I tried my luck at re-writing the program. It was quite fun. My program is 70 lines long. The original is 268 :)
There are places to optimize, but not without benchmarks.

In reaction to message about stop@home I started developing application for search of such symmetric prime tuples.

New application Up! Codename SCPN (will change maybe).
The workunit "scpn_beta1" was submitted to evaluate how it behaves.
Estimated runtime 30 minutes (i5-5200U).
Checkpoints do not work yet. The app checkpoints, but does not resume from checkpoint. That's why the quorum is set to 8 only.
This is non-interactive test (closed alpha stage). I will watch everything and analyse results/failures and I do not expect any feedback from users yet.
Eventually, I will call for interactive test, where feedback will be welcome.

I was so asleep that I forgot I coded the checkpoint resume already. The application should even resume from checkpoints.

First batch of workunits "scpn_xxx_b" in batch 42!
Estimated runtime: 2 hours (i5-5200U). The app checkpoints.
We scan the range 500002700000000000 to 500007900000000000 for symmetric prime tuples.
This range was already scanned and the results are here. We scan the same range to check consistency of this boinc app.
Previous batch 41 was faulty - it attempted to search for 0-tuples :)

I knew i forgot something. The app does not report the checkpoint to boinc, which also messes up other boinc aspect. I think the batch will finish, regardless.

I had the AmAZiNg (irony) idea of only storing half of the tuple coefficients (cause the other half is symmetric). But that is not unique encoding! So to modify the output format again.

For example, what is stored: 500002767573021263: 56, 48, 6, 166, 72, 96, 12, 2
there are 8 offsets. This could be expanded into symmetric dispacements like this:
0, 56, 48, 6, 166, 72, 96, 12, 2, 12, 96, 72, 166, 6, 48, 56 or like this
0, 56, 48, 6, 166, 72, 96, 12, 2, 2, 12, 96, 72, 166, 6, 48, 56.
The second one is wrong. So I fixed it already in code.

I had the AmAZiNg idea of only storing half of the tuple coefficients (cause the other half is symmetric). But that is not unique encoding! So to modify the output format again.

For example, what is stored: 500002767573021263: 56, 48, 6, 166, 72, 96, 12, 2
there are 8 offsets. This could be expanded into symmetric dispacements like this:
0, 56, 48, 6, 166, 72, 96, 12, 2, 12, 96, 72, 166, 6, 48, 56 or like this
0, 56, 48, 6, 166, 72, 96, 12, 2, 2, 12, 96, 72, 166, 6, 48, 56.
The second one is wrong. So I fixed it already in code.

Please keep the generally accepted tuple format!

500002767573021263: 0 56 104 110 276 348 444 456 458 470 566 638 804 810 858 914

or

500155744849852957: 0,24,66,90,120,130,144,154,186,196,210,220,250,274,316,340

Both of these formats are supported by my square builder program.

Please keep the generally accepted tuple format!

Yes.
I already fixed the issue in code and it will be deployed in next test batch.
I want to first process results of the current batch to catch any other problems.

Batch 43 checking in range 500266350516360713..500286766859602800.
Double-checking results from Xaver.

Batch 43 checking in range 500266350516360713..500286766859602800.
Double-checking results from Xaver.

What are the results of the previous batch?

Batch 43 checking in range 500266350516360713..500286766859602800.
Double-checking results from Xaver.

I'm seeing some strange behavior with these tasks.

First thing is they have a short deadline -- about 25 minutes, and they're going to take about 4 or 5 hours to run.

Second problem is even though there's 4 cores on the computer, and 4 tasks were downloaded, only 1 task is running. I don't know why. There's enough memory to run more than one task, so that doesn't appear to be the problem. I'm not sure if this is a problem on your end or my end.

Between the two problems, you can expect lots of automatically aborted tasks because of this. As soon as the (short) deadline passes, BOINC will automatically abort the three unstarted tasks, and immediately download three more. When their deadline passes, they will be aborted and replaced with three more, and so on.

---

Want to find one of the largest known primes? Try PrimeGrid. Or help cure disease at WCG.

I missed one digit while setting the deadline and I corrected it very shortly after. But some tasks are already downloaded and it does not propagate to clients. Also boinc should not download the task if it knows it can't meet the deadline, so not sure how that happened.

Not sure about the only one task running. Maybe that is the consequence of the short deadline? Or memory requirement. There is only placeholder value in the ram requirement field.

<rsc_memory_bound>100000000.000000</rsc_memory_bound>

That's from one of these tasks. That's 100 MB. (I think)

Actual usage is 357.91 MB according to BOINCTasks. Maybe that's got something to do with it?

---

Want to find one of the largest known primes? Try PrimeGrid. Or help cure disease at WCG.

I found out that the feeder respects priority only on startup. If high priority tasks run out, low priority tasks will take their place and wont go away even if low priority tasks appear. It seems the allapps option is needed even thou I dp not like it.

Batch 43 revealed that there is a error in detection of odd-k tuples. There were 37 k=17 reported, but none of them are prime above index 10.
All two k=18 tuples were correctly found, so that's good.
Later I will check whether all k=16 are correct.

k=17 ?
500266632926148299: 0 14 44 50 68 72 134 212 294 330 408 470 474 492 498 528 542
I checked for primality, and all all prime here.
Is this tuple correct? I need help with the odd-k tuples.
There are 40 odd-k tuples in the database from batch 43.

Error with central element 294.
In a symmetric tuple, the central element is d/2, d is the diameter of the tuple.
In your example, d=542.

Tomáš Brada
Pay attention to this definition

k-tuple (p + a1, p + a2, p + a3, ..., p + a [(k-1) / 2], p + a [(k-1) / 2 + 1], p + a [(k-1) / 2 + 2], ..., p + a [k-2], p + a [k-1], p + ak) for k odd called symmetric, if the following condition is satisfied:

a1 + ak = a2 + a[k-1] = a3 +a [k-2] =…= a[(k-1)/2] + a[(k-1)/2+2] = 2 a[(k-1)/2+1]

Example
symmetric 5-tuple

18713: 0, 6, 18, 30, 36

https://www.primepuzzles.net/problems/prob_062.htm

I used to check tuples in Wolfram Alpha,
for example

Select[Range[0,240],PrimeQ[1006882292528806742267+#]&]

result

{0, 6, 24, 36, 66, 84, 90, 114, 120, 126, 150, 156, 174, 204, 216, 234, 240}

Now it doesn’t work for me. I can not understand the reason :(