Community coordinated feedback on Tenacity

[hqdm]

The tests I had seen shows almost 1800 points making less than 5% difference. How are the new test results different?

The initial "tests" were done by some guy who had a single-digit sample size which effectively made them worthless despite good intentions.
There was a second one I saw after that, but only had a change of ~2xx TEN over a 150-sample test.

Gonna start off by pointing out that not one of the people who have tested the stat have done a proper test. They have either been 2 data points, or tested over a small range. Nobodies going to know the real value until someone does an incremental test from near base to near cap over large sample sizes. That being said, the fact that tenacity affects outgoing healing instead of incoming is ludicrous, and offers virtually no value as a bonus.

Incremental tests don't add any value for non-tiered stats.
Instead, you're making the dataset murkier because the main problem with testing reduction %s atm is having small damage ranges.
A tank taking ~1k dmg hits with 5% effective change in hp from TEN means ~50 dmg difference.
Splitting the tenacity delta to do incremental tests means a higher fluctuation / error margin because now you're measuring 200-300 TEN gaps where the difference will only be 15-20 damage and any 1 pt of dmg will have a greater %value.

To get a more accurate result than the recent test you'd need to find a mob that did more damage per swing, or wait until 4.x when everyone's swimming in higher ilvl raid gear.
The problem with that is that it's annoying af to test vs high damage mobs because you can't afk or you need an actual player healing you (outdmgs the bird) and healing aggro issues.
Desummoning the bird every 5min drops the aggro, but getting a player to rezone every 5min for a 7-8hour test is just not happening.

What constitutes a large sample size for you?
The last one was done with ~5k samples per point.
There was a fluctuation of about ~0.1% between 3k and 5k samples for some of the datapoints, but most reported back at the same rate.
The point of a larger sample size is to make sure any outliers don't affect the overall data - once it's settled, which it did by 5k, there's not much point in going higher.