Skip to content

docs(handoff): both ranks shipped — and the fan/AIO tuning question answered (no Linux-side control exists) - #1437

Merged
ZacxDev merged 1 commit into
mainfrom
docs/handoff-cooling-view
Sep 9, 2026
Merged

docs(handoff): both ranks shipped — and the fan/AIO tuning question answered (no Linux-side control exists)#1437
ZacxDev merged 1 commit into
mainfrom
docs/handoff-cooling-view

Conversation

@ZacxDev

@ZacxDev ZacxDev commented Sep 9, 2026

Copy link
Copy Markdown
Member

Updates claudedocs/handoff-nct6683-bar-integration.md in place. Both ranks are merged (#1411, #1430) and both hosts are converged and verified at a2b74e4b, so this closes the effort.

What the doc now carries that it did not:

  • The fan/AIO tuning answer, with the measurements. Under load 96 on 24 cores, Tctl is 81–83 °C against a ~95 °C Tjmax — the CPU is power-limited, not thermally limited, so better cooling buys nothing. And nct6683 exposes PWM read-only (no pwm*_enable, pwm1 not writable, the module's only parameter is force, a vendor-detection override), so a fan-curve daemon is impossible — tuning is BIOS-only. The one real finding is that the case fan sits at a flat pwm3=153 (60%) across every sample, loads 60→96 and temps 71→83 °C.
  • Three gotchas that cost real time this sessionhome.file giving each file its own store path (so .resolve() breaks sibling loading); a .pyc in the config dir outliving every same-size redeploy because store paths have mtime=1; and cbreak leaving ISIG on so Ctrl-C only works in the tty's foreground process group.
  • A near-miss worth keeping: git fetch does not move your local branch head. A worktree still at 9ae4fdb4 merged with origin/main and silently dropped a subagent's 170 lines; only the non-fast-forward rejection stopped it landing.

Honest about gate coverage: #1430's pytest tier timed out twice at 3600s with zero failing tests (box at load 70–96), so its coverage was assembled as 22-of-29 targets in one run plus the remaining 7 run directly. The nix sandbox tier was never run on that tree. Post-merge, origin/main content re-runs green (99 passed; 129 across all three fan suites, verified while writing this).

All four ranked next-steps are forcing: none — deliberately. Nothing external is asking for them, and rank 1 (a case-fan RPM floor) is explicitly conditional on the BIOS-curve decision in rank 2.

🤖 Generated with Claude Code

https://claude.ai/code/session_019LDvGed1jgLsz2UXSZNTdm

…ng view added and audited, and

Claude-Session-Id: fcc01bfd-a65d-4421-b048-8c23f99a003e
@ZacxDev
ZacxDev merged commit 270bc62 into main Sep 9, 2026
0 of 2 checks passed
@ZacxDev
ZacxDev deleted the docs/handoff-cooling-view branch September 9, 2026 06:24
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant