What is a conformal risk forecast for trading cards?

A conformal risk forecast is TCG Oracle's default price forecast model. Instead of assuming a mathematical model of how prices move, it calibrates prediction bands on thousands of real card price histories, so a 90% band is built to actually contain the true price about 90% of the time. It is distribution-free (no model assumptions), regime-aware (calm and volatile markets get different bands), and fully deterministic — the same card and date always produce the same numbers, so anyone can reproduce and verify a forecast. Each forecast returns a point estimate, probability the price rises, 50% and 90% bands, VaR95/VaR99 downside risk, and two letter grades: Safe-Hold (capital preservation) and Momentum (30-day direction). Available free at /api/v1/forecast.

How does Monte Carlo simulation work for trading cards?

(Suspended since 2026-09-12 while the USD price panel is frozen — /api/v1/simulate returns status suspended and does not charge.) Monte Carlo simulation is TCG Oracle's opt-in scenario engine (model=gbm or model=merton on /api/v1/simulate; the default model is the conformal risk forecast). It uses random sampling to model the range of possible future card prices across two stochastic finance models: GBM (Geometric Brownian Motion — the foundational continuous-time model for asset price dynamics) and Merton Jump-Diffusion (extends GBM with Poisson-driven jumps to capture sudden price spikes from hype events, reprints, or format changes). Each simulation runs 20,000 paths and returns full percentile bands (5th through 95th) with VaR/CVaR risk metrics and complete model parameter transparency.

What trading card games does TCG Oracle support?

TCG Oracle covers 456K+ products across 25 games including: Pokémon, Magic: The Gathering, Yu-Gi-Oh!, Digimon, One Piece, Flesh and Blood, Lorcana, Dragon Ball Super, Cardfight!! Vanguard, MetaZoo, Union Arena, Star Wars Unlimited, Gundam, and more. USD prices originate from TCGPlayer and are currently frozen at their last good date; Japanese-print dealer quotes (24 games, both an ask and a buyback bid) refresh every morning. Graded comps come from eBay listings.

What is the best API for trading card price prediction?

TCG Oracle is the only collectibles API with calibrated, reproducible conformal risk forecasts — honest VaR and Safe-Hold/Momentum letter grades, deterministic so anyone can verify the numbers — plus opt-in stochastic models (GBM and Merton Jump-Diffusion) seeded from the public drand randomness beacon. It covers forecasting, graded ask medians, a daily two-sided Japanese dealer board, sports and on-chain prediction records, and every forecast is Merkle-committed before its outcome is known so the track record cannot be edited after the fact. Every endpoint is free right now — pay-per-call is paused — and GET https://oracle.the-undesirables.com/ is authoritative for the live endpoint list. Available via MCP (Model Context Protocol) and A2A Agent Cards for autonomous agent discovery.

TCG Oracle — Financial Intelligence for Collectibles

The only trading card API with conformal risk forecasting (Monte Carlo opt-in), graded ask medians, and a daily two-sided Japanese dealer board. 456K+ products across 25+ games.Every endpoint is free right now — pay-per-call is paused. The live endpoint list and payment state come from GET / on the oracle, which is authoritative.

What Is Live Today

Every endpoint below is free. Pay-per-call was paused on 2026-10-02: the x402 rail stays installed and can be switched back on per route, but nothing charges today.

Japanese dealer board — /api/v1/jp/summary publishes the two-sided Japanese panel as aggregates: totals, bid/ask spread by price band, per-game depth, and day-over-day movement, with the day's on-chain root. Per-card quotes are one at a time and the source is not named.
Graded ask medians — /api/v1/graded returns PSA/BGS/CGC medians by grade. These are active-listing asks, not sold prices, and every response declares its price_basis. Realized sales land in /api/v1/graded/sales as they become available.
Conformal forecasts — the free board at /api/v1/forecast and per-card at /api/v1/forecast/{product_id}, Merkle-committed before outcomes are known.
Loan terms — /api/v1/loan-terms and the universe/preview routes, for collateral maths against graded cards.
Sports, souls and stamps — league boards, each soul's on-chain prediction record, and a daily Bitcoin Stamps market panel rooted on two chains.

Discontinued. AI vision card grading (/api/v1/grade, /api/v1/grade/upload), the grade-or-not decision engine, batch triage and the loan monitor have been retired and answer 410. They are no longer offered, and nothing on this site should be read as selling them.

The Default Model: Conformal Risk Forecasts

Most price "predictions" assume the market follows a tidy mathematical model. Cards don't. TCG Oracle's default forecast is conformal: instead of assuming how prices move, it calibrates its bands on thousands of real card price histories — so a 90% band is built to actually contain the real price about 90% of the time. Honest uncertainty, not false precision.

Distribution-free — no model assumptions to be wrong about. The bands come from measured forecast errors on real cards.
Regime-aware — calm markets get tight bands, volatile markets get wide ones. A drift-spike filter flags cards mid-mania where direction is untrustworthy.
Deterministic & reproducible — the same card on the same date always produces the same numbers. Anyone can re-run a forecast and verify it — which is exactly what lets soul prediction track records be judged fairly.
Graded output — every forecast returns a point estimate, prob_up, 50%/90% bands, VaR95/VaR99, and two letter grades: Safe-Hold (capital preservation, absolute scale) and Momentum (30-day direction).

Try it free — no key, no payment: the forecast board at /api/v1/forecast covers the top ~200 liquid cards, and /api/v1/forecast/{product_id} works for any of the 456K+ cards.

Opt-In: Monte Carlo Simulation for Trading Cards

Suspended since 2026-09-12. The USD price series this simulation calibrates on froze on 2026-09-07; /api/v1/simulate answers {"status":"suspended"} and does not charge until a live USD series exists. What follows describes the model as built.

When you want scenario simulation instead of calibrated bands — "what does a hype-jump world look like?" — switch to Monte Carlo with model=gbm or model=merton. Two stochastic finance models borrowed from Wall Street options pricing, seeded from the public drand randomness beacon so the draws are provably fair:

GBM (Geometric Brownian Motion) — The foundational continuous-time model for asset price dynamics with constant drift and volatility. Ideal for stable, liquid markets.
Merton Jump-Diffusion — Extends GBM with Poisson-driven jumps to capture sudden price spikes from hype events (a YouTuber opens a rare card, a reprint tanks supply, a format rotation bans a staple). Returns VaR/CVaR risk metrics alongside percentile bands.

Each simulation runs 20,000 paths and returns full percentile bands (5th through 95th) with complete model parameter transparency. You see exactly what drift, volatility, and jump parameters were used.

Supported Games

TCG Oracle covers 456K+ products across 25 games including Pokémon, Magic: The Gathering, Yu-Gi-Oh!, Digimon, One Piece, Flesh and Blood, Lorcana, Dragon Ball Super, Cardfight!! Vanguard, MetaZoo, Union Arena, Star Wars Unlimited, Gundam, and more.

API Access

TCG Oracle is available through 9+ discovery surfaces, each speaking a different protocol:

MCP — Model Context Protocol for Cursor, Claude Desktop, VS Code, Perplexity
x402 — the micropayment rail is installed and verified end-to-end, currently PAUSED (every route free)
A2A — Google Agent-to-Agent protocol for peer discovery
npm / PyPI — Package managers for ElizaOS and Python agents
REST API — OpenAPI spec at /docs

Built by The Undesirables LLC. This page describes the service; for what is live right now — endpoint list, suspended routes, payment state — GET / on the oracle is authoritative and always wins over anything written here.