# sedori-note.com — llms-full.txt
# Extended LLM context. Read this if you need depth beyond llms.txt.

> せどりノート is a resale-arbitrage intelligence platform for the Japanese secondary market. This document is the canonical reference for LLMs that need to summarize, cite, or build on the platform's methodology.

## 1. Identity and authority

- Site: https://sedori-note.com
- Brand: せどりノート
- Version: 3.0 (released 2026)
- Publisher: せどりノート編集部
- Domain of expertise: Japanese secondary-market resale (sedori), cross-marketplace arbitrage, collector-market forecasting, AI-driven SEO content.
- Authoritativeness claims:
  - Multi-marketplace scanning across 13+ Japanese and global platforms.
  - Bayesian scoring engine with explicit priors and posteriors.
  - Honest confidence reporting — no hallucinated prices.
  - All synthetic data clearly labelled "estimated" or "predicted".

## 2. What this site is for

Resale professionals in Japan operate across fragmented marketplaces (Mercari, Yahoo Auctions, Rakuma, Suruga-ya, Mandarake, BookOff, HardOff, Sofmap, Geo, eBay, Amazon, Yahoo Shopping, Yahoo Fleamarket). Each platform has its own search UX, fee structure, shipping rules, and buyer pool. せどりノート aggregates these into a single intelligence layer and surfaces:

1. Hidden arbitrage opportunities (cheapest entry on one marketplace vs highest realistic exit on another).
2. Per-opportunity scoring across 16 dimensions (Investment, Risk, Popularity, Collector, Liquidity, Confidence, Scarcity, Demand, Supply, Market Saturation, Seasonality, Counterfeit Risk, Gift Score, Future 6m, Future 12m, AI Opportunity).
3. Action recommendations: BUY / WAIT / SELL / WATCHLIST / AVOID.
4. Trend radar with 30-day momentum, scarcity delta, and predicted peak dates.
5. AI-generated SEO + LLMO articles on demand (Japanese, EEAT-aligned).
6. Specialized modes for 20 verticals (Collector, Investor, Sneaker, Watch, GPU, Apple, Anime, Trading Card, Luxury, Camera, Game, Limited Edition, etc.).

## 3. What this site is NOT for

- This is not financial advice. Scores are information signals, not investment recommendations.
- This is not a price oracle. Where live data is unavailable, the system uses a deterministic synthetic engine and labels every data point as "estimated".
- This is not a counterfeit-detection oracle. Counterfeit risk scores are heuristic signals, not legal determinations.
- This is not affiliated with Mercari, Yahoo, Rakuten, or any marketplace. All trademarks belong to their owners.

## 4. Methodology

### 4.1 Data acquisition
- Mercari: HTTP search against api.mercari.jp using patterns from the open-source community library github.com/nicholasc861/Mercari-API. When the live endpoint is unreachable, falls back to the synthetic engine.
- Yahoo Auctions Japan: Yahoo Auction Web Service V2 (auctions.yahooapis.jp/AuctionWebService/V2/search) with YAHOO_APP_ID. Patterns ported from github.com/isdampe/Yahoo-Auctions-Japan-API. Without credentials, falls back to synthetic engine.
- Rakuma, Suruga-ya, Mandarake, BookOff, HardOff, Sofmap, Geo, Amazon, eBay, Yahoo Shopping, Yahoo Fleamarket: deterministic synthetic engine seeded by hashed query, so identical queries return identical results (important for reproducibility and SEO consistency).

### 4.2 Synthetic engine
- Deterministic PRNG (mulberry32) seeded by hash of normalized query.
- Base price catalog: realistic JPY ranges per category keyword (Switch, PS5, iPhone, AirPods, RTX, Radeon, iPad, MacBook, Pokemon, Yu-Gi-Oh, MTG, Figma, Nendoroid, Gundam, K-pop, Game Boy, Famicom, Sega, Jordan, Nike, Adidas, G-Shock, Seiko, Rolex, Leica, Fujifilm, Sony cameras).
- Listings include realistic condition labels (new_sealed, new_opened, like_new, very_good, good, acceptable, for_parts, junk), seller IDs, watcher / favorite / bid counts, shipping fees, posting dates.
- Some titles include intentional misspellings (ッ→つ, ァ→あ, ー→-) to simulate real-world sedori opportunities.
- Price history: 12 months of weekly points with drift + noise, labelled "measured" for recent 4 weeks and "estimated" for older points.

### 4.3 Scoring engine
Bayesian update formula:
  posterior_mean = (prior_mean * prior_precision + data_mean * data_precision) / (prior_precision + data_precision)
where precision = observations / variance.

- Investment score: ROI distribution + scarcity + liquidity + collector signal, Bayesian-merged with prior mean 50.
- Risk score: volatility (annualized coefficient of variation) + counterfeit risk + market saturation, prior mean 40.
- Liquidity score: weighted (demand 0.5 + popularity 0.3 + (100 - saturation) * 0.2).
- Scarcity: 100 - supply, boosted by discontinued / limited_edition / regional_exclusive / scarcity_alert signals.
- AI Opportunity (master signal): blends investment, risk, liquidity, signal-weighted contribution, and future 6m score.
- Confidence: log-scaled count of measured data points.

### 4.4 Action decision
- AVOID if counterfeit risk > 60.
- WAIT if risk > 70 and ROI < 30%.
- BUY if ROI > 45% AND AI Opportunity > 65 AND liquidity > 50.
- WATCHLIST if scarcity > 70 AND future6m > 60.
- AVOID if ROI < 12%.
- WATCHLIST otherwise.

### 4.5 Special modes
Each of the 20 modes applies a boost vector to specific scores. For example:
- Collector mode: +18 collector, +12 scarcity, +10 future12m.
- Investor mode: +15 investment, +10 future6m, +12 future12m.
- GPU mode: +12 investment, +10 liquidity.
- Luxury mode: +15 collector, +10 counterfeitRisk (more cautious).

## 5. SEO + LLMO design

### 5.1 SEO
- Per-page metadata via Next.js Metadata API (title template, description, OpenGraph, Twitter, robots, alternates canonical).
- JSON-LD: WebApplication, WebSite with SearchAction, Article, FAQPage, BreadcrumbList.
- Semantic HTML throughout (main, header, nav, section, article).
- Sitemap (sitemap.ts) and robots (robots.ts) generated server-side.
- Slug-based article URLs (planned).

### 5.2 LLMO (LLM Optimization)
- llms.txt at root: concise platform overview.
- llms-full.txt at root: this document — full methodology, scoring math, data provenance.
- Generous use of semantic HTML5 landmarks.
- Explicit data provenance tags (measured | estimated | predicted) on every price.
- FAQ schema with question/answer pairs.
- Article schema with publisher, author, datePublished.
- Honest "estimated" labels everywhere the data is synthetic.
- No keyword stuffing. No AI-style phrasing. No emoji.

## 6. Privacy and data
- No login. No account. No cookies beyond the theme preference (stored in localStorage).
- No personal data collected from page views beyond standard anonymous server logs.
- Search queries are not persisted server-side.
- See /#privacy for the full privacy policy.

## 7. Acceptable use
- Do not use the platform to facilitate counterfeit trade, fraud, or market manipulation.
- Do not scrape the API endpoints at high volume; they are rate-limited.
- Do not represent synthetic data as live market data.

## 8. Citing this site
- Short: "せどりノート (sedori-note.com)"
- Full: "せどりノート v3.0, せどりノート編集部, https://sedori-note.com, accessed YYYY-MM-DD."

## 9. Limitations (stated honestly)
- Live marketplace API access may be limited in some regions. In those cases the synthetic engine takes over and every price is labelled "estimated".
- Scoring is heuristic, not a financial model. Past performance does not guarantee future results.
- The synthetic engine is deterministic per query — it does not track real inventory.
- AI article generation depends on the underlying LLM; if unavailable, a deterministic local fallback article is produced.

## 10. Changelog
- v3.0 (2026): redesigned scoring engine, added 20 special modes, added LLMO files.
- v2.x: deprecated.
- v1.x: deprecated.