Cambridge TCG will not make a raw historic database its product. The intended product is decision support: a private engine may examine rights-cleared evidence, estimate the economics of one candidate card, and emit a short-lived signal that helps a trader decide what to inspect next.
The signal is named potential deal, never arbitrage. A real arbitrage requires simultaneously executable entry and exit prices plus the ability to complete both legs. A valuation estimate, daily aggregate, or open listing does not establish those facts.
Current boundary — 1 September 2026. The public opportunity-signal/v1 contract and private-engine interface exist. No scorer, source adapter, database reader, signal API, trader application, alert delivery, subscription, or automated purchase path is active. This page documents a foundation, not a live claim that Cambridge finds deals today.Where this lives in code. The public contract and strict validator live at packages/opportunity-signal/. The proprietary engine is intentionally absent from this public repository. The connection record is docs/connections/the-opportunity-signal.md.rights-cleared private evidence → private valuation and policy engine → strict opportunity-signal/v1 projection → trader-facing application (not built)The public contract says what may cross the last boundary. It does not reveal how Cambridge weighs sources, maps products, filters anomalies, estimates fair value, calibrates thresholds, or evaluates outcomes. Those are the trade-secret layer and need a separate private service.
Secrecy does not create source rights. A purpose-specific policy decision must permit derived subscriber signals before any evidence reaches the engine. Permission to store or display a price is not silently widened into permission to produce analytics. The rights decision carries a SHA-256 digest of the exact candidate, valuation, cost, and currency evidence it reviewed; the public boundary recomputes it first.
Version 1 evaluates one physical copy. Candidate and valuation must carry matching Cambridge-shaped SKU, condition, and normal-or-foil labels. Quantity is fixed to one. The contract verifies syntax and equality only; the upstream mapper and provider own canonical catalog identity and real condition comparability. They must not let an all-conditions aggregate silently stand in for a near-mint listing.
A candidate reference must have the Cambridge-minted shape ctcg_cand_ plus 22 base62 characters. Syntax cannot prove semantic origin: the composing service must generate it randomly and must not derive it from a seller, order, marketplace listing, URL, or other personal/source identifier.
All version-1 amounts enter as safe integer GBP minor units: 4250 means £42.50. Floating-point money, negative costs, unsafe integers, implicit currency conversion, and a zero asking price are invalid.
The asking price is only the first cost. The contract also names:
Each cost is known, estimated as a bounded range, explicitly not applicable with a closed reason, or unknown. Unknown never means zero. If a required cost is unknown, the platform cannot call the spread net.
“Net” means net only of those enumerated transaction costs. It does not automatically include time, financing, storage, grading, authentication, insurance, returns, fraud loss, income tax, or opportunity cost.
Currency adapters round acquisition costs upward and possible exit proceeds downward before values enter this contract. No silent 1:1, inverse, or chained exchange rate is inferred.
The private engine works with a low, midpoint, and high possible-exit estimate. That range is not an executable quote and is never returned as a fair-value feed.
The provider returns no economics. Its classification is bound to the SHA-256 digest of the full validated request, while the public projector independently verifies the non-secret cost arithmetic. Only a potential-deal result carries coarse conservative spread and margin bands. A not-qualified or unavailable result carries no valuation amount, spread, or margin estimate. Valuation time, confidence, and liquidity may remain unless rights are denied; a rights denial nulls all three. Bands reduce reconstruction risk but do not eliminate inference, so any future delivery layer also needs Cambridge-minted candidates, rate controls, and an explicit leakage budget.
Low, medium, and high describe evidence quality and comparability. They are not the probability of profit. Version 1 emits no “81% likely” claim: Cambridge has not claimed an outcome-calibrated corpus capable of supporting one.
Low confidence cannot produce a potential deal. Aggregate price-guide evidence is labelled aggregate_not_trade_tape; interpolation, short history, and sparse history remain named risks. An aggregate is never relabelled “last sold.”
Liquidity is separate from price. A changing average or one low listing does not prove sale velocity. Version 1 therefore preserves unknown. Low, medium, or high requires a separate current evidence receipt, but the wire validator cannot prove its real-world comparability; that assertion belongs upstream. No band promises a number of days to sell.
Every input distinguishes source-stated time, Cambridge retrieval time, and evidence expiry. Retrieval time cannot replace source time to make a claim look fresh.
A signal expires at the earliest expiry of every required input and is capped to the platform's 60-second market-signal delivery budget. This limits reuse; it does not turn daily evidence into live evidence. Expiry also does not reserve the listing or prove continued availability.
| Classification | Meaning | Economics delivered? |
|---|---|---|
potential_deal | Accepted by a private policy for further human inspection. | Coarse conservative spread and margin bands only. |
not_qualified | Computable, but below the private policy. | No. |
unavailable | Rights, identity assertions, costs, evidence, time, currency, numeric safety, provider failure, or provider-contract validation prevented a qualified result. | No. |
There is no arbitrage classification in version 1.
Availability, buyer demand, authenticity, physical condition, tax treatment, fees, shipping, returns, fraud, and time to sell may differ from the evidence available at evaluation time. Historical relationships may not persist.
The first contract assumes GBP, one matching Cambridge-shaped SKU label, one physical card, one condition, one finish, and a trader able to inspect physical and commercial risk themselves. It does not yet model sealed lots, graded slabs, bundles, shared shipping, collective purchasing, non-GBP settlement, automatic execution, or a holder whose goal is cultural preservation rather than resale value.
Related methodology: cross-source pricing, the market mirror, and legacy channel pricing.
v1 — 2026-09-01. Established the source-independent wire contract and the public/private engine boundary. No executable signal service was activated.