How CardGameInsight prices a card, and why one live read beats four estimates
Every price on the site is either an observed read from a named source or an estimate badged as such. Here is the rule that keeps the two apart, the day it mattered, and the printing mistake it caught.
CardGameInsight stores every price it ever sees and never overwrites one. That is the whole design. A card's page shows the newest reading from each source, a chart of the daily series behind it, and one blended figure at the top. This article is about how that figure is built and what it refuses to do.
Observed beats modelled
Two kinds of rows live in the price table. An observed row is a number read from a named source on a stated date: TCGplayer's market price via PokemonPriceTracker, an eBay sold-by-grade average, a Tradera search. A modelled row comes from the built-in estimator that stands in for any source we do not hold a key for. Modelled rows are badged on the page and carry a flag in the database.
The blending rule is one sentence: once any observed row exists for a card and condition, every modelled row for it carries zero weight. The chart applies the same rule to the whole series, so a card with six real months of history shows only those six months rather than a modelled year stitched to the front of it.
The day this rule was written, Base Set Charizard had one live read at $869.02 and four estimator rows. Averaged together they printed $1,238 as the headline. One real number and four guesses had been treated as five opinions. After the rule, the headline read $869.02, and the estimator rows moved below the fold with their badge.
The printing mistake
On 13 September 2026 a spot check on Snorlax #11 showed a raw market price of $375.66. The unlimited print of that card, the one our catalog describes, traded at $153.26 the same day. The provider's headline price follows what it calls the product's primary printing, and for the Wizards-era sets that print is 1st Edition.
The fix was to price from the printing our row describes rather than the one the product leads with. Then came the part that matters more: 560 cards across Jungle, Fossil, Team Rocket, Gym Challenge, Neo Genesis and Neo Destiny had been stored with 1st Edition prices and six months of 1st Edition history. Those rows were deleted and re-pulled the same day, and the repair script stays in the repository as the record of what was removed. Base Set and every modern set were unaffected, because their printings are separate products.
The lesson is written into the sweep now: every card priced off a non-primary printing is listed in the run's output, so the next surprise of this kind is visible on the night it happens.
The eBay sold row
The provider also relays eBay's completed sales of raw copies: how many sold in the last 90 days, and the average, median, lowest and highest price. That is real transaction data, so it is shown on every card page as its own row, "eBay sold", and refreshed for the whole catalog once a week. It never enters the headline, for a reason the row states: eBay's bucket does not separate condition, and on the Wizards-era products it does not separate 1st Edition from unlimited either. Base Set Charizard on 14 September 2026: 131 sales, median $400, against TCGplayer's near-mint unlimited market of $869. Both numbers are true. They answer different questions.
What "estimate" means
A separate flag marks rows that are observed but weak: asking prices rather than sales, a graded average built on fewer than three sales, a figure the provider's own outlier filter corrected. These rows keep their weight; they carry a note. The distinction is deliberate. A thin sample is still evidence. A model is not.
Currency
Swedish prices are converted at the European Central Bank's daily reference rate, loaded the day of the read and stored with the row, so every converted figure can be reproduced from the row itself. Before this, the site carried a rounded static rate of 0.095 dollars per krona that nobody could source; the ECB rate on 13 September 2026 was 0.1032.
What the headline is not
It is not an appraisal, and it is not a prediction. It is the newest observed reading from the sources that carry weight, averaged by a fixed table of weights that the chart uses too, so the two can never disagree. If you want to know where a number came from, every row on every page names its source and its date.