaces underpriced listings from Goodwill’s online shop against internal comp thresholds.

Deal scoring
Cron poller

Amazon new releases

Curated feed of hot physical media pre-orders and drops, cross-referenced with collector interest signals.

Release tracking
Format filters

Collector guides

Long-form SEO guides (grading, hunting, format primers) generated from structured templates and JSON metadata.

Content pipeline
Schema.org

Wave analysis dashboard

0–100 heat scores blending eBay scarcity, Reddit velocity, and seasonality — with human-readable buy/sell/hold states.

Python analyzer
sklearn blend

Public collections

Community shelves and watchlists: share what you own, track wants, and discover titles trending in the collector graph.

Accounts & CSRF
Social discovery

Public read APIs

Versioned JSON endpoints for prices, heating-up titles, myth-bust stats, and integrator API keys with rate limits.

API keys
v1 REST

Architecture

PHP handles the web app and API integrations; Python runs ML and heavy analysis. PostgreSQL stores JSONB listing payloads; Redis caches hot reads. A cron-driven pipeline runs twice daily.

┌─────────────────────────────────────────────────────────┐
│                    Daily pipeline (cron)                   │
│  eBay crawl ──→ Reddit scan ──→ AI insights ──→ Wave     │
│  (PHP)          (PHP)           (Python)        analyzer │
│                       ▼                           ▼     │
│              ┌─────────────────┐        ┌──────────────┐│
│              │   PostgreSQL    │        │ ML predict   ││
│              └────────┬────────┘        └──────────────┘│
│                       ▼                                 │
│              ┌─────────────────┐                        │
│              │  PHP web app    │                        │
│              └─────────────────┘                        │
└─────────────────────────────────────────────────────────┘

PHP 8.x Python 3.12 PostgreSQL Redis scikit-learn Alpine.js OAuth 2.0 PWA Railway / Vercel

Data collected

Every estimate traces back to sold-listing and auction data plus optional community signal.

Marketplaces & APIs

eBay Browse API — sold listings, format, condition, grading metadata
Heritage Auctions — high-end memorabilia and graded tape results
TMDB — catalog metadata for ~3,700+ titles
Reddit — 16 collector subreddits for mention velocity and price language
Goodwill / Amazon — deal surfaces and hot new releases

Format coverage (snapshot)

Format	Listings	Titles	Avg sold
DVD	376,539	3,251	$12.14
Blu-ray	293,176	2,177	$18.21
4K UHD	154,007	852	$32.37
LaserDisc	73,453	747	$42.05
VHS	4,538	939	$22.03

Surprise valuations the data surfaced

Title	Format	Avg sold
Akira (1988)	VHS	$1,543.99
The Ballad of Buster Scruggs	Blu-ray	$1,293.44
The Matrix (1999)	LaserDisc	$1,098.44
The Iron Giant (1999)	LaserDisc	$446.10

Collection began August 2025 with twice-daily eBay batches (~100 catalog titles per run across formats).

Daily pipeline (5 steps)

TMDB batch crawler (PHP) — eBay across all physical formats; JSONB rows in price_history.
Reddit trending analyzer (PHP) — subreddit scans with catalog title matching.
AI price analysis (Python) — weighted pricing; optional Ollama market notes.
Wave analyzer (Python) — scarcity (40%) + Reddit (40%) + seasonality (20%) → 0–100 score.
ML predictions (Python) — weekly sklearn retrain on 7-day price change; blended with rules.

Skills & disciplines

Backend & APIs

eBay OAuth, Reddit, TMDB, rate limits, webhooks, public read APIs.

Data engineering

Cron ETL in PHP/Python, JSONB modeling, sitemaps at scale.

ML & analytics

Feature engineering, scikit-learn pipelines, confidence blending.

Frontend & UX

Design system, Alpine.js, PWA, SEO title/format pages.

Security & ops

CSRF, sessions, deployment automation, backups and monitoring.

Domain expertise

Physical media formats, grading, comps, collector guides.

Wave score → action

The wave system maps numeric scores to collector-friendly states used across dashboards and alerts.

Score	Status	Typical read
75+	Peak	Consider selling before cool-off
60–74	Heating up	Hold / watch momentum
40–60	Stable	Sideways market
25–40	Cooling	Wait for better entry
<25	Bottom	Default state for most titles — hunt inflection

Full case study

A standalone long-form narrative lives in this folder as case-study.md (metrics, pipeline phases, ML results, and lessons). Additional repo docs: docs/how-we-built-film-price-guide.md, docs/wave-analysis-system.md.