A1200 green screen - seemingly quite random chip mem issues with diagrom, any

[pat_ctrl]

I've got an A1200 with green screen, and I'm struggling to diagnose the issue.

Sadly I fear it's not as simple as a broken trace etc, or even bad RAM chip(s), but I'm still hopeful!

I inherited this machine and the first time I turned it on, it got to workbench disk screen, but then started to alternate with green screen, and now always green screen (with KS3.0 and 3.2.2).

Likely leaky cap damage you say? Well... the board was in immaculate condition with no obvious signs of electrolyte damage. I have de-capped (not yet recapped) and there was some leakage immediately below a couple of the caps which is cleaned up but nothing that seemed too bad. I'm not ruling it out, but given the diagrom output it's hard to understand what might be going on.

Diagrom:

I get random results from the chipram testing. Some addresses fail, Bit 10 is always "1" on read - this could indicate a problem with a data line or RAM chip (I think this is D10, but correct me if I'm wrong?), and it detects random amounts of "working" chip memory. Which then always fails later in working memory test, anyway. To me this looks like some kind of instability/inconsistency with the ram access/bus, but not sure where to start?

It's a REV 2B, so I understand it uses each chip to store 8 bits per 32 bit address.

I've checked VCC to the RAM chips
I've checked continuity on a bunch of the data lines and also CAS/RES etc, but all seem to be fine.
Alice is getting warm but not mental
I've checked the PSU voltages and they're rock solid, and the same PSU runs an A1200/pistorm with no issues.

Next step is to scope out some signals I guess (which requires me to get my scope back from a friend!), but I wondered if anyone had experience with similar issues to point me in the right direction?

I have some replacement ram chips but I'm loath to do SMD replacement until I'm a bit clearer on whether they're the issue or this indicates a custom issue (maybe alice or budgie?) or something completely separate.

Thanks in advance!

Example diagrom output:

Testing if OVL is working: OK
- Parallel Code $fe - Test UDS/LDS line
  - Test of writing word $AAAA to $400 FAILED
  - Test of writing word $00AA to $400 FAILED
  - Test of writing word $AA00 to $400 FAILED
  - Test of writing word $0000 to $400 OK
  - Test of writing byte (even) $AA to $400 OK
  - Test of writing byte (odd) $AA to $401 OK
- Parallel Code $fd - Start of chipmemdetection

Addr $00000400
       Write: $FFFFFFFF  11111111111111111111111111111111
        Read: $AAAA0400  10101010101010100000010000000000  FAILED

       Write: $AAAAAAAA  10101010101010101010101010101010
        Read: $AAAA0400  10101010101010100000010000000000  FAILED

       Write: $55555555  01010101010101010101010101010101
        Read: $AAAA0400  10101010101010100000010000000000  FAILED

       Write: $F0F0F0F0  11110000111100001111000011110000
        Read: $AAAA0400  10101010101010100000010000000000  FAILED

       Write: $0F0F0F0F  00001111000011110000111100001111
        Read: $AAAA0400  10101010101010100000010000000000  FAILED

       Write: $0F0FF0F0  00001111000011111111000011110000
        Read: $AAAA0400  10101010101010100000010000000000  FAILED

       Write: $00000000  00000000000000000000000000000000
        Read: $AAAA0400  10101010101010100000010000000000  FAILED
Addr $00010400
       Write: $FFFFFFFF  11111111111111111111111111111111
        Read: $14617741  00010100011000010111011101000001  FAILED

       Write: $0F0F0F0F  00001111000011110000111100001111
        Read: $FFFFFFFF  11111111111111111111111111111111  FAILED
Addr $00020400
       Write: $FFFFFFFF  11111111111111111111111111111111
        Read: $14617745  00010100011000010111011101000101  FAILED

...SNIP - SOME TIME LATER (sometimes it doesn't get this far, and the number of blocks found varies):

Addr $00160400   OK  Number of 64K blocks found: $01
Startaddr: $00150400  Endaddr: $0015FFFF
- Testing detected Chipmem for addresserrors
   - Filling memoryarea with addressdata

   - Checking block of ram that it contains the correct addressdata

   - Addresserror at: 00150404 00000000000101010000010000000100  A2EAA2AA
   - Addresserror at: 0015040C 00000000000101010000010000001100  2B15000C
   - Addresserror at: 00150424 00000000000101010000010000100100  00150424
   - Addresserror at: 0015042C 00000000000101010000010000101100  3A15002C
   - Addresserror at: 00150444 00000000000101010000010001000100  00150444

[pat_ctrl]

Anyone able to help intepret/debug possible causes of this type of diagram chipram test output?

Specifically - a mix of seemingly random wrong bits across whole address, plus a bit permanently returning “1” (in an rev 2b, which I understand uses all 4 chips for each address)

I’m thinking of replacing the RAM chips (perhaps starting with 2nd most LSB chip which I think is second from bottom). But any other pointers gratefully received.

Thanks

[pat_ctrl]

Also note the errors seem to be occurring randomly - each boot gives different results.