Every TVP read carries a footer linking here. We publish:
Per OS v8.0 §1.5: "Inaccurate data that looks accurate is worse than no data." This page is the antidote.
A TVP read flows through five stages. The same five stages every time.
Visual recognition via Anthropic Claude (Haiku 4.5 for quick reads, Sonnet 4.5 for research). Returns: player, year, set, parallel, serial, grade, vision-confidence (1–5). We do NOT publish a verdict when vision-confidence < 2 — we refuse with LOW_VISION_CONFIDENCE and ask for a better photo.
Every identified card is routed through the published TVP rule layer (Operating System v8.0 Rules 1–88) using empirical multipliers from Multiplier Reference v1.x. The engine determines active rules, expected price range, confidence band, and catalyst state. The engine produces structured ground truth — the LLM's job becomes "produce prose explaining the engine's read" instead of "produce a number." This is the structural fix for hallucination.
Comparable sales are pulled in parallel from multiple sources: Card Ladder (when authenticated), eBay Sold listings (public scrape), 130point.com (via ScraperAPI), and eBay Browse API for active asks. We prefer cleared sales over active asks. We attribute the source on every comp citation.
When live sources return fewer than 3 cleared comps, the engine consults a hand-vetted curated reference dataset (~110 cards as of May 2026). Curated comps are flagged with comp_source: "manual" so the user knows we are using vetted reference data, not pretending to scrape live.
The LLM produces a price midpoint and range, citing specific comps from the ladder. Every assertion must reference at least one comp. If comps are insufficient (<3 cleared in last 90 days) OR the LLM's read falls outside the rule engine's expected range without supporting comps, the engine returns INSUFFICIENT_COMPS or RULE_ENGINE_MISMATCH instead of a number.
This refusal is a feature, not a bug. Card Ladder gives you a 1-out-of-5 confidence and lets you draw your own conclusion. We refuse to publish a number we can't defend.
Every read carries a TVP Score (1–10) decomposed into five axes: Recency, Quantity, Cross-Source, Venue Tilt, Catalyst-State. Floor rule: any axis ≤2 caps the final score at 5. Modeled on Bloomberg's BVAL Score methodology. The TVP Score is a measure of INPUT QUALITY, not price accuracy.
Per Investment Advisers Act §202(a)(11)(D) and Lowe v. SEC, 472 U.S. 181 (1985):
True Value Protocol (TVP) is a publication of bona fide and impersonal price-truth analysis. TVP does not provide individualized investment advice. TVP does not custody, lend against, or take order flow on cards. TVP does not own a marketplace, grading service, or breaker operation. TVP does not represent any seller, buyer, or third party in transactions.
We do not:
Why this matters: the Publisher's Exclusion is the structural moat. Every other major actor in the hobby has a structural conflict of interest. Fanatics owns marketplace + Whatnot. Collectors owns PSA + SGC + Card Ladder + Goldin. Alt owns lending + custody + marketplace. TVP does not. That is the durable competitive advantage in an industry where every other read is plausibly conflicted.
Disagree with a TVP read? Submit a challenge — email challenge@tvpcards.io with: card summary, the read you disagree with, your cited evidence (sale link, comp screenshot, etc.).
Triage SLA: 7 calendar days standard, 48 hours during catalyst-active windows. Resolution classes (all public): NO_CHANGE, SCORE_ADJUSTED, METHODOLOGY_AMENDED, PENDING.
When a TVP read materially misses (actual realized price falls outside the published range by >20% within 90 days, OR a recommendation is contradicted by ≥3 catalyst events without updating), we publish a Confession entry within 30 days. Each names: the original read, the miss, the contributing factors, the methodology adjustment.
The credibility comes not from being right. It comes from publishing the rating, the evidence, and the correction in a permanent dated record.
Every TVP read response includes: verdict_id · active rules · routing path · engine's expected range · TVP Score axes · every comp cited (with date, price, grade, source, URL) · primary source · the R45 disclosure.
Aggregate scorecard: tvpcards.io/receipts. Public catalog of curated cards: tvpcards.io/verdict/catalog. Live demo: tvpcards.io/demo.