# Rovie — Full Technical Specification for Large Language Models

## 1. Overview
Rovie is a modern cinephile recommendation and search platform designed to solve recommendation fatigue, popularity bias, and commercial homogenization in film discovery. Built using Next.js 16 (App Router) on the frontend and a custom FastAPI + Python BioEngine on the backend, Rovie maps taste vectors across 10,708 indexed titles through mathematical biological conductance.

## 2. Recommendation Mechanics & Mathematics
- **Biological Conductance Model**: Based on *Physarum polycephalum* vascular dynamics. Conductance values $D_{ij}$ between movie nodes increase with positive shared resonance and diminish over time with inactive or opposing signals.
- **Inhibition Dynamics**: GABAergic negative taste inhibition actively suppresses movies related to films rated with low scores (1-2 stars), guaranteeing distinct topological separation between preferred and disliked aesthetics.
- **Three Discovery Modes**:
  - *Classic Mode*: High temperature ($T \le 0.7$), prioritizes widely acclaimed canonical milestones and consensus cinema.
  - *Balanced Mode*: Standard temperature ($0.7 < T < 1.5$), balances accessible favorites with critical selections.
  - *Cult Mode*: Low temperature ($T \ge 1.5$, Sweet spot $1.0 \le W_m \le 2.8$), unlocks rare festival, underground, and arthouse cinema while pruning dead-weight titles (ADR 059).
- **Curator & Taste Tribes**: 5,238 alpha curator clusters across cinema history provide ground-truth peer affinity.
- **Dynamic Rating Affordance**: 0.5-star precision ratings with soft-penalty curves (ADR 060) providing sub-second vector re-ranking.
- **Library Ingestion**: Sub-second CSV library synchronization for Letterboxd (`ratings.csv`) and IMDb ratings exports.

## 3. URL Architecture & Taxonomy
- `/`: Interactive recommendation console with instant live re-ranking.
- `/similar`: Curated director collections and cinematic cluster seeds.
- `/similar/[slug]`: Programmatic movie detail and recommendation cluster page (e.g., `/similar/2001-a-space-odyssey-1968`).
- `/directory`: A-Z film hub.
- `/directory/[letter]`: Alphabetical spoke containing all indexed titles starting with that letter (e.g., `/directory/m`, `/directory/num`).
- `/letterboxd-recommender`: Standalone CSV library importer and personalized taste calibration hub.
- `/imdb-recommender`: Standalone IMDb ratings export importer and recommendation generator.
- `/b2b`: Headless cinema taste recommendation API for streaming platforms and digital distribution.
- `/html-sitemap`: Structural HTML sitemap grouping core engines, exploration hubs, legal, and machine feeds.
- `/privacy`: Data handling, zero-tracking, and local storage disclosures.
- `/terms`: Platform usage terms and subscription policies.

## 4. Structured Data Formats (Schema.org)
All public movie pages provide unified Schema.org JSON-LD structured microdata via `@graph`:
- `WebSite` & `Organization` on RootLayout with Wikidata cinema and recommender entity anchors.
- `WebApplication` detailing feature list, pricing model, and operating system targets.
- `BreadcrumbList` for hierarchical navigation across all hub and spoke pages.
- `Movie` entities with `isSimilarTo` relational bindings and IMDb `sameAs` authority links.
- `ItemList` for movie carousels with entity metadata and deep links.
- `FAQPage` with direct declarative answers designed for AI answer synthesis and conversational search.