I'm confused:

  				# Builds completed	Total battery life	Avg. Build time
  2016 Touch Bar - 3.1 GHz i5	49			4:54:00			0:08:45
  2016 Fn key - 2.4 GHz i7	30			3:54:00			0:07:52
  2015 Retina - 3.1 GHz i7	24			3:30:00			0:06:02

suggests that the 2016 model did 19 more builds, and had a battery life of 1 hour longer, suggesting a _better_ battery life...

Looking at the data in his spreadsheet, I think he has the models and build times switched.

You're correct. Sorry about that, I fixed the table!

Ah, that makes so much more sense now!

2016 Touch Bar - 3.1 GHz i5 24 3:30:00 0:08:45 2016 Fn key - 2.4 GHz i7 30 3:54:00 0:07:52 2015 Retina - 3.1 GHz i7 49 4:54:00 0:06:02

Above is straight off of his website. Looks like the OLDER MBP has more builds/more battery, which is what he was saying. Your table looks completely wrong, but I understand that he had some errors on the original site. The comments in the article claim that the above is correct.

My comment was from prior the up date to the site, or the comments, hence the confusion

even consumer reports had enough issues with the battery to make it a "not recommended" machine.

"The MacBook Pro battery life results were highly inconsistent from one trial to the next."[1]

did you see the same?

[1]http://www.consumerreports.org/laptops/macbook-pros-fail-to-...

The results were highly consistent for me (I tested the Touch Bar model 5 times just to be safe). I presume that a lighter test (e.g. only browsing, watching videos, etc.) would probably allow macOS to apply more battery optimizing techniques, which may vary depending on the exact test, WiFi conditions, etc.

For my test, it was full bore, using lots of CPU (all cores), lots of bandwidth, and lots of I/O, with the screen on 100% brightness the entire time.

Using 100% brightness makes the test basically invalid. The backlight is one of the biggest power hogs in the whole machine, and the new Macbook is 65% brighter at maximum. That's helpful for e.g. outdoor use, but for office use the proper setting is 150-200 lux (about 50% brightness on the old machines, 30% brightness on the new ones). Testing the machines at different brightness settings is akin to testing them running different workloads.

I'm less concerned with the regression from 2015 to 2016 models, and more with the fact that the Touch Bar model does so much worse than the Function Key model.

At 100% the test might be invalid but I don't think your percentages are correct. I'm sure it's environment specific but I run my new 2016 Macbook Pro w/ Touchbar at about 60% brightness most of the time. At 30%, in office LED lighting, it's way to dark. At 60% it's still darker than a Monoprice gaming monitor that sits right next to it.

Note that Consumer Reports and Apple apparently had a pow wow and it was discovered that the settings they were using in Safari (to disable all caching) exposed some bug that was causing the fast run-down rates in those tests. (their tests on Chrome were consistent and long lasting)

They are doing a retest this week.

So on the battery front it seems not all is lost.

One important point is that you switched from i5 to i7, which has a fairly significant impact on battery life (in my experience).

Do you have your table labels muddled on the article? It's showing the 2016 with Touch Bar completing the most builds, but your spreadsheet and general disappointment disagrees with that.

Yes, sorry; I fixed the table just now; I had some of the data points switched between the Touch Bar model and the 2015

I switched from 2014 15" to 2016 15" with touch bar. Can't really complain about the battery life. Is it the 13" with battery issues?

This is my last 12 hours. http://c.ekin.io/3o0S3e1j3i36 Have been working on several projects with a VM and PgSQL running at background. This is comparable to my older device.

Yes, it seems to be a much larger problem on the 13" Touch Bar model especially. See the links in the blog post to the iFixit reviews—the battery in the Touch Bar model (which is driving more things and a higher TDP CPU) is actually _smaller_ than the one in the Function Keys model.