What a PSA 10 Is Selling For Right Now, and When the Median Lies

Two numbers describe the same slab. One is the median of every PSA 10 sale we kept for that card. The other is the median of the sales inside a trailing window: 90 days when 90 days holds at least three sales, and wider when it does not. Ask what a PSA 10 is worth and you get whichever of the two the surface in front of you leads with, and across the 8,862 PSA 10 cards priced on 2026-10-11 the pair sits more than 50% apart on 742 of them. Bibarel GG25 from Crown Zenith Galarian Gallery reads $295.00 on eleven sales in 90 days and $84.99 across the 39 sales in its whole kept pool. Mega Dragalge ex 116/083 reads $64.99 on six, against $140.00 on eleven. Same card, same grade, same feed, two answers. The three notes closest to this one answer different questions: Which Market Price? sorts out the three numbers a raw card carries, The Ask Is Lying to You sets a raw listing against a graded sold median, and How Many Sales Is a Price? censuses the sample counts behind our PSA 10 series. None of the three sets a card's recent graded median against its own long one. That is the only comparison in here, and it comes with both sample sizes printed.

The table is not the whole population. It is the eight widest gaps below and the seven widest above, drawn from the 3,972 cards whose recent figure came from a true 90-day window and rested on five or more sales, with one card pulled out by hand for a reason given below. Call that cut the gated pool. Every steepest and widest in this note is ranked inside it, on the day named beside it, and the cut holds a different number of cards each day because the cohort that refreshed is different. The gate is there because a thin window and a wide one are different claims, and ranking them together puts a card with three sales somewhere in the last year at the top. The full population is the 8,862, and the two columns in it are not two data sets. The recent median is a subset of the long one: the same kept comps, after the same bulk-lot collapse and the same outlier trim, counted again with everything older than the window, or carrying no sale date at all, dropped. On Mega Dragalge ex 116/083 the six recent sales sit inside the eleven, and the eleven have a median of $140.00, so at least three of the five outside the window sold at $140.00 or more. That is all a gap ever is.

The window is also not always 90 days. The rule asks for at least three dated sales inside 90 days, widens to 180 when it cannot find them, then to 365, and the figure carries the window it actually used. Of the 8,862 cards, 5,556 got a true 90-day read, 2,269 needed 180 days and 1,037 needed a full year. Anyone who reads every recent median as a quarterly figure is wrong on 37.3% of the book.

In the 90-day cohort, and for a reason that is arithmetic rather than market behaviour. When the window widens to a year it often swallows the entire kept pool, and then the two medians are not two readings of one price. They are the same number twice. Of the 1,037 cards on the 365-day window, 755 have a recent median identical to their long-window median to the cent, and on 726 of those the sample counts match as well. Across all 8,862 cards, 2,008 pairs are exactly equal. Restrict to the 7,340 cards whose long pool is strictly deeper than its recent window, so that the two figures genuinely cover different stretches of time, and the median gap rises to 10.4% with 10.1% past the 50% bar. That is the honest denominator for anyone asking how often the long figure has fallen behind: one card in ten, not one in twelve.

Here is the correction that stops this being a stock-picking list. Across the 8,862 cards the signed gap has a median of plus 2.8%. 5,069 of them sit above their own long-window median, 2,008 land on it to the cent and 1,785 sit below, which is 57.2% above, 22.7% level and 20.1% below. On the 5,556 with a true 90-day read the drift runs harder, median plus 5.4% with 63.6% above. So a 20% premium is not the going rate. Across the same 8,862 cards, 2,027 sit more than 20% above their long pool. That is 22.9% of the book, which puts a plus-20% card in the top quarter of the distribution rather than in the middle of it, and inside the 7,340 whose long pool is strictly deeper it is 27.6%. The far end is where the lopsidedness lives. Ten of the 8,862 read more than 50% below their long pool against 732 above, and under the table's own gate the split is one against 303. Part of that is the bar: 50% is asymmetric in ratio terms, because clearing it upward needs a recent median above 1.5 times the long one while clearing it downward needs a long median above 2 times the recent. On a symmetric one-and-a-half-times bar the full-book split is 732 to 49 and the gated split is 303 to 14. The direction holds either way. The failure people describe out loud, a graded median stuck high above what the card now trades for, is real and we can name the cards it happens to. It is just the rare direction.

Our card pages print the long-window median beside the recent one only when the two differ by more than 50%. That bar is a publishing rule, not a property of the market, and it means eight of the fifteen rows above are fully visible on the card page while seven are not. For Mega Zygarde ex 113/080 at -48.1% the page shows $69.00 on sixteen sales and says nothing about the $132.99 behind it. So the eight rows past the bar are checkable one card at a time on their own pages, and the seven under it are not: their long-window figures live in this note and in the tape behind it, and nowhere a reader can click. There is no board for the comparison either, and the ranking exists only inside this note until we build a surface for it.

The long-window column is also not all-time, and calling it that would be the easiest mistake to make here. It is the provider's current set of kept comps for that product, with no fixed left edge, and the row does not record how far back the oldest one reaches. Worse for anyone reading it as history: the pool re-forms. Of the 8,835 cards carrying a long-window median on both 2026-10-06 and 2026-10-12, the comp count grew on 5,805, the median moved more than 20% on 419 of them and more than 50% on 85. Umbreon H30 from Skyridge is one case: its long-window PSA 10 median read $100,000.00 on five kept sales on 2026-10-06 and $55,200.00 on eleven on 2026-10-08. The card did not fall 45% in two days. The pool got deeper and the middle of it moved. Read that column as a deeper sample of the same feed, never as a price history.

Of the 9,667 PSA 10 rows written on 2026-10-11, 805 get no recent median at all, because fewer than three dated sales landed inside a year. Those are not cheap cards and they are not errors. 597 of them carry a recorded last sale, median 94 days old, the oldest 596 days old. Charizard 103 from Stormfront is the dearest of them: two kept sales, a long-window median of $32,400.00, and a last recorded sale of $45,600.00 on 2026-03-02. That $32,400.00 is not on its page either, under the same publishing rule: the page leads with the last sale, and $45,600.00 against $32,400.00 is 40.7% apart, inside the 50% bar. Ash Greninja EX XY133 is the next dearest and the same shape, two kept sales, $26,350.00, last sold $48,000.00 on 2026-09-21. The other 208 rows carry nothing dated at all, which is why they hold a provider aggregate and no window. An absent median means we would not stand behind one. It does not mean the card is worthless, and it does not mean nobody is buying.

This tape is written by an hourly cron that refreshes a rotating slice of the book, so one day is a cohort of whatever got refreshed and not the whole keyspace. On 2026-10-11, a complete day, it wrote 14,781 distinct cards across all grades and 9,667 PSA 10 rows. On 2026-10-12, read through the 21:00:45 UTC write, the last one that had landed when we closed this file, it held 11,901 cards and 9,687 PSA 10 rows. Over 2026-10-06 to 2026-10-12 the PSA 10 rows written per day ran from 9,074 to 9,687, against 10,045 distinct PSA 10 cards for the week as a whole, so no single day holds the book.

A share that holds at 8.4%, 8.5% and 8.6% across a complete day, a day closed at 21:00 UTC and a week of freshest rows is not an artefact of which cards happened to refresh. That is the result that matters most in this note. The cohort problem is real and it moves individual rows hard. Mega Greninja ex 114/083 is the widest discount in the gated pool on 2026-10-12 at -63.8%, $250.83 on 32 sales in 90 days against $692.80 on 91, and it has no row at all on 2026-10-11, which is why it is absent from the table above and present here. Rows move. The share does not.

It cannot tell you which of the two numbers is right. A 90-day median on five sales is thin and a long pool with no left edge is soft, and this note ranks the distance between them rather than adjudicating it. For the 2,008 cards whose medians are identical, the comparison is empty by construction and the only honest reading is that the window never bit. It cannot resolve printings either. The tape is keyed per TCGplayer product rather than per printing, so a product whose collector number is shared by several printings can blend them into one pool, which is why Base Set Mewtwo is out of the table. Each of the 20 cards linked in this note sits on a TCGplayer product id that exactly one catalog row claims, which we checked one id at a time. The catalog does hold the other shape elsewhere, a Japanese set written once in Japanese and once in English with a single product id on both rows, and we have two confirmed cases of it, product ids 665870 and 665919. The ranking is per product, so nothing double counts, and these links pick the English row. And no price in this note is a TCGplayer raw quote. Every figure is a median over graded sold comps from our own daily recording, printed with its window and its sample count.

Source: graded_snapshots, grade 'PSA 10', read with the dated model columns recent_median, recent_n, recent_window, all_median, all_n, last_sale_price and last_sale_day, written by the hourly cron from the provider's own dated sold comps. Primary population: day bucket 20737, which is 2026-10-11, the last complete day before publication (snapshot_at 00:00:45.358 to 23:00:45.117 UTC, 24 distinct write stamps, 109,361 rows over 14,781 distinct cards at all grades, 9,667 of them PSA 10, one row per card). Exclusions on that day: 0 rows in the catalog-id keyspace, which is stale forever and is dropped on sight by requiring an all-digit card_key; 805 rows with no recent_median, being the cards with fewer than three dated sales inside a 365-day window, of which 597 carry a last sale and 208 carry nothing dated; 0 rows with a recent median and no long-window median. That leaves 8,862 cards ranked. The gap is (recent_median minus all_median) divided by all_median, and the 50% bar is HEADLINE_DIVERGENCE in packages/core/src/gradedmodel.ts, the same bar the card pages use to decide whether to print both figures. That bar is asymmetric in ratio terms, so the up and down counts either side of it are reported again against a symmetric 1.5x bar wherever the lopsidedness is named. The gated pool applies two further gates to the 8,862, recent_window = 90 and recent_n of at least 5, leaving 3,972 cards on that day, and the table of widest disagreements then removes Base Set Mewtwo by hand for the printing reason above. The 3,972 count and the splits quoted against it, 303 to 1 and 303 to 14, include Mewtwo; the hand removal is from the table only. The same gate on day 20738 holds 3,978 cards, and that is the population behind the 2026-10-12 widest discount. Every steepest and widest in this note is ranked inside the gate on the day printed beside it. Check run: all_n is at least recent_n on all 8,862 rows, so no recent sample exceeds the pool it came from. The provider's own aggregate in the median column is not a median and appears in no figure here. Second and third populations, both reported: day 20738, which is 2026-10-12, read through its last landed write at snapshot_at 1791838845358, that is 21:00:45.358 UTC (98,817 rows over 11,901 distinct cards at all grades, 9,687 of them PSA 10, 11 catalog-id keys dropped, 804 with no recent median, 8,872 ranked); and the freshest row per card_key across days 20732 to 20738 (10,045 distinct PSA 10 cards, 11 catalog-id keys dropped leaving 10,034, 876 with no recent median, 9,158 ranked). A day 20738 figure is pinned to that epoch because the 21:00:45.358 write lands 4,986 rows under a single timestamp, so a cutoff inside it is not a cutoff; immediately before it the day held 93,831 rows over 11,064 cards. Pool-drift figures compare day 20732 with day 20738 on the 8,835 cards carrying a long-window median on both. Medians are the mean of the two middle values on an even sample, and every published median here is unchanged under the nearest-rank convention. Every query filtered on day, which uses graded_snap_day_idx; no LIKE predicate and no api_cache read was involved. No PSA population table was joined, so no gem rate is read here, as a zero or otherwise.