Page Comparison

Truth Score depends on three metrics:
1) % Percentage of Audit Log Messages (40% of the score).

This is (Audit Log Messages for a Dataset / Total Audit Log Messages). Programs reads are omitted to avoid redundancy.

 
2) % Percentage of Unique Programs Read (40% of the score).

This is (Total Unique Programs reading from dataset / Total programs present)

 
3) Time Since Last Read (20% of the score, by rank of the most recent read among most recent reads for all datasets)

   Sample Output:   Example: if there are 10 datasets, they are sorted based on time since the last time the dataset was read. The dataset that was read the most recently gets 10/10 * 20 = 20 (rank / total datasets * 20) points, as it's ranked first. The second most recently read dataset receives a score of 9/10 * 20, the third most gets 8/10 * 20, and so on. As time since the last read can vary from never to 0 to a very large number, a relative score seems necessary.

Sample Output 1

Calculation Example:

DS1: 70% of the Audit Log Messages are for DS1. 70*40/100 = 28 (40% of the score)

         60% of the programs access DS1: 60 * 40/100 = 24 (40% of the score)

         Among the two datasets, DS1 has been accessed the most recently, so 2/2 * 20 = 20 (20% of the score)

         Total: 72

DS2: 30 * .4 + 50 * .4 + 1/2 * 20 = 42

Sample Output 2

Calculation Example:

DS1:

25 * 0.4 + 30 * 0.4 + 1/4 * 20 = 27

Sample Output 3

Problems with the design

• Scores go down (on average) as number of datasets that are tracked increases (example: sample output 2)
• Most scores are on the lower end. Even dataset that look popular on paper have a score of around 65-80.