# RareIndex methodology

This page describes how RareIndex turns fragmented collectible markets into structured market data. Every number on the site carries its sample size, confidence and freshness; when the evidence is insufficient we display "Data unavailable" rather than an estimate.

# Data sources and provenance

Data is collected from public marketplaces, auction houses, dealers, price guides, grading companies, manufacturer catalogues and official datasets through a connector framework. Each connector declares what it collects (sales, listings, catalogue, population, auctions), its engine priority (direct API → structured feed → Firecrawl → Scrapfly) and its trust score. Every stored record keeps its source URL, crawl timestamp, native currency, parser version and connector version, so any displayed datapoint can be traced back to its origin.

RareIndex never bypasses logins, paywalls or access controls, respects robots policies and rate limits, and attributes third‑party data to its source.

# Canonical assets and variants

A canonical asset is the collectible object independent of its condition (e.g. Charizard #4/102, Base Set 1st Edition Holo, 1999). A variant is a grade/condition slice of that asset (e.g. PSA 10, CIB, deadstock size 10). Sales, listings and valuations are attached to both.

Entity resolution uses deterministic identifiers first (card numbers, set codes, model references, UPC/ISBN/SKU, certification numbers), then a normalised canonical key, then fuzzy text similarity and, for ambiguous cases, embedding and LLM verification. Merges are recorded and auditable.

# Sales versus listings versus guide prices

Sales are observed transactions with a date and a price. They drive valuations and indices.
Listings are asking prices. They are never treated as market value; the vs RIV column shows the discount or premium of an ask to the valuation.
Price observations are guide values published by price guides or card databases. They are informative priors, weighted below transactions, and are labelled as such.

Prices are stored in their native currency and converted to USD at the historical exchange rate of the sale date (ECB reference rates). Historical sales are never converted at today's rate.

# RareIndex Valuation (RIV)

RIV is an ensemble estimate computed per asset and per variant from verified sales within a rolling window:

latest sale, medians of the last 5 / 10 / 20 sales, volume‑weighted price, trimmed mean and exponentially weighted price (recent sales weigh more);
comparable-, grade- and condition-adjusted prices when the asset has too few direct sales;
a low / high range from the dispersion of the sample;
a confidence derived from sample size, recency, dispersion and source trust.

Outliers (data errors, bundles, cancelled auctions, misidentified items, abnormal prices) are flagged and excluded from the calculation but never deleted; the audit log records every decision.

# Grading

Grades are tracked per grading company. A PSA 10, a BGS 10 and a CGC 10 are different variants; their relative premiums are measured empirically from sales of the same asset in different grades and published with their sample sizes. Population reports, when available, feed rarity and population momentum (growth of supply).

# Scores

Rarity (0–100) combines production quantity, surviving and graded population, listing and sale frequency, variant rarity and distribution exclusivity. Unknown inputs stay unknown.
Liquidity (0–100) combines transactions per month, number of marketplaces, active listings, days between transactions and the spread between asks and sales.
Momentum measures price and transaction acceleration over 7d / 30d / 90d / 1y.
Trending is the normalised product of price, volume, search, listing and news momentum.
Value opportunity flags listings materially below RIV, with confidence. It is analytical data, not advice.
Data quality (0–100) combines source trust, field completeness, entity confidence, image match, date and price confidence.

# Indices

Each subindex (RARE‑TCG, RARE‑WATCH, …) is a chain‑linked index of priced constituents in its families, base 1,000 on 2024‑01‑01, equal‑weighted by default. The flagship RARE aggregates the subindices. Repeat‑sales methods are applied where the same identifiable asset trades repeatedly, reducing composition bias. An index publishes nothing until it reaches its minimum number of priced constituents, and every daily value stores its coverage (share of constituents with a fresh valuation).

Benchmarks (S&P 500, gold, Bitcoin, inflation) are shown for context only: collectibles trade infrequently, carry high transaction costs and are valued from observed sales rather than continuous quotes.

# Market capitalisation

Category market capitalisation is estimated as estimated population × representative value summed over priced assets, and is always displayed with its confidence label. It is indicative, never precise.

# Disclaimers

Valuations are estimates. Listing prices are not confirmed transactions. Past performance does not guarantee future results. RareIndex does not authenticate items. Third‑party names and marks belong to their owners.

spb/rareindex