7.4 KiB
REPORT — release 1 of four, the „dates” release (2026-11-14, daytime, attended at two STOPs)
1. Claims in the prompt that turned out wrong, named first
- „The hub's box facts hold no list of files” — true for the PARSED facts, not for what the box sends. The box's
host report carries a sha256 per installed root file (
system.facts.host.config_bundle.live); the hub's parser (sysfacts.ConfigBundle) drops it. The fix did not need it. - „The vouch holds only one checksum of the bundle” — true. But the hub can get more at any time: it already has a read-only client for the package registry, and the bundle is a published file. That made a hub-only fix small.
- „Go 1.27.4 and its images are published” — true (go.dev; Docker Hub
golang:1.27.4-trixieand-alpine3.24, both pinned by digest). - The baseline's „golden 0.352.0 baked 2026-11-11 … plus 7 days” is right, and moot: a new golden is vouched.
- CC's own wrong claim, in the 30-minute hub watch: the second read was taken 12 seconds before the 30 minutes were over (10:28:03 against 10:28:15 UTC) and its heading said they had passed. A third read is in the file.
2. Five lines
- Every box that reports runs controller 0.352.1 (Tester 1, demo-hp, demo-felhom, scratch 9202), built with
Go 1.27.4 (read from the running binary). The HTTPS header is unchanged:
max-age=86400on all 26 public names — the one-year setting is NOT in this release (built on 11-14; the words „one day early is OK” were not given). - Hub 0.181.0 is live since 09:58 UTC, on the operator's yes; Synced/Healthy; 0 ERROR lines.
- The false „root files behind” warning is stopped by the fix, not the stopgap. The hub now compares the files of the two bundles. Proof that no mail is due on 11-17: the three „behind since” marks in the hub's database are empty since 09:59:20 UTC, the alarm check sent 0 alarms, and the System page reads „All boxes look fine”.
- Golden 0.352.1 is baked, proven by one fresh install, and vouched by the operator (10:25:43 UTC).
golden_currency_gate.py: OK, rc 0. - Not proven: that a box's report never timed out in the hub's slow first 8 minutes (R-1393); the vouch through a second channel (read from the hub's log only); that no https trouble happened since 10-15 (not searched, because the date condition already failed); the agent on Go 1.27.4 (release 2).
3. Details
Rulings recorded (Part 0): 09-update-architecture.md §3 decisions 303 (D80 = the root CA copy), 304 (D81 = yes,
with the stopgap word), 305 (the N100's time service: no hand repair; 2026-12-15), 306 (the build order). D80 and D81
are out of the decide table; dated lines on R-1380, R-1392, R-1301, R-1388.
| Version | Digest | Commit | CI (run id, by commit) | |
|---|---|---|---|---|
| Controller | 0.352.1 (rc.1 first) | sha256:1a6e8945211a…13b6bb |
felhom-controller 8747744b |
2637 success |
| Hub | 0.181.0 | sha256:be2e9e5676ee…d1f429 |
felhom.eu 83316fc9 (image), 63f69b99 (manifest) |
2636 success, 2638 success (read before the sync) |
| Golden | 0.352.1 | 4f411cbeb4b2…01346d |
build-golden.sh v3.2.0 | — |
| Agent | 0.173.0, not released | — | — | — |
HSTS (R-1050). Condition 1 (the date) failed, so the other two were not judged. The check date moved to
2026-11-21 with the reason in the row; STATUS.md says how to ask for it.
Go, the 20-minute reads. Controller: seven MaxBytesReader places, each reads the body to its end in the
handler's own goroutine before it answers; no TimeoutHandler. Hub: no MaxBytesReader at all. Neither matches the
advisory's crash pattern. The agent has one place; its turn (release 2) is hygiene. Written on R-1391.
R-1392, the fix. hub/internal/bundlefiles (new): fetch both bundles, accept each only by a sha256 already
trusted, compare everything but agent_version. Fail direction: anything unread → „behind”, as before. Seven tests;
the three the prompt asked for were seen red with the fix line switched off. The two real bundles (0.172.1, 0.173.0)
hold the same 32 files. Design text: 11-os-updates.md §5.4.2 [FACT]. No [DESIGN] text was needed.
Rows: 250 before, 250 after. Opened R-1393 (hub slow after a restart). Closed R-1392. Narrowed R-1391 (the agent is left). Dated lines: R-1050, R-468, R-1380, R-1301, R-1388. (R-1391 and R-1392 were opened by this morning's health check, before this run.)
Machines changed, and how each was left.
- Tester 1, demo-hp, demo-felhom: controller 0.352.1 by the normal update path; per-box floors cleared.
- Scratch 9202: rc.1, then 0.352.1 by hand through its image file.
- demo-hp: VM 9385 made and destroyed with its disks; the installer ISO fetched and removed.
- DooPlex: two image builds, the bake in the drill VM (now off, disk back on
virgin), the hub deploy, two copies of the hub database for reads (shredded). Nothing else. - Hub: host
Tester4-46836fis left for the operator to delete (and its appliance entry, if listed). The hub re-issued Tester4's off-site key and carried its sealed package by itself. No ep0 change by CC. - The N100: took 0.352.1 like any box; no hand step.
Architecture documents read: 09-update-architecture.md (§3 decisions 247, 249, 250, 251, 304), 11-os-updates.md
§5.4.2, 05-hub-architecture.md §17, runbooks/publish-train-rules.md, runbooks/RUNBOOK-manual-build.md §4,
tests/golden-0.352.0-2026-11-11/README.md, the felhom-build-deploy skill. Not read: 08-alarm-ladder.md
(the alarm's wording and audience were not changed).
Capability map: no row changed — nothing a box newly proves. System poster: no listed fact changed; none of
the three acts was needed. unproven.py --summary: 32 walked, 15 partial, 4 built, 4 missing, 23 not walked —
the tool's numbers; no number was moved by this run.
Instruction files: none edited.
Unreleased after this run: felhom.eu hub source holds one commit after 0.181.0 — the „same files” log line is
written once per pair, not at every page view (hub/CHANGELOG.md „Unreleased”). It rides the next hub release.
4. Observations
- The hub is slow for 8 minutes after a restart (24 slow statements, the System page 12.7 s, an enrolment 11.4 s).
FILED: R-1393. - The hub logged the „same files” line 6 times per page view. Fixed in source, unreleased.
NOT-A-FINDING: small, fixed in the session (the size rule). - Tester 1 pushed „household cleared its mail address” at the controller's start.
NOT-A-FINDING: the box's own stored setting says so (email_cleared=true) and the start only repeats it; demo-felhom kept its address through the same update. Not checked: when Tester 1's address was cleared. - demo-felhom's controller start reads 09:18:23Z, 24 s before the floor that caused it.
NOT-A-FINDING: the N100's clock, R-1301. - The hub keeps
bundle_behind_sincemarks of two deleted Tester4 hosts.NOT-A-FINDING: decision 293 (D73) already rules that a host delete removes such rows; not built yet. os_config_bundle_behindfired 4 times before today (last 2026-11-06).NOT-A-FINDING: the same defect, now closed (R-1392).- On the LAN the installer address answers 404.
NOT-A-FINDING: R-1389, already on the operator's list. - The claim's two status lines are not kept in a file (the script printed them to the session only).
NOT-A-FINDING: an evidence gap of this run, said in the golden's README.