How PriceSquirrel Works

This page documents the exact methodology behind the Buy/Wait/Neutral verdict engine β€” real thresholds, real data sources, real limitations. No marketing language.

1. How the verdict is calculated

A tiered engine that adapts to how much data we have

The verdict engine has four stages. Which stage a card is in depends entirely on how many days of price data we have for it. Each stage uses a different baseline to avoid generating false signals from thin data.

πŸ”„CALIBRATING< 4 days

No verdict. We are collecting baseline readings and have no meaningful history to compare against yet.

πŸ”„EARLY PULSE4–13 days

Trend direction only (↑ rising / ↓ falling / β†’ stable). Not enough history for a statistically meaningful BUY or WAIT signal β€” we show the trend without a verdict.

πŸ”„SHORTAGE PULSE13–30 days

Uses the 13-day rolling mean β€” calibrated for the short, volatile cycles typical of shortage markets. BUY triggers when price is 5%+ below the 13-day mean, or drops 3%+ at a single store compared to its price 18–48 hours ago. WAIT triggers when price is above the mean and the trend is up.

BUY can also trigger from a velocity dip β€” a 3%+ price drop at a single store compared to its price 18–54 hours earlier, even if the absolute price hasn't crossed the 5% threshold yet.

πŸ”„STABLE ENGINE30+ days

Uses the 30-day moving average. More data means a more stable baseline and fewer false signals. Same 5% threshold applies.

Why not MSRP? In the 2026 EU GPU market, MSRP is functionally irrelevant β€” cards routinely trade 20–60% above it with no sign of converging. Using MSRP as a baseline would produce a permanent WAIT verdict for nearly every card. We benchmark against observed real-market prices instead.

2. Safety mechanisms

Three filters that protect verdict quality

Outlier filter β€” 50% spike hold

If a price changes by more than 50% in a single scrape cycle (6 hours), that data point is held for confirmation in the next cycle before it affects the verdict or rolling average. This prevents misread prices, temporary errors, and flash sales from poisoning the baseline.

Shortage Tax Lock β€” 48-hour WAIT

If any single tracked store spikes 15%+ in price within a 24-hour window, the verdict locks to WAIT for 48 hours across all stores for that SKU. This protects against sudden manufacturer MSRP hikes or shortage-driven price gouging being misread as a normal market fluctuation.

Cross-store aggregation

The verdict is computed using data across all tracked stores, not per-store in isolation. A single elevated reading at one retailer is diluted by the others. We track retailer storefronts only β€” marketplace listings and scalper prices don't appear here.

3. Data sources and update frequency

Where the numbers come from

πŸ‡©πŸ‡ͺ

Alternate.de

Germany

πŸ‡«πŸ‡·

Alternate.fr

France

πŸ‡©πŸ‡ͺ

Coolblue.de

Germany

πŸ‡«πŸ‡·

LDLC.com

France

πŸ‡³πŸ‡±

Azerty.nl

Netherlands

πŸ‡³πŸ‡±

Megekko.nl

Netherlands

πŸ‡ΈπŸ‡ͺ

Webhallen.se

Sweden

Minimum store requirement: Only products tracked at 2 or more stores are shown in listings. Single-store products are tracked in the background and will appear once a second store carries them.
Note on Webhallen.se: Webhallen prices are tracked in SEK and displayed on product pages for reference. They are excluded from verdict calculations, which use EUR prices only.
Scrape frequencyEvery 6 hours
Data retention90 days per SKU per store
Tracking startedMarch 10, 2026
Per-retailer last updatedShown on each price row
Price stale thresholdAmber warning after 8 hours

4. What the verdicts mean in practice

Plain language, no hedging

🟒

BUY β€” This price is statistically low

The current price is more than 5% below its rolling mean (13-day in SHORTAGE_PULSE mode, 30-day in STABLE_ENGINE). If you need this card, this is a reasonable time to buy. It doesn't mean the price won't drop further β€” it means it's below its recent norm.

πŸ”΄

WAIT β€” This price is elevated

Price is above its rolling mean AND trending upward. A price above the mean but flat or falling does not trigger WAIT. The Shortage Tax Lock (15%+ spike in 24 hours) also triggers WAIT for 48 hours. Unless you need the card urgently, historical data suggests patience is likely to be rewarded. Not a guarantee β€” just a signal.

🟑

NEUTRAL β€” Price is in line with recent history

The price is within Β±5% of its rolling mean. No strong signal either way. The decision should be based on your personal need and budget, not market timing.

πŸ“Š

EARLY DATA β€” Trend direction only

Between 4 and 13 days of data. We can see which direction prices are moving but the baseline isn't stable enough for a BUY or WAIT verdict yet. Use it as directional context, not a signal.

πŸ”„

GATHERING DATA β€” No verdict yet

Fewer than 4 days of data. We don't issue any verdict during this window. Check back in a few days.

5. What we don't do

Honest limitations

βœ—

We don't predict the future. Every verdict is backward-looking β€” we analyse historical price data, not forecasts.

βœ—

We don't account for your personal budget or use case. A WAIT verdict on a 1,200€ card is irrelevant if you need it today.

βœ—

We don't factor in retailer reputation, shipping costs, or warranty differences between stores. Two identical prices at different stores get identical verdicts.

βœ—

Our history goes back to March 10, 2026. Verdicts will become more accurate as the 30-day and 90-day windows fill with real data.

⏳

Shipping cost comparison is on the roadmap but not yet implemented.

⏳

Retailer trust scoring (reviews, return policy) is on the roadmap.

FAQ

How often are prices updated?

Every 6 hours. The scraper runs automatically across all tracked stores and every data point is stored permanently.

Why is MSRP not used as the baseline?

In the 2026 EU GPU market, MSRP has no relationship to actual retail pricing. Cards routinely sell 20–60% above MSRP with no sign of converging. Using MSRP as a reference would make every verdict WAIT permanently. We benchmark against real observed market prices instead.

How accurate are the verdicts with limited history?

The engine is tiered to match the data available. Under 4 days: no verdict (CALIBRATING). 4–13 days: trend direction only (EARLY_PULSE). 13–30 days: 13-day mean with shortage-market thresholds (SHORTAGE_PULSE). 30+ days: 30-day moving average (STABLE_ENGINE). Verdicts issued in SHORTAGE_PULSE mode are valid signals β€” they are intentionally calibrated for volatile, short-window markets. Tracking started March 10 2026.

Which GPUs do you track?

RTX 5050, RTX 5060, RTX 5060 Ti, RTX 5070, RTX 5070 Ti, RTX 5080, RTX 5090, RX 9060 XT, RX 9070, RX 9070 XT, and Intel Arc B570/B580 from major AIB partners across 6 European stores.

Why only Europe?

EU GPU pricing is structurally different from US pricing β€” VAT, import duties, and local supply chains mean US price history is irrelevant to EU buyers. Sites like CamelCamelCamel are US-focused. We fill the EU gap.

See it in action

Every GPU card page shows the live verdict, per-retailer timestamps, and 90-day price history.

Browse GPUs β†’