REPORT: the pilot's English quoted in full for the operator's read (R-560)
gates / gates (push) Failing after 2s
gates / gates (push) Failing after 2s
The three apps' English text, the per-app table, the unverified UI labels for the slice-6 walk, the gate's five checks and the three defects its own decoys found in it, and the live proof on both demo boxes. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS
This commit is contained in:
@@ -1,32 +1,203 @@
|
||||
# REPORT — vaultwarden closed by default, paperless-ngx sized for a batch (2026-09-15)
|
||||
# REPORT — localisation slice 5, Part B: the copy freeze, the gate, and the English pilot (R-560)
|
||||
|
||||
Task: *before the volunteer — the big night's P1 fixes*, Part E. No image or version change.
|
||||
2026-09-20 · catalog `main` `e81d41e` · needs controller **v0.257.0** to render · Part C not started.
|
||||
|
||||
## What changed
|
||||
- **vaultwarden (R-512):** `SIGNUPS_ALLOWED` defaults to `false`; the select explains both choices; „Első lépések" is
|
||||
invite-first (admin panel with the generated token → invite → create account); the compose header no longer tells
|
||||
anyone to change a setting after install.
|
||||
- **paperless-ngx (R-514, R-515):** `PAPERLESS_TASK_WORKERS=1`, `THREADS_PER_WORKER=1`, webserver `memory: 1280M`
|
||||
(`mem_limit` hint 1664M); `default_creds: admin / admin` removed; first steps point at the generated password.
|
||||
---
|
||||
|
||||
## Measured first
|
||||
- **E.1 spike** (throwaway container on scratch 9202, SMTP off as on a box without app-email): stranger register **400**,
|
||||
admin invite **200**, invited register **200**, second stranger **400**. So invite-first needs no controller change.
|
||||
- **E.2** (1 worker × 1 thread, no cap): 20 three-page PDFs → **20/20**, `memory.peak` 783 294 464 B → cap = peak × 1.5
|
||||
rounded up to 256M = **1280M**. Caveat: the PDFs carried text, so OCR on scanned images may need more.
|
||||
## For the operator — the one thing you have to do
|
||||
|
||||
## Proven live after the change (scratch 9202, deployed from the catalog through the controller API)
|
||||
- Vaultwarden: container `SIGNUPS_ALLOWED=false`; stranger `send-verification-email` via traefik → **400 „Registration
|
||||
not allowed or user already exists"**; the app page shows the invite steps (fragment 1, negative control 0).
|
||||
*One invalid attempt first: a hand-built register body got 422 (did not parse) — marked invalid in the evidence.*
|
||||
- Paperless: 20 PDFs posted at once → tasks **20 SUCCESS**, documents **20**, limit 1 342 177 280 B, `memory.peak`
|
||||
**772 370 432 B**, OOMKilled false, 0 kernel OOM lines.
|
||||
- Forcing an OOM (cap 128M) showed Docker reports nothing in an LXC guest (R-528, a controller/agent row). **Mistake,
|
||||
stated:** restoring the cap on the throwaway with a broad `sed` also changed its redis cap; moot, the stack was removed.
|
||||
**Read the English below (it is short) and say "go", or say what to change.** Fifty more apps follow
|
||||
whatever you say. Ten minutes now against a day of rework later. Everything else in this report is
|
||||
for the record.
|
||||
|
||||
---
|
||||
|
||||
## The three pilot apps, in full, for your read
|
||||
|
||||
### PrivateBin
|
||||
|
||||
> **What it is:** Encrypted note and text sharing
|
||||
> **Tagline:** Encrypted text sharing - the server never sees the content
|
||||
|
||||
**What it is for**
|
||||
- Share sensitive text safely
|
||||
- End-to-end encryption - the server cannot reach the content
|
||||
- You choose how long it lasts (5 minutes to 1 year, or never)
|
||||
- Optional: delete it automatically after it is read
|
||||
- Password protection for extra safety
|
||||
|
||||
**First steps**
|
||||
1. Open paste.DOMAIN in your browser
|
||||
2. Type your text and select Send
|
||||
3. Share the link you get - the encryption key is part of the URL
|
||||
|
||||
**Install settings:** Domain — "The server domain name" · Subdomain — "The subdomain this app
|
||||
answers on"
|
||||
|
||||
### Paperless-ngx
|
||||
|
||||
> **What it is:** Digitise your documents and keep them in order
|
||||
> **Tagline:** A digital filing cabinet - scanning, OCR and automatic sorting
|
||||
|
||||
**What it is for**
|
||||
- Turn paper documents into files and keep them in order
|
||||
- Automatic OCR text recognition on scanned documents
|
||||
- Smart automatic sorting and tagging
|
||||
- Full text search across every document
|
||||
- Documents that arrive by e-mail are processed automatically
|
||||
|
||||
**First steps**
|
||||
1. Open paperless.DOMAIN in your browser
|
||||
2. Sign in: user "admin". The password is on the Settings page, under Automatically generated values
|
||||
3. Upload a document in the app, OR drop it into the import/paperless folder in the File manager
|
||||
(FileBrowser) - it is read in and processed with OCR from there. The File manager sign-in is on
|
||||
the File manager app page
|
||||
4. You can upload many documents at once: they are processed one by one, and a larger batch takes a
|
||||
few minutes
|
||||
5. Create tags and correspondents so new documents sort themselves
|
||||
|
||||
**Before you start**
|
||||
- An external hard drive is recommended to keep the documents on
|
||||
- At least 1 GB of free RAM (for the OCR work)
|
||||
|
||||
**Install settings:** Database password · Encryption key · Admin user name · Admin password ("For
|
||||
the first sign-in. You can change it in the app afterwards.") · Data storage path ("The path to the
|
||||
external hard drive where the documents are kept") · OCR language (Hungarian / English / Hungarian +
|
||||
English / German + English — "The language the text recognition reads") · Folder label: "Documents
|
||||
to read in"
|
||||
|
||||
### RomM
|
||||
|
||||
> **What it is:** Retro game collection manager
|
||||
> **Tagline:** Retro game collection manager, browser and player
|
||||
|
||||
**What it is for**
|
||||
- Sort and browse a retro game collection in your browser
|
||||
- Game details download themselves - cover art, descriptions, ratings
|
||||
- Filter and search the whole collection by platform
|
||||
- Upload and download ROM files in your browser
|
||||
- Several people in the household can have their own account
|
||||
|
||||
**First steps**
|
||||
1. Open arcade.DOMAIN in your browser
|
||||
2. Sign in with the default admin / admin account
|
||||
3. Change the password straight away, in the Settings menu
|
||||
4. Upload the ROM files into the roms/ folder in the File manager (FileBrowser), sorted into
|
||||
platform folders (for example roms/gba/, roms/snes/)
|
||||
5. Start a Scan from the menu on the left to read the ROMs in
|
||||
6. Optional: set up metadata providers so cover art and descriptions download themselves (see below)
|
||||
|
||||
**Before you start**
|
||||
- An external hard drive is needed to keep the ROM files and cover art on
|
||||
- At least 1 GB of free RAM is recommended (MariaDB + Redis + RomM)
|
||||
- ROM files sorted into platform folders (for example roms/gba/, roms/snes/)
|
||||
|
||||
**Optional settings — "Metadata providers":** "So that cover art, descriptions and ratings download
|
||||
themselves. IGDB is the one worth setting up first. All of them are free after you sign up." · six
|
||||
key fields with their own sign-up instructions · Folder label: "ROM collection"
|
||||
|
||||
---
|
||||
|
||||
## Per-app table
|
||||
|
||||
| app | copy strings | English | list-length exceptions | push | live-checked |
|
||||
|---|---|---|---|---|---|
|
||||
| privatebin | 14 | 14 | none | pilot | **yes** — app page, Apps list |
|
||||
| paperless-ngx | 33 | 33 | none | pilot | **yes** — app page, deploy page, Apps list |
|
||||
| romm | 42 | 42 | none | pilot | **yes** — app page, deploy page, Apps list |
|
||||
| the other 50 | 943 | 0 | — | — | — |
|
||||
| **total** | **1 032** | **89** | **none** | | |
|
||||
|
||||
No app needed a list-length exception: every English list has exactly as many entries as its
|
||||
Hungarian twin, which is what the gate requires unless the file carries a stated reason.
|
||||
|
||||
## Unverified UI labels — for the slice-6 walk (R-561)
|
||||
|
||||
These English lines name a button or a menu in the app's OWN interface. No gate and no endpoint can
|
||||
tell whether the app actually shows that word in English; somebody has to open it.
|
||||
|
||||
| app | the line | what has to be checked |
|
||||
|---|---|---|
|
||||
| privatebin | "Type your text and select **Send**" | PrivateBin's send button's English label |
|
||||
| paperless-ngx | "under **Automatically generated values**" | this is Felhom's own Settings page — taken from the controller's English bundle, so it is right by construction unless that bundle changes |
|
||||
| paperless-ngx | "Create **tags and correspondents**" | Paperless's English words for these two |
|
||||
| romm | "Change the password straight away, in the **Settings** menu" | RomM's English menu label |
|
||||
| romm | "Start a **Scan** from the menu on the left" | RomM's English button label |
|
||||
|
||||
## What changed in this repo
|
||||
|
||||
**Three commits, in this order, and the order is the point.**
|
||||
|
||||
1. `0c19b45` — **the freeze alone.** `scripts/copy_freeze/hu.json`: every customer-facing Hungarian
|
||||
string in the catalog, 1 032 of them across 53 apps, captured from the commit before any
|
||||
translation existed. Its own commit so that a baseline cannot be said to have arrived with the
|
||||
checker that reads it.
|
||||
2. `84e0584` — **the gate.** `scripts/check-copy-i18n.py`, fifth row of `catalog_gates.py`, static
|
||||
and in the pre-push hook: the Hungarian freeze, the English block's structure, its language, its
|
||||
credential tokens, and a coverage ratchet. 18 new decoy cases.
|
||||
3. `e81d41e` — **the pilot.** Three apps, 89 strings, `EN_MISSING_CEILING` 1032 → 943.
|
||||
|
||||
**No image, no pin, no `catalog_since`, no compose line changed.** A translation is not a version
|
||||
bump of an app.
|
||||
|
||||
## What the gate checks, and what it cannot
|
||||
|
||||
| # | check | convicts on |
|
||||
|---|---|---|
|
||||
| 1 | FREEZE | any Hungarian copy string that differs from the freeze, is new, or has gone. Runs on all 53 apps whatever scope is named — a scoped push that quietly edits a neighbour is what a freeze is for. A new app must be admitted with `--add-app NAME --reason "…"` |
|
||||
| 2 | STRUCTURE | a key the English block may not carry; an `env_var`, option `value`, `match_group`, `target` or `path` with no Hungarian twin (it would be INERT on the box and the translator would never learn); a list whose length differs from the Hungarian |
|
||||
| 3 | LANGUAGE | an accented Hungarian letter; **ASCII-only Hungarian**; "please"/"kindly"; an English retrieval promise the Hungarian does not make; a dropped app name or „Felhom" |
|
||||
| 4 | CREDENTIALS | a login token inside `default_creds` or an initial-credentials note that did not survive translation, matched on WORD boundaries |
|
||||
| 5 | RATCHET | a coverage count above **or** below `EN_MISSING_CEILING`, counted over the whole catalog whatever scope is named |
|
||||
|
||||
**It cannot judge whether the English says what the Hungarian says**, and it cannot judge whether an
|
||||
app's first steps match that app's real screens. The first is your read above; the second is the
|
||||
table before it.
|
||||
|
||||
## Numbers that were wrong in the plan, corrected by measurement
|
||||
|
||||
| | the plan said | measured on `94bc5febaca2` |
|
||||
|---|---|---|
|
||||
| copy strings | 835 | **1 032** (it omitted 13 `placeholder` values, and every ASCII-only label) |
|
||||
| with a Hungarian letter | 832 | **832** ✔ |
|
||||
| ASCII-only Hungarian | three — „Igen", „Nem", „Nincs" | **~120**, and those three do not occur in this catalog at all |
|
||||
|
||||
The last row decides an instrument rather than a footnote. „Aldomain" appears 53 times and „A szerver
|
||||
domain neve" 53 times; both are Hungarian with no accent in them. **A gate that scans for accents
|
||||
passes every one of them inside an English block.** So check 3 folds the text and matches word-bounded
|
||||
stems, and prints a positive and two negative controls on every run.
|
||||
|
||||
## Three defects in the gate, each found by a decoy rather than by reading it (R-592)
|
||||
|
||||
1. **Coverage was counted only for the apps named on the command line.** `check-copy-i18n.py
|
||||
privatebin` reported 47 more untranslated strings than the same tree unscoped — a ratchet anybody
|
||||
could loosen by naming one app.
|
||||
2. **The credential check matched its token as a bare substring**, so „admin / adminadmin" translated
|
||||
to "administrator / hunter2" passed: "admin" is inside "administrator".
|
||||
3. **The ASCII-Hungarian stems matched as bare substrings too**, so „ird be" convicted "the third
|
||||
best" and „angol" convicted "Angola".
|
||||
|
||||
All three now have their own case. 33 decoy cases in total, every one seen to convict or to pass as
|
||||
intended.
|
||||
|
||||
## Live proof
|
||||
|
||||
Endpoint-level, from inside the guest, with the customer `Host` header and a real session — the same
|
||||
handlers and templates the browser drives, only the drawing skipped. Said plainly because "the page
|
||||
shows X" has to name how it was seen.
|
||||
|
||||
- **demo-hp (controller 0.257.0):** the three English app pages, the Apps list and two deploy pages
|
||||
show the English catalog text. The seven **Hungarian** pages are byte-identical before and after the
|
||||
push apart from the per-session CSRF token.
|
||||
- **demo-felhom (controller 0.255.0, deliberately old):** with the pilot synced onto the box —
|
||||
confirmed positively, the sync named the three apps and the block is present in both the cache and
|
||||
the stack copy — its pages hash identically before and after and its log carries no parse warning,
|
||||
in a 93-line window that includes the sync's own lines.
|
||||
|
||||
Evidence: `felhom.eu/documentation/audits/i18n-slice5-2026-09-20/`.
|
||||
|
||||
## Teardown
|
||||
Machine: scratch 9202's throwaway `paperless-ngx` and `vaultwarden` stopped and removed (volumes removed; the paperless
|
||||
data folder on the scratch drive kept, the product default). Host: nothing. Hub: nothing (9202 reports to no hub).
|
||||
|
||||
## Gates
|
||||
`catalog_gates.py --fast` OK on every push. Evidence: `felhom.eu/documentation/audits/evidence-p1fixes-2026-09-15/E1-*`, `E2-*`, `teardown-9202.txt`.
|
||||
Nothing installed, nothing removed, no app deployed. Three template files changed on each demo box,
|
||||
which is what a catalog sync does. The staged password file and the capture scripts were shredded from
|
||||
both Proxmox hosts and both guests. Both boxes left on Hungarian.
|
||||
|
||||
Reference in New Issue
Block a user