PART A -- the poster. documentation/architecture/felhom-system-poster.html
(307 KB). Secret scan first: ZERO IPv4, zero PEM blocks, zero ssh keys, zero
Bearer. The one EAA... match is base64 inside an embedded "mime":"font/woff2"
blob, not a Facebook token. "token"/"secret"/"password" appear 11 times and
every one is a NAME ("6. ep0 read token", "the hub seal key"); the poster
itself says "Names only; no secret values". All five long base64 blobs are
declared assets: 1 image/png, 3 text/javascript, 1 font/woff2.
It renders with NO network: the source mentions cdn.jsdelivr.net and Google
Fonts, but the loaded requests are only the HTML plus blob:/data: URLs -- the
bundler inlined everything. Measured, not assumed, and it matters: this is a
disaster-recovery document, so needing the internet to draw would be a defect.
No console errors.
The operator's three Claude Design fixes are all present: (a) no "WG" badge,
WireGuard only for the tunnel, no badge on the ep0-copy tile; (b) the box ->
ep0 arrow reads "encrypted on the box, sent through WireGuard"; (c) "Known
gaps" holds two items and NOT the household-keys sentence, which is now a
neutral "By design" note under the ep0 household namespace.
ONE FACT ON IT WAS WRONG. The felhom.eu tile said "served from DooPlex through
Cloudflare". It is not: Cloudflare is DNS only and the traffic goes direct --
measured this morning for the privacy notice, which states exactly that. The
poster would have contradicted a published page. Fixed in place (a label):
"served from DooPlex, Cloudflare DNS only". The first wording overflowed the
fixed-size tile, so it was shortened to fit and the evidence lives in the
facts file instead -- checked by re-rendering, not by hoping.
PART B -- the facts and the rule. DESIGN-PROMPT-...md is renamed
felhom-system-poster.facts.md (one home per fact), with the three fixes folded
in as explicit instructions so a regeneration cannot undo them, plus a new
"Badges" section saying a "WG" chip must never come back.
New rule, section 6 "The system poster stays true", added IDENTICALLY to all
five copies of unprompted-work.md (the four repos and the workspace root on
DooPlex; verified identical by diff before and after) and to
PROMPT-TEMPLATE.md's end-of-session checklist as a FIFTH coupled artifact.
scripts/poster_facts_gate.py WARNS when the facts file has a newer commit than
the poster. It never fails a push, deliberately: a refresh needs Claude Design
and the operator, --no-verify is forbidden here, so a blocking gate would leave
deleting it as the only way out. It compares COMMIT times, not mtimes, because
a checkout rewrites mtimes and every fresh clone would shout.
RED-PROOF -- and it found a real bug in the gate. The first run warned
correctly but exited 1: a single non-ASCII character in its own warning raised
UnicodeEncodeError on this cp1250 console. A gate whose entire contract is
"never fails a push" was failing pushes. Fixed (ASCII output + an encode
guard), and the decoy now asserts BOTH the warning and exit 0. Three branches
proven: facts newer -> warns, rc 0; poster newer -> quiet, rc 0; poster
missing -> "could not tell", rc 2, not a false all-clear.
The decoy itself was seen to fail, twice, on Linux (the suite needs fcntl and
cannot run on Windows): breaking the warning gives STALE_WARNS=False, and
making it exit 1 gives RC_STALE=1. All 80 felhom.eu decoys behave.
PART C -- do box reports pass through Cloudflare? NO. Two channels. DNS from
PUBLIC resolvers (not DooPlex's own, which answers the LAN address):
hub.felhom.eu is a CNAME to dooplex.hopto.org -> 37.191.56.193, not a
Cloudflare address, and no cf-ray comes back. The manifest: an ordinary k3s
Ingress, Cloudflare named only in a DNS setup comment. THE CONTROL that makes
the negative mean something: iso.felhom.eu resolves to 172.67.x / 104.21.x,
real Cloudflare addresses -- so the method does detect proxying.
So nothing is added to the Cloudflare row: the hub path does not touch it.
06-offsite-connectivity.md section 1 claimed the public edge is a
Cloudflare-Tunnel and "DooPlex has no public IP" -- both untrue today. Kept
and marked STALE with the measurement rather than rewritten, because that
paragraph is the reason ep0 exists and the argument needs its premise visible.
total-loss-of-dooplex.md's "today a CNAME to dooplex.hopto.org" is confirmed
correct.
Register: 137 before, 137 after, 0 opened, 0 closed -- every finding here was
small and fixed in the session.
Decision 150 (operator 13:25): sessions correct stale facts in instruction files themselves,
naming each edit; never loosen a rule. Added to unprompted-work.md §5 (all copies) and
PROMPT-TEMPLATE.md §9. CLAUDE.md gates paragraph (R-891), architecture pointer, docs/website
rules, the doubled R-286 sentence in the workspace CLAUDE.md.
Decision 149: vaultwarden's step proven on bench 9401 and box 9202 (evidence here).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS
The hub has written every customer e-mail in Hungarian whatever the box was set
to. The box has published its language since controller v0.247.0; nothing read
it. Now it does.
Nothing an operator reads changes. The Hungarian mails are byte-identical, and
that is a diff rather than a reading: 56 goldens per language captured from
v0.117.0 BEFORE any string moved, and all 56 Hungarian ones pass unchanged after
every sentence was routed through the new bundle.
- internal/i18n: flat bundle, 79 keys, hu authoritative + hu fallback, ceiling 0.
- customerMessages/severityLabels are DERIVED from the bundle, so a sentence is
written in one place and all 40+ tests that read those maps still work.
- Language order: last reported -> created-with -> hu. reports.language defaults
to EMPTY, never hu: "never told us" is not "chose Hungarian".
- message_customer on POST /api/v1/event, additive and optional forever, for the
sentences the box composes and the hub cannot translate.
- The bind page is per-language, and its `expired` state stays Hungarian: it is
the state an unknown token lands in, so rendering a real English customer's
token in English would make the LANGUAGE answer what the TEXT refuses to.
Two defects found inside the release:
- R-581: the newest report was picked by received_at, which has SECOND
granularity, so same-second reports tied and the winner was arbitrary. Ordered
by the autoincrement id now. GetCustomers() still has the shape - row open.
- R-582: the English copy-guard stems, ported word for word from Hungarian,
convicted 141 honest sentences. The English claim is a phrase with a modal.
R-555 closed: the language allowlist entry is out of wire_contract_gate.py.
hub_copy_gate.py follows the sentences into the bundle - without that it would
have scanned four files that no longer hold any customer text and reported
success. Three new decoys incl. an innocent control.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS
10-localisation.md §10.3: the saved notes follow the box language at write time, with the
one-night consequence stated rather than hidden; the globe, and the table of WHO reads which
page and where its globe posts — getting that wrong makes the button do nothing, which it did
on /recovery until the live probe found it. Decision 6 superseded a second time; decision 8
(a claim carries the visitor's language) recorded. Decision 5 of §11's anonymous-surface line:
changing what a VISITOR reads is within what an anonymous request may do; changing anything the
household owns is not, and POST /lang can do only the first.
R-578 — the deadlock, and why it is a row rather than a fixed bug: UpdateOffboxStatus holds the
settings write lock while running its callback, boxLang() wants the read lock, sync.RWMutex is
not reentrant. On a real box an off-site run would have hung FOREVER holding that lock. The
symptom was a test suite going from 8 minutes to a 25-minute timeout. Fixed and guarded, but the
guard covers one package and three helper names; the class needs a gate.
R-577 — a guest share visitor still has no way to pick a language, and the household's setting
is the wrong default for a stranger. Deliberately left, pinned by a test, and the operator's to
decide because it is a promise the share feature makes.
.claude/rules/live-probes.md, unconditional: never send a deploy request for an app that is not
installed, not even expecting a refusal — the endpoint accepts first and validates later. Two
sessions made that mistake in two days, the second WITH a prompt line forbidding it. A prompt is
read once; a rule file is loaded every session.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS
R-549 (operator ruling A): controllerStatus hardcoded 30m/1h while both
staleness checkers and hostStatus read alerting.stale_threshold. Moving the
threshold to 45m would have painted a customer amber 15 minutes before the
alarm could fire - the second definition rollup.go's header forbids. It now
reads the same value, down at 2x. Both 'checker initialized' log lines print
the threshold, which no line did before.
R-539 (ruling 3 of 2026-09-16): controller_slow_crashloop (warning,
operator-only), minted when the agent's slow_crashloop_since moves, with the
fast sibling's first-sight rule.
R-550: restore_interrupted (warning, for the household) allowlisted with a
Hungarian customer message.
Red-proofs, each seen failing then passing: the status test with the old
hardcoded numbers; the checker test with the movement branch removed; the
operator-only test with the registration removed; the household-message test
with the Hungarian entry removed (asserted on the SUBJECT - the body
legitimately repeats the raw message, which my first version of the test
mistook for a fallback).
go build/vet/test ./... green, 18 packages.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS
227 -> 115 effective lines, split into .claude/rules/{hub,website,manifests,docs}.md, and
repo_gates.py gains gate 6. Trim first, register second: a registered-but-failing gate refuses
every push through the pre-push hook, which is why this repo -- the one that OWNS the gate --
was the only one not running it.
Register discipline and the R-110 installer fence deliberately stayed in the core; both have
triggers no fixed glob covers, and scoping them would have rebuilt the failure class they exist
to prevent.
Scoping proven from the InstructionsLoaded hook log in two fresh sessions, not from frontmatter.
admin