Track record
BacktestedEvery signal our model would have fired across each card's price history, scored by what actually happened next. A Snipe is a “hit” if price was higher 30 days later; a Sleep hits if price was lower.
Live track record
Out of sampleEach figure carries what those same cards did on their own, over the same window and the same 8-sale bar. That comparison is the number that matters: prices rose broadly over this period, so a Snipe rate near its baseline is the market rather than a result. It cuts both ways — a Sleep rate that looks lower than a Snipe rate can be clearing a much harder bar.
Snipes split by strategy · Value = into weakness · ↗ Trend = ride the winner
Value entries @14d
17%
68 cards · med -2% · avg -2.3%
These cards rose 38.8% of the time on their own · -21.8 vs that
Trend rides @14d
57.9%
464 cards · med +0.9% · avg +3.9%
These cards rose 38.8% of the time on their own · +19.1 vs that
Sleep @14d
79.2%
275 cards · med -2.8% · avg -2.4%
These cards fell 61.2% of the time on their own · +18 vs that
Value entries @30d
19.2%
55 cards · med -5.1% · avg -4.6%
These cards rose 47.1% of the time on their own · -27.9 vs that
Trend rides @30d
68.2%
408 cards · med +3.1% · avg +5.9%
These cards rose 47.1% of the time on their own · +21.1 vs that
Sleep @30d
83.3%
230 cards · med -6% · avg -6.3%
These cards fell 52.9% of the time on their own · +30.4 vs that
Counts are distinct cards, not daily snapshots. We show the median return next to the average, since one mispriced grail can swing an average. A card with no fresh sale near the horizon stays unscored rather than guessed. A resolved call is never revised. It reflects the prices available on the day it was scored, and later historical backfills do not change it.
These figures leave out calls we have since corrected. See corrections
Corrections
When part of the live record turns out not to have measured what it claimed, we mark those calls and stop counting them. They stay stored exactly as published. Each correction below says what went wrong, shows the figures with and without those calls, and gives a test you can check yourself.
Published Oct 8, 2026
Calls logged at another card's price
Some recent sets have a rarer version of a card with the same name and number, printed with a Poke Ball or Master Ball pattern and sold as its own product. Until late August 2026 our price feed matched cards by number, so on some days the rarer version's price was filed under the ordinary card. These calls were logged at that price.
When the feed was changed to match by product, each card got its own price back. For these cards that was a small fraction of the price we had logged, and the record scored the difference as a fall of more than 99%. No card fell that far. The calls measured a filing error. They were not right or wrong about the market.
How to check it
A card with steady sales does not lose 99% of its price in two weeks. Among these calls, 22 of 22 14-day results and 22 of 22 30-day results show a fall of 99% or more. For every other call counted on this page, it was 0 of 10,863 and 0 of 7,425. 14 of the 22 14-day results and 14 of the 22 30-day results counted as a hit.
What changed
We marked 44 calls on 2 card versions as corrected. They no longer count toward the live figures above. It covers calls logged through Aug 26, 2026. A call logged on the same cards after that date counts as usual. This correction moves some of the figures above up and some down. We would have made it the same way if it moved them the other way.
| Live figure | Counting them | Not counting them |
|---|---|---|
| Value entries @14d | 17%68 cards · rose 38.7% on their own | 17%68 cards · rose 38.8% on their own |
| Trend rides @14d | 57.8%465 cards · rose 38.7% on their own | 57.9%464 cards · rose 38.8% on their own |
| Sleep @14d | 79.3%276 cards · fell 61.3% on their own | 79.2%275 cards · fell 61.2% on their own |
| Value entries @30d | 19.2%55 cards · rose 47% on their own | 19.2%55 cards · rose 47.1% on their own |
| Trend rides @30d | 68%409 cards · rose 47% on their own | 68.2%408 cards · rose 47.1% on their own |
| Sleep @30d | 83.4%231 cards · fell 53% on their own | 83.3%230 cards · fell 52.9% on their own |
Published Sep 16, 2026
Calls on vintage cards priced from a source that doesn't distinguish condition
For some versions of cards from sets released between 1999 and 2002, mostly reverse holos and 1st Edition holos, the only price we had came from a market feed that doesn't say what condition a card is in. We treated it as a Near Mint price, and it rarely changed from one day to the next.
So we made calls on a price that didn't move, then scored those calls against the same unmoving price. They measured nothing. They were not right or wrong about the market.
How to check it
A call scored against a price that never moves returns exactly 0%. Among these calls, 83% of 14-day results and 78% of 30-day results were exactly 0%. For every other call counted on this page, it was 13% and 6%. 1 of the 95 14-day results and 3 of the 89 30-day results counted as a hit.
What changed
We marked 508 calls on 31 card versions as corrected. They no longer count toward the live figures above, and results still to come on the same cards are marked the same way as they arrive. This correction raises the figures above. We would have made it the same way if it moved them the other way.
| Live figure | Counting them | Not counting them |
|---|---|---|
| Value entries @14d | 16.9%69 cards · rose 38.5% on their own | 17%68 cards · rose 38.8% on their own |
| Trend rides @14d | 57%472 cards · rose 38.5% on their own | 57.9%464 cards · rose 38.8% on their own |
| Sleep @14d | 79.1%276 cards · fell 61.5% on their own | 79.2%275 cards · fell 61.2% on their own |
| Value entries @30d | 19.2%55 cards · rose 46.6% on their own | 19.2%55 cards · rose 47.1% on their own |
| Trend rides @30d | 66.9%414 cards · rose 46.6% on their own | 68.2%408 cards · rose 47.1% on their own |
| Sleep @30d | 83.2%231 cards · fell 53.4% on their own | 83.3%230 cards · fell 52.9% on their own |
Backtest in-sample · for reference
Headline figures are the high-conviction calls we actually promote: momentum and breakout entries and confirmed downtrends, above our confidence floor. The full per-signal breakdown, including the weaker reads we don't lead with, is in the table below.
High-conviction Snipe @ 30d
71%
2,941 of 4,164 · avg +10.3%
High-conviction Sleep @ 30d
59%
478 of 810 · avg +1.8%
Signals evaluated
5,344
as of 2026-10-10
By lens
By signal
Snipe signals are tagged by strategy · Value = into weakness · ↗ Trend = ride the winner
| Signal | Call | Hit rate | Avg return | Sample |
|---|---|---|---|---|
| Momentum BuildingTrend rides | Snipe | 72% | +11.0% | 2,843 |
| Healthy BreakoutTrend rides | Snipe | 67% | +8.7% | 1,347 |
| Downtrend · Sub-threshold | Sleep | 59% | +1.0% | 717 |
| BreakoutTrend rides | Snipe | 55% | +7.5% | 138 |
| Price Below AverageValue entries | Snipe | 55% | +26.7% | 125 |
| Sleep | Sleep | 63% | +5.1% | 88 |
| PullbackValue entries | Snipe | 49% | +18.1% | 81 |
How this is measured
For every card, we replay our model across its full price history: at each past date it sees only the prices up to that day (no look-ahead), exactly as it runs live. Each time it flagged a Snipe or a Sleep, we then look 30 days forward and check what price did.
A Snipe counts as a hit if the price was higher 30 days later; a Sleep counts as a hit if it was lower. “Watch” calls aren't directional, so they're excluded. Buckets with fewer than 20 samples aren't shown.
This is an in-sample backtest (our current model measured against past prices), not a log of live, real-time calls. It shows how the model behaves on history; live forward results will differ.
Not financial advice. Snipe or Sleep is an informational tool. Signals are generated by an automated model from historical prices, may be inaccurate or out of date, and are not a recommendation to buy or sell anything. Collectibles are volatile and can be illiquid; you may not be able to sell near a quoted price. Do your own research; you alone are responsible for your decisions. Not affiliated with, endorsed by, or sourced in partnership with TCGPlayer, eBay, or The Pokémon Company.