How the Prism grade fuses five signals into one number.

01Why fuse signals into one grade instead of five charts?

Why one grade instead of five charts?

The bottleneck on the floor of a card show isn't data — it's fusion. Anyone with a phone can pull a TCGplayer recent-sales chart in thirty seconds. Almost no one can hold realized comps, asking-price spread, sentiment, pop counts, and a forecast in their head at the same time, in five seconds, while the other party is waiting for an answer.

The job a serious operator is actually doing in that moment is fusion — collapsing five overlapping signals into a single yes-or-no. The Terminal is built for that job, and Prism is the part that does the collapsing. The output is one letter, one sentence, one confidence number. You can drill into any of the underlying vectors at any time — every cell on the page is interactive — but the default is a verdict, not a worksheet.

This is the same principle behind the founding post: a desk works because every signal lives on one screen and the trader never has to leave the cockpit. Prism is the screen.

02What are the five Prism signals and what does each measure?

What does the Prism grade actually measure?

Prism is five independent vectors. Each one runs on the same realized-price feed but answers a different question. Each is computed on its own and weighted by recency and confidence before the fusion step.

Liquidity

Realized sales volume over rolling 7- and 30-day windows. Liquidity is the gatekeeper. Thin liquidity widens the confidence interval on every other vector and caps the upside of a grade — a card that has traded four times this month cannot earn an A+ no matter how clean the rest of the picture looks.

Volatility

Standard deviation of realized prices over the trailing 30 days, normalized against the comparable peer set. High volatility pulls a grade toward neutral. A card that traded between $90 and $160 this month doesn't have a defensible "fair" price; the model refuses to take a strong position on it.

Floor Strength

How tightly the 5th-percentile of realized prices is holding. A rising floor is the strongest bullish signal in the model — it means even the rushed sellers are getting paid more. A collapsing floor is the strongest bearish signal, even when the median looks fine, because medians are slow. For the supply-side context — how set supply dilution shapes the floor signal — see the 2026 English-set ranking.

Spread

The gap between the asking price you're testing and the 30-day realized median. Spread is the biggest input to short-horizon grade swings. A 20% positive spread on a liquid card with a strong floor is the textbook A-grade buy. A 20% negative spread is the textbook "pass."

Momentum

Slope of the realized price series, blended with watchlist movers and forecast direction. Momentum is the only forward-looking vector in the model. It is also the noisiest, which is why Prism weights it lightly and refuses to let it drive a grade on its own.

Five vectors. Five questions. One picture per card.

03How does Prism weight and fuse the five signals?

How does Prism weight the five signals?

Each vector is weighted by recency and confidence. Older data fades. Thin samples count for less. The fusion step then runs an asymmetric rule the model is explicit about: a single bad signal can floor the grade, but no single great signal can ceiling it. This is the most important property of Prism, and it's the one that makes the grades feel different from a back-of-the-envelope average.

Two textbook cases:

• A great spread on an illiquid card still grades B at best. If a card has only traded twice in the last 30 days, the liquidity vector is shouting "I don't know" loud enough to cap everything else. The grade reflects that the data isn't there to defend an aggressive call.
• A weak floor with elevated sentiment is the textbook "trim" signal. Sentiment can stay hot for weeks while the floor quietly collapses. Prism is built to catch that asymmetry — the floor vector pulls the grade down even when momentum and spread look fine on the surface.

The fusion output is three numbers in trench-coat: a letter grade from A+ to F, a plain-language one-sentence verdict ("Buy. Strong spread vs. 30-day realized."), and a confidence score between 0 and 1. The confidence score is the part most users underweight on first encounter. A B+ at 0.92 confidence and a B+ at 0.41 confidence are not the same call — the second one is the model telling you to look harder before you act.

04Where does Prism's data come from?

Where does Prism's data come from?

Realized comps are aggregated from public sales venues across the major TCG marketplaces and refreshed every few minutes during US market hours. Pop reports are synced from PSA, BGS, and CGC, daily. The sentiment input blends realized-vs-asking spread compression with movers breadth — it deliberately ignores social-media noise.

That last clause is a design choice, not a limitation. Social sentiment is a momentum amplifier in collectibles markets; it's also a coordinated-pump amplifier. We treat the realized-vs-asking spread as the honest signal and treat movers breadth as the breadth check. If the spread is compressing across a wide slice of the market, that's real sentiment. If it's a hashtag, Prism doesn't care.

The forecast is a separate model, surfaced inside the Terminal but not directly fused into the Prism grade. It's a quantile model that returns 5th, 50th, and 95th percentile price paths over 1-, 3-, and 12-month horizons. It is not a prediction. It is a range you can use to reason about position sizing. The point of separating the forecast from the grade is to keep the grade about what is true now, and let the forecast carry the load of what might be true later.

05Where the model holds back.

What won't the Prism grade tell you?

Prism is a data summary, not a recommendation. It does not tell you what to buy. It does not tell you what to sell. It tells you what the realized market is doing right now, in one number, with a confidence score attached. The Terminal is a screen. You are the trader.

The reasons for that line in the sand are practical. Recommendation engines invite regulatory attention this product does not need and does not deserve. They also flatten what makes a desk useful — the operator's read of context the model can't see. We'd rather ship a sharp grade and a wide-open verdict line than a soft grade and a buy button.

When the confidence score is low, the model is telling you to drop down a level — look at the underlying vectors, see which one is dragging, decide whether the picture matches what you're seeing on the floor. The grade is the default. The vectors are the appeal. Both are always available; nothing is hidden inside a black box.

The next two posts in this thread will cover the rest of the founding stack: what we learned from the de-financialize pass (why we refused to ship a recommendation engine) and the broader investment thesis hub (why this market deserves real price-discovery infrastructure regardless of where you sit on the speculate-versus-collect spectrum).