The number the world cites — every field reproducible.

GET /api/v1/cards/{catalog_card_id}/price
Authorization: Bearer vsprice_<your-key>

{
  "version": "1",
  "card_id": "9f4e…b3a1",
  "name": "Charizard",
  "set_code": "BASE",
  "fair_price_cents": 38912,
  "ci_lower_cents": 36420,
  "ci_upper_cents": 41440,
  "confidence": "high",
  "sample_size": 24,
  "window_days": 60,
  "observed_at": "2026-05-08T19:14:11.000Z",
  "last_sale_cents": 39500,
  "velocity_per_week": 5.8,
  "sources": ["VaultStore", "VaultStore TCG graph"],
  "source_breakdown": { "vaultstore_sale": 8, "tcg_graph": 16 },
  "currency": "USD",
  "license": "https://vaultstore.live/legal/pricing-license",
  "methodology_url": "https://vaultstore.live/pricing/methodology"
}

• version — pin this on every consumer. New fields are additive within a major version; field semantics never change without a version bump.
• fair_price_cents — trimmed median of weighted comparable sales over the last 30 days. Null when confidence === "insufficient".
• ci_lower_cents / ci_upper_cents— central 95% of the trimmed distribution. Student-t when n < 30, normal-approx 1.96·σ/√n when n ≥ 30. Null with insufficient data.
• confidence — one of high, medium, low, insufficient. Gate UI on confidence !== "insufficient"before rendering a headline number — that tier explicitly says “we don’t know”.
• sample_size — comp count after trimming. Useful as a depth signal independent of the confidence tier.
• window_days — width of the comparable-sales window the trimmed median was computed from. The endpoint reports the window it actually used so consumers can reproduce the math without guessing.
• last_sale_cents — price of the most recent realized comp in the window. Null when none exists. Pair with observed_at for the freshness of that specific sale.
• source_breakdown— per-source comp count over the window, keyed by source identifier. Useful for auditing which feeds drove the median. Don’t render these keys directly to end users — route through methodology labels when displaying.
• velocity_per_week — realized sales per week, averaged over the last 28 days. Null when no sale-typed events exist. A high-velocity card with a tight CI is liquid; low velocity with a wide CI is illiquid even if the median looks confident.
• sources — attributed source labels we hold an active license to cite. Filtered through external_price_licenses; you can render this array verbatim.
• observed_at — most recent comp timestamp in the window. Pair with Cache-Control: s-maxage on the response to drive your own freshness display.
• license — URL of the data-license terms covering this response. Show or link this when you republish the value.
• methodology_url — link to the public formula stack. Surface this alongside the price so end-users can audit your number without leaving your app.

<script src="https://vaultstore.live/embed/price.js" async></script>
<div data-vs-price-card-id="9f4e…b3a1" data-vs-price-theme="dark"></div>

• The widget injects inline-styled HTML into each matching div. No shadow DOM, no global CSS, nothing to clash with the host page.
• Themes: light, dark, and minimal. The script always includes “Powered by VaultStore” attribution and links to the public methodology page.
• SPA hosts can re-scan after route changes by calling window.VaultStorePriceEmbed.refresh().
• Rate limit: 60 requests / minute per origin IP. The script caches at the edge for 15 minutes, and price reads cache for 5 minutes so a popular blog with N embeds renders all of them on first paint without throttling.
• Embed renders fire the standard price_engine_queried capture event with surface="embed" so we can see third-party adoption without collecting personal data from the host page.

• Responses ship with Cache-Control: public, s-maxage=N, stale-while-revalidate=M. N defaults to 300 seconds and is brain-tunable per methodology.
• Per-key rate limits and per-day cost caps are enforced via the standard X-RateLimit-* response headers. Free tier is 60/min, 5k/day, and 100k/month; growth tiers raise the ceiling.
• Errors use the standard envelope: { error: { code, message } } with HTTP status. Auth failures are always 401/403 with code scope_required or tier_required.

• Auditable provenance — every response carries sample_size, observed_at, sources, and a link to the formula stack. A consumer disputing your number reads methodology and reproduces it.
• An honest empty state — when comparables are missing, the API says insufficient and returns null. No fabricated headline numbers, no quietly-wide CIs hiding the lack of signal.
• Velocity — published alongside the price, not an afterthought. You learn liquidity at the same time you learn the median.
• A drift watchdog — every 24 hours we compare our published number against external comparables on the same cards. When the gap exceeds 5% across 50+ cards an action-tier insight reaches the methodology team. The number gets investigated before the next refresh ships.