Card Price Pipeline

Current implementation status (relayer/src/sources/card.ts): three sources. pokemontcg is live and keyless: the TCGplayer market price for a RAW (ungraded) card variant via pokemontcg.io, refreshed daily, with retry-and-backoff and a low/high sanity band (set POKEMONTCG_API_KEY to raise rate limits). A PSA-graded market must never use this source; raw and graded are different assets. manual is an operator-attested USD price for markets whose real source is pending. gradedApi is the shaped client for subscription graded-price APIs and refuses to run without TCG_GRADED_API_KEY. All three convert through a live Pyth SOL/USD read. The multi-source graded estimator described below is the production design those keys unlock. Verified live 2026-07-24: swsh7-215 RAW ($2,137.76) pushed on chain within 2 bps of source. Graded cards are the flagship asset class; see Asset Classes for why.

The underlying unit

A card market’s underlying is one card and grade pair, for example “PSA 10 of card X”. Grade is part of the market identity: a PSA 10 market never blends PSA 9 or BGS 9.5 prices. Those are different assets with different markets.

Sources

The estimator pools four independent source types:

Source	Role
Aggregator card-price APIs (per card, per grade, computed from eBay and TCGplayer data)	Workhorse estimate, refreshed roughly daily
eBay sold listings for the exact card and grade	Ground-truth sale prints the aggregators compress
Collector Crypt sales of the tokenized card, plus their instant-buyback quote	On-chain sale prints, and a capitalized bid-side anchor: the buyback offer (roughly 85 to 93 percent of modeled value) is money the platform will actually pay, so the implied price is expensive to fake
Courtyard sales of the same card	Cross-chain sanity check

Estimator

Per market, per cycle:

Pool all sale prints of the card and grade within the lookback window (default 14 days; longer for slower cards).
Reject outliers by median absolute deviation: prints more than about 3 MAD from the rolling median are dropped. This kills single shill sales.
Compute a recency-weighted median of the survivors.
Apply the sanity band: the estimate is floored by the best capitalized bid (the Collector Crypt buyback implied price, minus a tolerance) and capped by the lowest ask across venues (plus a tolerance). Asks are free to fake, so they only ever cap the estimate. They never set it.

Gates

Two gates can block a push entirely:

Cross-source disagreement: if the aggregator estimate, the eBay estimator, and the tokenized-venue estimator disagree by more than about 15 percent, the price feed (the relayer) holds the last value and alerts. A gap that large is either a real dislocation or someone gaming one venue; both deserve human eyes.
Liquidity gate: if fewer than min_sales prints exist in the window, no new value is pushed. A card that stops trading goes visibly stale instead of tracking noise.

USD to lamports

Card prices are dollar-denominated; the on-chain index is lamports per unit. Each cycle takes one Pyth SOL/USD read, then:

index_lamports = usd_estimate / sol_usd_price * 1e9

Because collateral and settlement are in SOL, synth holders are long the card in SOL terms. If the card is flat in dollars and SOL rallies, the index falls. This is the correct product for a SOL-collateralized market, but interfaces should state it.

Basket indexes

A basket market (for example a vintage PSA 10 index) composes the same pipeline: run it per member card, then take a weighted sum with weights fixed at listing (equal-weight or population-weight). The basket pushes only if enough members pass their gates to meet a quorum. Baskets are the hardest markets to manipulate, since moving the index means moving many independent card markets at once.

Example market config

markets:
  - market: 7xK...   # market PDA
    kind: card
    game: pokemon
    card: { set: "SWSH07", number: "215", name: "Umbreon VMAX Alt Art" }
    grade: "PSA 10"
    lookback_days: 14
    min_sales: 5
    sources: [tcg_api, ebay_sold, collector_crypt, courtyard]

Listing policy favors mid-tier cards with large graded populations and steady weekly sales over thin grails; see Listing a Market.

Core Concepts

Oracle & Relayer

Program Reference

Card Price Pipeline

The underlying unit

Sources

Estimator

Gates

USD to lamports

Basket indexes

Example market config

​The underlying unit

​Sources

​Estimator

​Gates

​USD to lamports

​Basket indexes

​Example market config

The underlying unit

Sources

Estimator

Gates

USD to lamports

Basket indexes

Example market config