Every app tells you the price.
HoldOrSell tells you what to do, and why.
Price trackers hand you a chart and leave the decision to you. This is a decision engine: seven weighted signals, a regime classifier, and one call per card: Sell Now, Hold Short or Hold Long, with the case for it, the case against it, and how much of it we actually believe.
Both of these charts are falling. Only one of them is a sell.
Scored live, right now, by the same engine that scores your collection, not a screenshot. Reload the page and the numbers move when the archive does.
A falling price is two completely different stories.
Supply went up, or demand went away. Every chart looks the same either way, and the difference is the entire decision. Telling those two apart is what the regime classifier does before any verdict is written.
Hold Long
Reprint Trough
Price is falling because supply rose (a reprint or a restock), not because demand died.
Reprint supply washes through and popular cards recover, so the chart that looks like a sell is usually a hold. It only qualifies if the set clears the popularity floor and has a history of bouncing.
Sell Now
Structural Decline
Price is falling and there is no supply event to explain it: demand itself is fading.
Unlike a reprint trough there is no scheduled end to the pressure. A confirmed reprint on the same dates would reclassify it, which is why reprint evidence is reviewed rather than scraped and trusted.
Not just the two
Four more regimes
Spike / Distribution, Accumulation, Rotation Fade and Stable / Illiquid each carry their own verdict and their own price band.
A call you can argue with.
Nothing here is a black box score. Every verdict ships with the machinery that produced it, including the parts that disagree.
The call and the band
One of three verdicts, plus a regime-dependent price band: the range the model thinks the card trades in from here. Thin markets get a wide band, and you are told to read the band rather than the midpoint.
Both sides, stated
Signals that score high enough form the case for holding, signals that score low enough form the case for selling. When the strongest signal contradicts the verdict, it is printed as a named counterpoint, not quietly dropped.
Confidence, honestly
Confidence is a function of how many real sold comps exist and whether the price series is real rather than simulated. Simulated shapes are never allowed to drive a directional call.
Verdict basis
Every call is labelled trend + signals or trend only. When no signal supports the call, the card says so outright and tells you to treat it as a prompt to look, not a conclusion.
Your position, separately
If you recorded what you paid, your unrealised gain is reported as a fact about your position. It can shorten a hold, but "it went up a lot" is never allowed to force a sale on its own.
Three price layers
Market Ask, Last Sold and Graded, side by side. A layer we cannot source is shown empty with the reason attached, never backfilled with the nearest number to hand.
Export, upload, decide.
Bring your collection
Export a CSV from Collectr, or search the 44,000-card archive and add cards one at a time. Sealed product counts too.
The desk scores it
Every line is matched against a local archive of weekly TCGplayer prices going back to February 2024, enriched with reprint state, rotation and cohort base rates, then classified.
You get a queue, not a table
Sell Now first, sorted by what it is worth to act on. Hold Short next, with the catalyst that should make you look again. Hold Long left alone.
What this is not.
English market only
The price archive is TCGplayer, which means English printings. A Japanese restock is real news our data cannot corroborate either way, and the app says so rather than guessing.
Pokémon is the calibrated game
The reprint, rotation and era logic was fitted to Pokémon. Other games carry catalogue and price data, and the card page tells you plainly that the verdict engine is not applied there.
Sold prices are partly unavailable
A true completed-sales median needs a restricted eBay scope this account has not been granted. That layer stays empty with the reason attached until it is.
Popularity is partly curated
Durable collector demand is hand-curated for a minority of sets; the rest use a neutral default. The methodology page names the number.
Stop reading charts. Start clearing decisions.
Free covers a 50-item portfolio. Pro removes the cap, drops the 24-hour delay, and turns on alerts when a verdict changes.