igneum-network/igneum
1
0
Fork 0
Code Issues Pull requests Releases Activity

adv-mixer-3 report: box-hour totals and the partial rows named

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
igneum-labs 2026-10-08 00:03:43 +00:00
parent 9a14f0c88e
commit 1017b2ed66
1 changed files with 9 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
docs/analysis/cryptanalysis/report-mixer-3.md
View file

@ -308,6 +308,15 @@ CNF out of the worktree mirror into /srv/builds/_adv-mixer-3/. Both applied befo
| Pre-emption at any size (lease sha ce30e357) | both | 21:21, applied 22:2x | | a release or v5 waiter that has waited 120 s pre-empts adv holders of any size with SIGTERM; both drivers now wrap every shard in run_shard (retry the same lease line until the log carries a result, log each PREEMPTED line with its time, skip rows already done). Queue 17 restarted while waiting on its sac 20729 k = 8 lease (nothing lost); queue 18 restarted on CaDiCaL k = 3 about 15 min into its hour (that partial solve lost, restarted from zero) |
| Queue 07 (8 random days) | build-1 | 18:5x | running | claimed and run by the sibling lane adv-mixer from the shared queue; its logs are index-<day>-k<k>.log and sac-<day>-k<k>.log under /srv/builds/_adv-mixer-3/logs/ on build-1, read by this lane |
Totals at 00:05 UTC on 8 October: about 3.0 wall-hours of sweep time on the two boxes (build-1 about 1.9, build-2
about 1.1; every sweep ran beside other lanes on boxes at load 100 to 600 on 96 cores, so these are loaded-box
hours, not core-hours) plus 4 single-core CaDiCaL hours; pod-hours 0 (no GPU on either box). Under the 8-hour
reading line.
Partial rows, named: Q3 k = 8 was killed at 19:20 UTC and not re-queued (k = 2..7 are clean on both days); Q2 at
2^27 (k = 2, 3) and Q2b day 20733 at 2^28 (k = 2, 3) are in queue 17; Q6 day 20733 k = 2..4 in queue 18; Q7 has
day 20730 only (queue 07 is run by adv-mixer); multi-bit linear masks and a MILP trail bound were not attempted.
## The round margin, stated
Internal adversarial pass, not an independent review. Against the 8 applications between dependent cache reads