Over the past week, I've been working with @xeroc's Piston tool to track Steemit post interaction in the hopes of digging into some of the numbers. I feel like i've built up enough data now to share some of the trends I'm finding. In this post, I will concentrate on the first 30 minutes a post is active, as this is the pivotal decision making time for maximizing your curation reward. This is due to the reverse auction-style penalty in the first 30 minutes, which has been covered extensively by authors greater than me. From the official @steemitblog release:

all curation rewards earned by voting in the first 30 minutes after a post is made will be shared with the author. If you vote immediately after a post is made, then 99.94% of the curation reward will go to the author. If you vote after 15 minutes, then 50% will go to the author. Any votes made 30 minutes or later 100% will go to the curator. This only applies to the curation rewards. The author still gets at least 50% of the total rewards.
...
Under the new system bots must balance speed vs payout without waiting too long. Which ever bot is willing to share the most with the author for the privilege of going first wins. To maximize their reward bots want to wait the longest time possible without letting other bots go first.

It is important to note that this function is linear and continuous, meaning there is no sudden jump at 15 minutes, and no sudden jump at 30 minutes. Before digging into the data, I'd like to outline my methodology.

Methodology

I picked a set of popular authors, who consistently receive fairly high payouts on their posts. Although this data may be reflective of a wider range of authors, I can only speak to the activity on popular authors such as these. Here are the authors included in this dataset: @ausbitbank, @donkeypong, @furion, @fyrstikken, @gavvet, @good-karma, @hilarski, @jerrybanfield, @jesta, @joseph, @kevinwong, @kingscrown, @kyriacos, @ned, @officialfuzzy, @papa-pepper, @stellabelle, @thecryptofiend, @timcliff

From those authors, I tracked a total of 54 posts, collecting data continuously for the first 30 minutes the post was active. Grouping by minute, i looked at the total votes and payout accumulated within that minute alone, to try to identify trends in when post interactions are taking place.

Ok, enough build up. Let's look at the data ...

Results

Seems like there's not a lot here. However, there is one important thing I notice. There is a large drop-off between minutes 1 and 2. This is not a surprise to me, as there are probably still some bots which are coded to jump on posts from popular authors immediately. Not smart in terms of curator rewards, but oh well. Let's move on to payout ...

Wow! I just love significant results. There is a huge spike after minute 15, which is maintained the rest of the way. It's important to note here that these are not minutes elapsed, meaning that minute 1 goes from 0:00-0:59 and minute 15 really goes from 14:00-14:59. Therefore, minute 16 on these graphs is when the curation reward penalty drops below 50%. Remember, though, that the curation reward penalty is a continuous linear function going from 100% to 0% over the first 30 minutes. It is a smooth, constant decline in penalty. It seems to me like there is some human psychological bias going on here. The way the curation reward calculation was announced, people seemed to consistently make specific mention of the 15 minute mark, even though that point in the function is completely arbitrary. It's just a pleasant way of breaking it down mentally. This spike could simply be due to a natural human tendency to break things up into halves. It seems people are weighing the tradeoff between voting too early and incurring a high penalty, and waiting too long and missing out, and they've generally decided that they should wait until the penalty drops to at least 50%. There is a lot of activity from bots, but it's important to remember humans make bots and build in their own human biases when they do so.

My first thought was that this seems to be the type of interaction you'd expect from less experienced users. But then It occurred to me that, if the number of votes per minute didn't shoot up dramatically, and the payout per minute did, then the votes being made after the 15 minute mark are worth more. There are higher reputation users who suddenly start voting after the 15 minute mark. We can prove this out by combining the first two graphs to look at payout per vote on a minute basis.

BUSTED. There seems to be a spike in high reputation users voting after 15 minutes. Not that this is a terrible strategy, but it is somewhat arbitrary and clearly leaves room for improvement. There are smart, dynamic ways to optimize for this problem, but it also seems that a bot which voted just before 15 minutes is up would perform pretty damn well right now. It would get in right before the spike.

Conclusion

This data is averaged over 54 different popular posts, so it will obviously not apply the same to every post. However, on average, there seems to be an arbitrary pattern being followed, in which the number of high-value votes dramatically increases after 15 minutes. This pattern can be exploited by those who know about it.

The brilliant thing about this curation penalty system, however, is that it's fluid. Bots are no longer trying to exploit the rules, but instead trying to exploit the patterns of other users and bots. As bots change their patters, other bots will need to adapt. There will never be a perfect answer, meaning that bots can never fully dominate this system. There will always be a premium paid to those who can actually find good content that others have not found, and help bring it to the forefront.

I've included a full summary of this data below for those who would like to look in more detail. Feel free to copy it and utilize it however you see fit.

minute	votes	payout	payout_per_vote
1	7.7358	3.64918867924528	0.47172439024390
2	3.6226	0.54069811320755	0.14925520833333
3	6.0189	1.55686792452830	0.25866457680251
4	7.9623	2.54709433962264	0.31989573459716
5	10.0000	1.30950943396226	0.13095094339623
6	9.5849	1.65207547169811	0.17236220472441
7	8.1132	2.41967924528302	0.29823953488372
8	6.4717	2.35328301886792	0.36362682215743
9	7.0566	2.20415094339623	0.31235294117647
10	5.0755	1.56543396226415	0.30843122676580
11	5.0377	1.83067924528302	0.36339325842697
12	5.7736	1.95928301886792	0.33935294117647
13	4.4717	2.25788679245283	0.50492827004219
14	5.4717	2.54196226415094	0.46456551724138
15	3.7925	1.17828301886792	0.31069154228856
16	5.6981	6.35503773584906	1.11528807947020
17	5.3585	4.35354716981132	0.81245774647887
18	6.6981	7.19352830188679	1.07396338028169
19	8.9245	8.68620754716981	0.97329598308668
20	8.4906	5.45086792452830	0.64199111111111
21	6.9434	4.97794339622642	0.71693206521739
22	6.7547	5.02092452830189	0.74332122905028
23	7.1509	7.01739622641509	0.98132453825858
24	8.0189	5.73875471698113	0.71565647058824
25	9.6415	8.82286792452830	0.91509197651663
26	8.5849	7.11818867924528	0.82915164835165
27	9.2264	7.44003773584906	0.80638445807771
28	8.4151	8.26913207547170	0.98265470852018
29	7.2075	7.11028301886792	0.98650523560209
30	7.6226	5.13794339622642	0.67403712871287

---

Thanks For Reading! Follow, Upvote, and Resteem @steem.engine