Exponents to re-release for first-time testing: "harmful" errorcodes

[GP2]

Now that HRF3.TXT contains error codes, we can also use that to automatically generate lists of exponents to release for first-time testing.

We define a "harmful" error code as any nonblank error code that is not zero, AB00AB00 or 00XX0000 (or any of those with the high bit 80000000 set), as described in this thread.

Stats over the range 2M-9M show that results with a "harmful" error code have an error rate of 55 - 60 %.

The overall philosophy, as before, should be that any exponent without at least one "presumed good" result should be released for early re-testing, whereas any exponent with at least one presumed good result will be double-checked in due course (perhaps years from now).

A result is "presumed good" unless:
1) it was returned by an error-prone machine, regardless of error code.
2) it was returned with a harmful error code


George, I guess you've been manually releasing these harmful-error-code exponents all along, but I can now generate a complete list.

[GP2]

We look for all exponents between 10.4M and 19M where all the results returned have been returned with a "harmful" error code.

As usual, we filter out any exponent that currently appears in status.txt or cleared.txt.

We get 582 exponents, of which 74 are triple-checks where both original tests had harmful error codes (for status.txt and cleared.txt as of Oct 9 2003 17:00 UTC).

Slightly more than half (315) of these are exponents that were already released once by George as first-time tests, got converted to double-checks by the server sync, and have since expired.

I would advocate that these could be released right away, without waiting for the first batch to complete, because they are selected by completely different criteria and especially because such exponents have already been being released for quite some time.

If there is no disagreement I can grab the latest versions of status.txt and cleared.txt and filter against them and post the list.

[Prime95]

That's fine. It will save me the time to create such a list.

I'll spot check it once you have it ready.

[GP2]

Originally posted by Prime95
I'll spot check it once you have it ready.

Filtering against Oct 11 2003 00:00 UTC versions of status.txt and cleared.txt gives 586 exponents, of which 74 are triple-checks or higher.

[GP2]

Some of you may be looking at
http://www.mersenne.org/primenet/summary.txt
and wondering what are all those exponents between 13M and 17M in the "available" column for double-checking.

Well, all of them are exponents with "harmful" errorcodes that George re-released once-upon-a-time for "first-time" tests. They got converted to double-checks with the server sync back on Sept 23 and have since expired.

Attached is the full set of 180 exponents as of today at 16:00 UTC. Again, I would advocate that these could be released right away as first-time checks.

There were roughly 2275 exponents that got converted from first-time checks to double-checks with the server sync. These have had a fairly stunning expiry rate. Probably more than half will have expired before the dust settles. More expiries will trickle in over the coming weeks.