Clarendon to London

I've done 7 Clarendon marathons since 2007.  Let's take a look:
2008 - 3:16:41, 3rd place, age-grade 64.7%
2009 - 3:07:05, 4th place, age-grade 68.5%
2010 - 3:21:04, 12th place, age-grade 64.3%
2012 - 3:13:41, 5th place, age-grade 67.8%
2013 - 3:11:05, 3rd place, age-grade 69.3%
2016 - 3:12:39, 2nd place, age-grade 70.5%
2017 - 3:15:05, 6th place, age-grade70.3%

A varying pattern of ups and downs, very much dependent on (a) the ground conditions and (b) who turns up.  Nice to see age-grade scores heading in the right direction.

However, none of this gives me much of a clue about what to expect next April.  I have a good-for-age place in the 2018 London marathon; the first time I'll have done it since 2006.  Here's my history at this event:
 

It's been all downhill so far

Not for a minute do I expect to replicate 2006 result of 2:48:44, when I was a tender 30-something, in 2018 as an M50.  But could I match the result by age-grade?
Age 38, marathon time 2:48:44 = age-grade 74.07%.
To match this age 50, I'd need to achieve 3:05:14. 
Knocking ten minutes off my Clarendon time, with all those hills and all that mud, seems more than achievable.

Let's look at it another way - what would other recent races suggest?
Southampton half marathon, 1:24:41 = 76:93% = marathon in 2:58:20.
Lymington 10k, 0:37:49 = 79.46% = marathon in 2:52:40
Parkrun in 0:18:04 = 81.64% = marathon in 2:48:03.

What a completely mixed bag of data... the algorithms suggest that I will finish anywhere between 2:48 and 3:05.  Que sera sera... but I'd really like to go under 3 hours.