gates / gates (push) Successful in 2m9s
Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS
Releases 2b and 3 — evidence (2026-11-15)
Report: REPORT-release-2b-3-2026-11-15.md (repo root). Golden and fresh install:
documentation/tests/golden-0.353.0-2026-11-15/. The ids of the deleted boxes whose rows the sweep removed are NOT
here: /mnt/5_hdd/felhom.eu/evidence/release-2b-3-2026-11-15/sweep-preview-ids.txt (off git; an id can be a name).
| File | What it shows |
|---|---|
part0/redproof-agent-cutwrite.txt |
the cut-write fix of the agent's config: 9 mutations, each seen red; 3 redone and one hollow test removed, said in the file |
part0/sweep-preview-counts.txt, sweep-preview.sh |
on a copy of the hub's own 08:00 snapshot: the first sweep deletes 26 of 504 os_reports rows, 13 owners, 0 in the five other tables; the control (a row to delete whose owner exists) is 0; the copy is deleted |
part1/hub-0182-start-log.txt |
hub 0.182.0's start: „Purged 26 row(s) … map[os_reports:26]”, the daily ledger's first lines, the lost-mail scan |
part1/hub-system-page-0950.txt |
the System page on 0.182.0 with the two new cells |
part2/t1-before.txt, t1-after-agent.txt, t1-moved.txt |
Tester 1: pinned → agent 0.175.0 (the wrapper refuses the unknown mode; the false trust problem) → root files → „MOVED … cleared AFTER a working call with the copy” |
part2/hp-moved.txt, n100-moved.txt |
demo-hp and the N100 with 0.175.1: no trust problem in the window, then MOVED |
part2/hp-cert-round1.txt, hp-cert-round2.txt |
two forced new Proxmox certificates on demo-hp; after each: the same agent process, 0 trust lines, 0 ERROR lines, 14 and 40 calls by the agent's token answered 200 in Proxmox's own access log |
part2/hp-ca-wrong.txt |
the CA copy replaced by another CA and the agent restarted: the right root is written first; after the next start the file is identical to the saved one |
part2/redproof-agent-root-step-too-old.txt |
the fix of the false trust alarm and its narrowing after the security review: 7 mutations seen red, 3 redone |
part2/t1-clock-A-*.txt |
the clock 150 s back with the OLD controller: one app recreated at the step back and again at the step forward; the kernel's boot id unchanged |
part2/system-1107.txt |
the „Security profiles” cell filled for the three boxes |
part3/boxwalk-0353.txt, watch-boxes.txt, watch-hub.txt |
the three boxes on controller 0.353.0 (image 61cf2a8b…), 0 ERROR lines; 25 minutes after the global floor |
part3/https-only.txt, hsts-headers.txt |
the https-only check green; 26 of 26 public names answer exactly max-age=31536000 |
part3/hub-webcerts.txt |
from a copy of the live hub database (shredded after): the three boxes' reports carry web_certs; the ledger's and the end-date baseline's marks |
part3/t1-clock-B-*.txt |
the same clock step with the NEW controller and agent: the same container ids before and after, in both directions |
part3/t1-after-reboot.txt |
one real restart of Tester 1: a new boot id, the boot pass recreates the drive-backed app |
part3/per-box-floors-cleared.txt, hub-system-page-after.txt |
the per-box floors cleared; „No per-customer floors”; global floor 0.353.0, vouched agent 0.175.1 |
The hub's two mails of the day (both to the operator's address only, read in the test mailbox): proxmox_trust_problem
for tester-1 at 10:04 CET (false, fixed), box_clock_wrong for tester-1 at 11:05 CET („150 seconds behind”). The
second clock alarm at 11:20 was suppressed by the one-hour rule (the hub's log).