…res + warnings scroll past too fast; empirical from physical hardware-support test; 3rd empirical anchor in same test session as B-0832 + B-0833)

Per operator 2026-05-26: "i got some failures and warings on install
of nixos not sure if it matters it scrolled by to faster have gh
login this is exactly what i'm hoping you can log and test in ci"

Two observations:
1. Install failures + warnings scroll past faster than human read speed
2. gh login not reached; scroll-past blocks diagnosis

Files B-0834 as P2 substrate-engineering target. 2-approach scoping:

- Approach A (preferred): tee install output to /tmp/zeta-install-*.log
  + copy to /mnt/var/log/zeta-install.log on completion
- Approach B (upgrade path): script(1) wrapper records full session
  including ANSI + timing (replayable via scriptreplay)

P2 priority — diagnostic enabler, not hard install blocker.

This row is the OPERATOR-SIDE analog to B-0831's CI-SIDE cascade #6
serial-console-as-workflow-artifact. Both solve the same fundamental
problem (preserve install output for later review) at different
surfaces (real-hardware operator vs QEMU CI). Both compose; both land
separately.

THREE empirical anchors in ONE physical hardware-support test session
(B-0832 nmtui WiFi rescan + B-0833 interactive-login vs baked-keys
tension + this row B-0834 log preservation) is strong validation of
B-0831's reframing: physical-test-as-first-class-hardware-
compatibility-matrix-substrate produces real-world substrate-
engineering targets that CI emulation cannot reproduce within
minutes of B-0831's own landing.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>