vpao3

No seed fits this run under any model tested. Unlike the other short pieces, it has enough information (seven values of range 45 or more, about 40 bits) that a matching seed would stand out anywhere in the search space. None exists under the count model that explains every placed run, so something about how these values were drawn must differ.

The run

Seven posts, 27 June 2012 at 20:07 UTC to 28 June at 01:11 UTC. The first is 20 minutes after the last post of run 08, and then the posts come hourly at about :11. The title clock is on UTC−7 for six of them and on UTC−4 for the last one. Run 09, the first run with Next(8, 64), follows about 30 hours later.

#	Post	Posted (UTC)	N
1	vpao3	27 June, 20:07:40	14
2	vpawr	27 June, 20:11:21	32
3	vpeod	27 June, 21:11:30	42
4	vpi8j	27 June, 22:11:33	24
5	vplrd	27 June, 23:11:37	52
6	vpp7k	28 June, 00:11:40	21
7	vpsnt	28 June, 01:11:45	17

What has been tested

Test	Model	Space	Result
Full search	Next(8, 64), one draw per post, up to one skipped draw, up to eight extra draws, offsets up to 72	every seed	0 matches
Continuation	Every placed count stream, each with its own range	the first 20,000 draws, up to three skips	0
Continuation of runs 1 to 8	Any Next(lo, hi), lo 0 to 14, hi 53 to 128, up to nine draws between posts	the first 3,000 draws	Run 08: 0. Runs 02 and 06: six chance paths, none consecutive
Fresh start	1,140 ranges (lo 0 to 14 by hi 53 to 128), up to two extra draws, offsets up to 72, launch between 12:00 and 20:07 on 27 June	the two June boots	5 hits against 5.6 expected. Controls found 3 of 3 in 20 batches
Own GUID seed	One draw of each post's own GUID-list seed, the same draw and range for all seven	seven known seeds	0

An empty result means something here. If the model were right, the true seed would appear in these searches, and at about 40 bits chance matches are few: about 0.09 for the narrowest test and about 1.6 for the broadest. So an empty result rules the model out. The five hits in the fresh-start test all sit at implausible starts (offsets of 25 to 62, or two samples per post). None is at draw 0 with one draw per post, as every 2012 run is.

Open readings

• A different draw rule in a test build. 27 June falls between the Next(8, 24) runs of 26 and 27 June and the first Next(8, 64) run on 29 June. A build in transition could draw counts differently, from a range not tested here, with more draws per post, or as a sum of draws.
• The last post is not part of it. vpsnt has a different title zone. Without it, the six-post version carries 35 bits, and it was not part of every test above.
• Not from the generator at all. The values may have been entered or derived by hand, for a test.

What could decide it

A model change that also explains another unplaced piece from June or July 2012, such as vvjmg or the burst in vsb-burst. On its own, this run cannot select among the readings above.