jan2013-A1

Two seeds reproduce all nine posts. Against the two hits expected by chance in the window, the pair is not evidence, so the run is not counted as reproduced. Both boots that would be expected a priori are excluded.

The run

Nine posts, 3 to 8 January 2013, every 16 hours, with the title clock on UTC−2. The 2013 build draws 4 × Next(8, 16) per post. Draw 0 is the startup draw, and the posts run from draw 1.

Draw	Post	Posted (UTC)	N
1	15vhjm	3 Jan, 06:00:00	44
2	15wtz8	3 Jan, 22:00:00	48
3	15y7po	4 Jan, 14:00:00	44
4	15zuqj	5 Jan, 06:00:00	44
5	160zew	5 Jan, 21:59:59	60
6	1626ov	6 Jan, 13:59:59	32
7	163u1f	7 Jan, 05:59:59	36
8	1656hd	7 Jan, 21:59:58	32
9	166lht	8 Jan, 13:59:58	48

Why it is hard

Each post has eight possible values, so nine posts carry 27 bits. That is enough to make any window of up to about 100 million seeds meaningful, but the boot is unknown, and the launch is only bounded to 1.1 posting intervals, about 18 hours.

How it was narrowed

The neighbours give the first constraints. Run 39 last posted on 12 December 2012, the morning of the December reboot after Patch Tuesday (11 December), and the unseeded fixed-N posts that followed had several breaks. The next piece, jan2013-A2, starts on 9 January, and run 40 is on a boot between 9 January, 03:28 and 10 January, 05:52.

Step	Constraint	Candidates	Expected by chance	Seeds found
Every seed on the tick grid	First post at draw 1	about 137 million	1.0	2
Run 39's boot	Run 39's boot continued (20 November)	about 4.5 million	0.03	0, so excluded
December reboot	A reboot on 12 or 13 December, 09:00 to 12:30 UTC	about 9.7 million	0.07	0, so excluded
Any boot	A boot after run 39's last post (12 December, 06:00:08), launch within 1,056 minutes	about 122 million	0.91	2

The candidates

Seed	Startup draw (never posted)	Draws 1 to 9	Boot (UTC)	Fits a break in the fixed-N run
162,853,390	36	44, 48, 44, 44, 60, 32, 36, 32, 48	31 December, 15:09 to 1 January, 08:45	Yes: the fixed-N gap from 31 December, 07:00 to 1 January, 07:56
1,000,996,750	40	44, 48, 44, 44, 60, 32, 36, 32, 48	21 December, 22:20 to 22 December, 15:56	Yes: a fixed-N restart before 23 December, 05:00 (title clock and seconds change)

Both reproduce draws 1 to 9 against the stream of the .NET generator.

Reading. Each candidate's boot lines up with a break in the fixed-N stretch. A break can also be an ordinary reconfiguration, so that does not choose between them. Two hits against 0.91 expected is chance-level: the window covers most of the seed space, which holds only two hits on the tick grid at draw 1, and the empirical probability of a pair this good is about 0.66.

What could decide it

Showing that one of the fixed-N breaks was a reboot rather than a change within the same process, or a second piece on the same boot.