• Problem:

Minimum Price, especially for Classic WoW, tends to be incredibly useless. This is because often times players or bots will randomly post items for low price, causing 'Minimum Value' to believe the auction is a price that it is not. E.g. [Arcanite Bar] usually costs 100g. One day a bot posts one [Arcanite Bar] for 40g. You scan the auction house and AuctionDB now claims [Arcanite Bar] is 40g Minimum Buyout. Then, you might accidentally sell [Arcanite Bar] for way less than it is actually worth.

• Solution

A 'Market Value' price source can take all of the scans in the last month, remove the outliers (e.g. [Arcanite Bar] for 950g), then return the average using some statistics functions to produce a reliable 'price' of what the item is worth.

To do this, AuctionDB would need to save not just one scan, but many. I noticed that it seems like AuctionDB seems to overwrite the past scan with the current one. This essentially means that AuctionDB has no concept of what items actually cost, only what they cost right now.

To enable saving multiple scans over time, and delete ones that age too long (e.g. remove scans from 5 months ago automatically), how much changing would need to be done to MoLib, AHDB, etc?

it does save all the scans but the tooltip is indeed the latest scan... I think TSM has better averaging but in theory yes AHDB should use all the available information

I discovered this a few days after my post. I was incredibly confused why AHDB never seems to get rid of scan data (this can lead to memory leaks). Unfortunately TSM is not available on all Classic clients, especially on launches of new classic realms. For example Classic-Era did not have TSM available for years. Classic Anniversary did not have TSM available for a week. There is certainly a use case for AHDB to list min / market-value / etc.

If all the scan data is saved, I recommend the following:

• Have an option to remove scans that are X days old (this option may exist? not sure)
• When calculating market value, disregard extremes (e.g. someone posting auctions for 1 copper or 99999999 gold). Some mathematical formulas can be of help here.
• Weight the prices based on how soon they were after the previous scan. This is very important, because if you scan once on Monday, scan 15 times on Wednesday, and scan once on Friday, your market value will be overwhelmingly scaled towards whatever the prices were on Wednesday. This would be a catastrophe. Thus, prices must be scaled based on how frequently the user has been scanning.
• Alternatively, you could reject any scan that's done between X hours of the previously scan. This is not a very ideal route but should be relatively effective.

Theres no magical function that will remove outliers. Upper and Lower are not really useable because people may undercut by 1c meaning that there may be 3 uppers that are 50G over priced. This means we have to rely on a form of standard deviation. 3 Particular Functions would help us figured this out. Mean Median and Mode. Lets break this out now. Some Information we need to understand. For this to work we have to do a fullscan. Meaning we have to look at every every listing and its number of auctions at various prices. With the AH change this should be alot faster then pre-dragonflight.

Lets work with an example of Dream Dust. It has the following auctions organized from lowest to highest.
Dream Dust (Price is per item)

10 - 1g
20 - 1.5g
30 - 1.57g
5 - 30g
1 - 300g

Conversion rates

100c = 1s
100s = 1g
10,000c = 1g

For the sake of making this easier this is how the array would look.
{10000,10000,10000,10000,10000,10000,10000,10000,10000,10000,15000,15000,15000,15000,15000,15000,15000,15000,15000,15000,15000,15000,15000,15000,15000,15000,15000,15000,15000,15000,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,15700,300000,300000,300000,300000,300000,3000000}

I'm not going to explain what Mean Median Mode are in detail but I will give a synopsis

Mean = Average of All the Numbers
Median = When Organized from Least to Highest the Middle Number
MarketValueMode = The Number (Value) that occurs the most often

Below is the value for each.

Mean = 81379
Median = 15700
Mode = 15700

As you can see doing an average out the gate is not beneficial because of the high numbers screwing the dataset.

Lets change how the Mean works then

lets call this ModifiedMean

ModifiedMean = (Median + Mode + MinBuyout)/3

or

ModifiedMean = (15700 + 15700 + 10000)/3
Answer = 13,800

Now lets find a real market value

RealMarketValue = (ModifiedMean + Median + Mode)/3

or

RealMarketValue = (13800 + 15700 + 15700)/3
Answer = 15066

Hope this helps.