docs: §8.2f — no notice, no deletion, as built; the texts after decision 316; what a taken-over box can still do; 08 §6.5 the two new operator warnings
gates / gates (push) Successful in 2m11s

Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS
This commit is contained in:
admin committed 2026-11-15 16:42:53 +01:00
1 parent c9d67309e2
commit b8a48f7020
4 files changed
+157 -27

No files matched your search

+1 -1
View File
@@ -80,7 +80,7 @@
| `store.NoteAppList` / `AppPresences` + `offsitekeys.agreeEnded` (D72/D84, `07` §8.2f) | hub/internal/store/offsite_ended.go, hub/internal/offsitekeys/ended.go | `NoteAppList(customerID, deployed, claimed, now)`; `agreeEnded(now, presence, notices, listing, haveListing, claims)` | The hub's OWN record of when an app left a box's app list (fed by `handleReport`), and the decision which snapshot ids of ended apps a window may remove outside its cap | **The box's dates decide nothing.** An id is agreed only when its app has been absent 30 days by the hub's clock AND the hub's own listing shows the file older than that absence. `WindowCredit.CapExplains` is the cap's part, `EndedIDs` the agreed ids — a reader that explains ids by name must not also spend `Explains`. **After the review of 2026-11-15:** one window agrees to at most the cap's size from the hub's OWN count (`endedCap`/`boundEndedToCap`); a request for more than half of the store waits a week — its first sight is the hub's own row (`store.NoteLargeEndedRequest`, key `largeSetKey` = a hash of the whole app set), never a search of text a box could write; `store.NoteAppListClaims` returns never-seen apps once for the operator's warning; a claimed name that is not an app name is written nowhere (`ledgerName`). |
| `notify.householdReceiptEvents` + `Dispatcher.SendEndNotice` (D84; the notice lock, `09` §3 decision 315) | hub/internal/notify/end_notice.go | register; `SendEndNotice(customerID, app, ends) (EndNoticeOutcome, why)` | The mail that tells the household about its own press (one type: `offsite_end_scheduled`). **The box's event mails nobody**: the hub sends the notice from its own state (`api.endNotices`, at every report, until the mail service accepts it) | Bypasses the household's group switches; the hub writes it from a checked app name and its OWN day. `SendEndNotice` keeps no state and tries once: only `EndNoticeSent` tells the household; the daily limit (5 per household, counted in the sent log) answers `EndNoticeLimit` — the caller tries again, it never drops. **Every new entry needs its own ruling** — pinned by `TestD84_Receipt_OpensNothingElse`, `TestNotice_*`. |
| `store.NoteEndNotices` / `EndNotices` / `MarkEndNoticeTold` + `api.endNotices` (the notice lock, `07` §8.2f) | hub/internal/store/offsite_end_notice.go, hub/internal/api/end_notice.go | `NoteEndNotices(customerID, deployed, claims, tierOn, now)`; `EndNotices(customerID)`; `MarkEndNoticeTold(customerID, app, press, to, now)` | The hub's OWN record that a household (or the operator in its place) was told that an app's off-site copies will be deleted. `offsitekeys.agreeEnded` agrees only when it says „told” 30 days ago | **Written only by the hub, only after the mail service accepted the mail; never derived from the sent log or from a box's date.** `press` (the box's `removed_at`) is a key only: another value VOIDS the notice. An install, a take-back (a report that shows the tier ON and no record) and a household delete remove the row. A new condition of a deletion goes HERE, not into `notification_log`. |
| `Dispatcher.SendOperatorNow` | hub/internal/notify/end_notice.go | `SendOperatorNow(customerID, eventType, severity, message, detailsJSON) error` | An operator mail whose DELIVERY is a condition of something else: it answers only when the mail service has (nil = accepted) | Operator-only types only; no cooldown and no background retry — the caller keeps its own „told once” record and tries again. For every other operator mail use `ProcessEvent`. |
| `Dispatcher.SendOperatorNow` | hub/internal/notify/end_notice.go | `SendOperatorNow(customerID, eventType, severity, message, detailsJSON) error` | An operator mail whose DELIVERY is a condition of something else: it answers only when the mail service has (nil = accepted) | Operator-only types only; no cooldown and no background retry — the caller keeps its own „told once” record and tries again, **and bounds itself** from the sent log (`store.CountOperatorSentSince`, `OperatorSentAboutSince`: per household a day, per thing a week — `api.endNoticeToOperator`). For every other operator mail use `ProcessEvent`. |
| `accesslog.New` / `Middleware` / `Mark` (v0.158.0, R-1140) | hub/internal/accesslog/accesslog.go | `Mark(ctx, kind)` | The hub's own access log (`<data_dir>/access-log/`, 90 days) and marking the identity KIND an auth check established | Never log a value: the path only (no query; `/bind/<token>` and 32+-char segments redacted). A new auth path calls `Mark` with its kind. A logging failure never fails a request. |
| the timed SQLite driver `sqlite-timed` + `SetSlowStatementThreshold` (v0.178.0, R-1346) | hub/internal/store/timeddriver.go | `store.New` opens it — nothing to call | Names a statement over 1 s: `[WARN] slow statement: <ms> (over 1s): <SQL>` | The SQL text only, never a bound value. Put a heavy statement's SQL in a const and pin its plan (`r1346_speed.go`, `TestR1346_HeavyStatementsUseTheirIndex`): on the hub's network disk a whole-table read is seconds to minutes. |
| `(*Store).SaveOneTimeSecret` / `OffsitePassword` / `SealLegacyOffsiteSecrets` | hub/internal/store/offsite_seal.go | — | Storing / reading the sub-account password | **Sealed AES-256-GCM; no key → refused (fail-closed).** Under `go test` every store gets a fixed key (`testing.Testing()`); production needs `OFFSITE_SECRET_KEY`. Never serve the value to a box. |
@@ -1835,7 +1835,7 @@ minutes). **[FACT]** 9202 rc.6: an image that exits after 20 s → not started,
unit back → success, copy released 3 minutes later. Tester 1 on 0.343.0: a unit restore and a set-aside restore, each
„ran stably for 3m0s … the restore copy is released" (`audits/release-2026-10-26/`).
### 8.2f The off-site copies of a removed app end 30 days after „delete backup data” `[FACT — built, not released: controller `f628cb5` and hub `01a106b7` + `a63f5b2b` (after v0.182.0), 2026-11-15; NO box has run it]` (D72 and D84, `09` §3 decisions 292 and 309, R-1374 part 1; designed 2026-11-14)
### 8.2f The off-site copies of a removed app end 30 days after „delete backup data” `[FACT — built, not released: controller `f628cb5` and hub `01a106b7` + `a63f5b2b` (after v0.182.0), 2026-11-15; the notice lock and the text changes: hub `42ef25c2`, `9ef8cee1`, `c9d67309` and controller `10df99b`, 2026-11-15 evening; NO box has run it]` (D72 and D84, `09` §3 decisions 292 and 309; the notice lock and the texts, decisions 315 and 316; R-1374 part 1; designed 2026-11-14)
**The ruling.** When the household removes an app and presses „delete backup data” (`remove_backups`), that app's
off-site copies end 30 days later, at the next clean-up after that day — the number of a restore copy (D62). When the
@@ -1876,6 +1876,7 @@ section adds one thing to that window.
names and upload times) holds it, uploaded before that absence (`offsitekeys/ended.go`). A refused app is named,
with the reason, in the answer, on the window's own row and on the customer page's window block; a clear
disagreement raises `offsite_ended_app_refused` to the operator. The box's dates decide nothing on the hub.
**And only when the hub's own record says the household was told 30 days ago** — „No notice, no deletion”, below.
For an app the hub never saw installed (removed before the hub kept this record) the hub's clock starts when the
box first names the app in a pending end record (the report's `offsite.ended_apps`).
5. **Local copies are unchanged.** They are deleted at once, as before. The dialog says both.
@@ -1890,13 +1891,74 @@ cap plus those. `store.WindowCreditsBetween` gives the two parts apart.
itself agreed to — are explained by name and raise nothing. Every other vanished file is judged against the cap's
part alone, exactly as before, and still raises `offsite_snapshots_vanished`.
**The mail (D84)** `[FACT — built, not released]`. At the press the box sends `offsite_end_scheduled`; the hub mails
it to the household's own notification address — once per press, in the household's language, whatever its group
switches say (a receipt: `notify.householdReceiptEvents`), and not to the operator. **The hub writes the mail, not the
box** (review of 2026-11-15): subject and body come from the hub's own bundle, with an app name the hub checked and
the end date it parsed from the details; the box's sentence is never mailed. A household gets at most 5 receipts in
24 hours (counted in the hub's sent log); the rest are dropped and said once a day in the hub's log. **A household with no
notification address gets no mail: the page is all it has.** A blocked household gets none either.
**No notice, no deletion — the notice lock (D84; `09` §3 decision 315, operator, 2026-11-15 evening)**
`[FACT — built, not released: hub `42ef25c2` + `c9d67309`; no box has run it, no mail has reached a real inbox]`.
*Why.* In the first build the household's mail left once, when the box's event `offsite_end_scheduled` arrived. A
box's event is lost when the hub is away for nine seconds (R-1361); a send that failed was not tried again; a sixth
receipt in 24 hours was dropped. In each case the hub still agreed 30 days later, because the absence alone decided.
The mail D84 promises was not a condition of the deletion. Now it is.
*The design, as built.*
1. **The hub sends the notice itself, from its own state.** Every controller report names the box's pending end
records (`offsite.ended_apps` — while the off-site tier is switched on). The hub keeps one row per (household,
app) such a report names (`offsite_end_notices`, `store.NoteEndNotices`) and, after each report, mails every row
that is not told (`api.endNotices` → `notify.SendEndNotice`) — at that report and at later ones until the mail
service accepts the mail. Only then does the row say „told”, with the hub's time. One row per press: one notice
per press, however many reports name it. **The box's event mails nobody**; it is the operator's record only. So a
lost event, an event sent twice and an event that never comes all change nothing.
2. **The hub agrees only 30 days after the notice.** `offsitekeys.agreeEnded` now needs BOTH: the app absent from
the box's app list for 30 days by the hub's clock (as before), AND the hub's own row saying „told” at least 30
days ago by the hub's clock. No row, a row not told, a row told later: refused, and the window's answer names the
reason. The day in the mail is the hub's own — the day of the mail + 30 days, in Budapest time — not the box's.
3. **Where „told” is kept.** In the hub's own table, written only by the hub and only after the mail service
accepted the mail. Not derived from the sent log (`notification_log` — nothing prunes a living household's rows
today, but a condition of a deletion must not rest on a log), and never from a box's word. The box's `removed_at`
is kept only as the KEY of a press: another value voids the notice; no value can make one.
4. **A take-back or an install voids the notice.** The row is deleted when a report lists the app as installed, or
when a report that shows the off-site tier ON no longer names the app (a take-back; also after the window has
run). Another press (another key) resets it. A new press needs a new notice and a new 30 days. A report that
does not show the tier (it is off) changes nothing: its silence is not a take-back.
5. **The limit per day delays; it never drops.** At most 5 notices per household in 24 hours, counted in the hub's
sent log; the sixth stays „not told” and goes at a later report.
6. **A household with no address, or a blocked one — the operator is told in its place, and that starts the 30
days.** One mail per app (`offsite_end_notice_operator`, operator-only, sent with `SendOperatorNow`, which
answers only when the mail service has). When the household gets an address later, the hub mails the household
and **the 30 days start again from that mail**. Bounded (security review of the commit): at most 5 such operator
mails about one household in 24 hours, and the same app again no sooner than 7 days; a notice that waits for
these limits is NOT told. *Both halves are a choice — decided by CC unattended on 2026-11-15; the operator may
reverse it.* The other choices: no deletion ever for a household with no address (cost: a household that never
gave an address can never have off-site copies deleted this way), or the operator's mail stays the one that
counts when an address comes later (cost: a person who holds the login can remove the address, press, and put
the address back on day 29 — the household then reads „deleted tomorrow”).
7. **A notice that cannot go out** (the mail-hold marker, a send that fails, no mail service): nobody is told, the
row keeps the reason in the hub's own words, the next try is at a report at least one hour later, and nothing is
agreed. **After 3 days** without a notice the operator gets one line (`offsite_end_notice_stuck`, once per app, at
most 5 per household a day, not while the mail hold is on). *Why 3 days:* it is longer than any outage of the
mail service the hub has had, longer than the hold on a restored copy is meant to last, and longer than the
daily limit delays an honest household (15 apps removed at once are all told inside it) — so the line means
„this will not heal by itself”; and it is short enough that a deletion the household asked for has lost three
days, not a month, when the operator hears of it.
8. **A box that asks a window for an app it never named in a report** is refused („the hub has no record that the
household was told”) and the operator is told (`offsite_ended_app_refused`): an honest box names a pending record
in every report.
9. **The hub's database put back from a copy** that lacks the row: there is no notice. The hub mails again at the
next report and waits 30 days again. (A restored copy is started with the mail-hold marker in place — `hub/internal/mailhold` — so until the
operator releases it nobody is told and nothing is agreed.)
10. **The controller is not changed.** It already reports its pending records. An older controller reports none:
nothing is told and nothing is agreed, as before.
*Accepted, and the operator may reverse it (not a decision):* the notice ignores the household's notification
switches; no end record on a box whose off-site target is the household's own NAS; „delete in 30 days” is offered for
every removed app with copies and no mark.
*A limit of the built design* `[DESIGN — proposed, not built]`: the controller names its pending records only while
the off-site tier is switched ON (`OffboxStatusForReport`). A household that presses „delete backup data” while the
tier is pinned but switched off gets no notice until the tier is on again — the page shows a date, the mail comes
later, and the 30 days count from the mail. No copy is deleted meanwhile (no window runs either). To close it the
controller would report `ended_apps` also with the tier off; that changes the report's shape for a tier that is off
and needs a test on a box with an older hub. Not built; named in the build's report for the register.
**Edge cases, each with its rule and its test.**
@@ -1906,7 +1968,20 @@ notification address gets no mail: the page is all it has.** A blocked household
| The app is installed again after the copies are gone | Nothing special: a new history starts | — |
| The household kept the backups at removal | No record. The copies stay with no end. The off-site list offers „delete in 30 days” (the next row) | `TestD72_NoRecordNoDeletion`, `TestD72_RemovalWritesTheEndRecordOnlyWithDeleteBackupData` |
| The household changes its mind | „Keep it after all” until the window has run; after that the page says the copies are not marked or are already deleted | `TestD72_StartAndKeepHandlers`, `TestD72_AfterTheTakeBackTheWindowRemovesNothing` |
| Two presses for the same app | One record, one mail; the date does not move | `TestD72_EndRecord_ASecondPressChangesNothing`, `TestD84_OnePressOneMail`, hub `TestD84_Receipt_OncePerPress` |
| Two presses for the same app | One record, one mail; the date does not move | `TestD72_EndRecord_ASecondPressChangesNothing`, `TestD84_OnePressOneMail`, hub `TestNotice_AReportNamesAnApp_TheHubMailsTheHouseholdOncePerPress` |
| **The notice lock's wrong cases** | | |
| The box's event never arrives, or arrives twice | The event mails nobody; the notice comes from the report | hub `TestNotice_TheBoxesEventMailsNobody`, `TestNotice_AReportNamesAnApp_…` (no event is posted in it) |
| The box names the app only in the window request, never in a report | No notice row: refused, the operator is told | hub `TestNotice_Window_TheNoticeRecordDecides` (a), `TestNotice_AgreeEnded_NoNoticeNoDeletion` |
| The hub's database is put back from a copy that lacks the notice | No row: the hub mails again and waits again | hub `TestNotice_ADatabaseWithoutTheNoticeTellsAgain` |
| The address is added after the press | The operator was told in its place; the household is mailed when the address appears, and the 30 days start again | hub `TestNotice_NoAddress_TheOperatorIsToldInItsPlace_AndAnAddressLaterRestarts` |
| Press, take-back and press again within one day | The take-back deletes the notice; the second press gets its own notice and its own 30 days | hub `TestNotice_ATakeBackOrAnInstallVoidsTheNotice`, `TestNotice_Record_WhatVoidsIt` |
| Two apps pressed in one minute | Two rows, two notices | hub `TestNotice_TwoAppsInOneMinute_TwoNotices` |
| A household deleted between the notice and the window | The notice rows go with the household; nothing is agreed, nobody is mailed | hub `TestNotice_ADeletedHouseholdHasNoNotice`, `TestNotice_Record_GoesWithTheHousehold`, `TestNotice_Window_…` (e) |
| The send fails, or the mail hold is on | Not told; tried again an hour later; told only when the mail went; one operator line after 3 days | hub `TestNotice_AFailedSendIsTriedAgainUntilItHasGone`, `TestNotice_TheMailHoldTellsNobody`, `TestNotice_StuckForThreeDays_TheOperatorGetsOneLine` |
| Seven apps removed in one day | Five notices that day, two the next; none is dropped | hub `TestNotice_TheDailyLimitDelaysAndNeverDrops` |
| Told 29 days ago; told, but absent for 10 days only | Refused: both 30 days are needed | hub `TestNotice_AgreeEnded_NoNoticeNoDeletion` |
| A box of a household with no address names presses without end | The operator gets at most 5 in-place notices a day for that household, the same app once in 7 days; the rest wait and are not told | hub `TestNotice_NoAddress_ABoxCannotFloodTheOperator` |
| A box changes the press's key in every report | The notice is reset each time: the household gets at most 5 mails a day, and nothing is ever agreed | hub `TestNotice_Record_WhatVoidsIt`, `TestNotice_TheDailyLimitDelaysAndSurvivesARestart` |
| The box is off at the date, or the store cannot be reached | The next window that runs does it: the record does not age | `TestD72_EndRecord_Due` |
| The off-site tier is switched off at the date | No run, no window, no call to the store; the record stays | `TestD72_TierOffAtTheDate_TheRecordWaits` |
| The household is deleted or RESET before the date | Nothing to do: the account goes with all copies; the hub's app record goes with the household | hub `TestD72_AppRecordGoesWithTheHousehold` |
@@ -1938,6 +2013,9 @@ notification address gets no mail: the page is all it has.** A blocked household
see sizes, and the box does not send them.
- *The receipt does not read the household's group switches.* D84's reason is a household whose login someone else
uses; that person could switch a group off first. **The operator may reverse this.**
- *The mail leaves at the first report after the press, not at the press* (the notice lock): up to one report
interval later (15 minutes). Its day is the hub's (the mail's day + 30), so it can be one day later than the day
the page shows, and later still when the mail waited.
**A limit, said plainly (rewritten after the review of 2026-11-15).** The hub cannot read a snapshot's tags (the
repository is encrypted; the password never leaves the box — decision 68), and its record of which apps a box runs is
@@ -1957,11 +2035,27 @@ clock, and then name snapshots of other apps under it. What bounds that now, all
younger than 30 days can go this way.
- A box that removes more than was agreed raises the drop alarm at the close and the vanished-files alarm at 07:10.
**What is still open:** a box that stops listing a living app is NOT told at the start (it looks exactly like an
honest removal); the hub hears of it only in the window, and then only if the request is for more than half of the
store. Under that size, such a box can remove up to the cap's size per week beside the cap, without a mail — twice
what it could remove silently before this feature (R-822's accepted limit, `09` §3 decision 166). And one window can
still take the cap AND the cap's size again; after a large request that is a week after the operator was mailed.
**What a taken-over box can still do, after the notice lock (rewritten 2026-11-15 evening).** In plain words:
- **It cannot have anything removed in silence.** To have an app's copies agreed it must name the app in a report —
and the hub then mails the household at once: the app's name, the day, „if you did not ask for this, take it back
and change your password”. This holds for an invented app and for a living app the box stops listing. A box that
names the app only in the window request gets nothing and the operator is told.
- **Size and time, if the household does not act on the mail:** nothing for 30 days after the mail. Then at most the
size of the window's cap per weekly window, beside the cap — the cap is half of the store's snapshots (at least
5). A request for more than half of the store waits one more week after a mail to the operator: 37 days. The
files it names must be older than the day the hub saw the app leave; the hub cannot check that they are that
app's copies (it cannot read tags).
- **The take-back goes through the box.** A household that reads the mail and presses „keep it after all” is safe
when its login was used by somebody else (the mail's case, D84). When the box itself is taken over, the box can
show „kept” and go on naming the app. The mail's last line then does not help; the operator does (the mail's
sign-off), by switching the weekly windows off for that household. **Open.**
- **A household with no address is not told; the operator is.** What happens next depends on the operator.
- **The address is the one on record at the mail.** A person who holds the login and changes the notification
address first gets the mail himself. Whether the old address is told of such a change was not checked in this
build. **Open.**
- **Noise, not loss:** a box can make the hub mail its household 5 notices a day (by changing the press's key), and
the operator 5 in-place notices and 5 „stuck” lines a day per household. Nothing is agreed while it does.
**A narrowing of D84 — decided by CC unattended on 2026-11-15; the operator may reverse it.** *Question:* may one
window remove every snapshot the hub agreed to for removed apps, however many? *Option A (built):* no — at most the
@@ -1973,11 +2067,16 @@ list for 30 days can have the whole store removed in one window, and the first m
the ruling's own reason for the allowance was that the hub's 30-day check does the cap's job; the review showed the
30-day check rests on the box's own reports, so it cannot do that job alone.
**The texts, as built — NOT published; the operator reads them before a release.** The Hungarian date carries its own
final dot (`2026. 12. 15.`), so the Hungarian sentences have no dot after it.
**The texts, as built — NOT published; the operator reads them before a release.** **One date form in every place
of this feature** (decision 316, built): Hungarian `2026. 12. 15.`, English `2026-12-15` — the controller's
`i18n.FormatDate` on the page and in the dialog, the same form made by the hub (`notify.receiptDate`) in the mail's
subject and body. The written-month form („2026. december 15.”) stood only in the first build's note; no code makes
it. The Hungarian date carries its own final dot, so the Hungarian sentences have no dot after it. Rows marked
**(316)** changed on 2026-11-15 evening.
| Where | Hungarian | English |
|---|---|---|
| The remove dialog, under „Mentési adatok törlése”, on a box that shows NO date (no hub off-site tier) — the older sentence, unchanged; **(316)** it is no longer shown where the next row shows | „Az éjszakai restic pillanatképek nem törölhetők egyenként — a megőrzési szabályok szerint automatikusan elavulnak.” | "The nightly restic snapshots cannot be deleted one by one — they expire automatically under the retention rules." |
| The remove dialog, under „Mentési adatok törlése” | „A helyi mentések most törlődnek. A távoli másolatok még 30 napig megmaradnak. Törlésük napja: {dátum} Addig a Mentések oldalon visszavonhatod.” | "The local backups are deleted now. The off-site copies stay for 30 more days. They are deleted on: {date}. Until then you can take this back on the Backups page." |
| The off-site list, the app's line (also the notice after „delete in 30 days”) | „Törlésre kijelölve. Törlés napja: {dátum}” | "Marked for deletion. Deleted on: {date}." |
| The button | „Mégis megtartom” | "Keep it after all" |
@@ -1987,18 +2086,25 @@ final dot (`2026. 12. 15.`), so the Hungarian sentences have no dot after it.
| A removed app with no record | „Ennek az alkalmazásnak a távoli másolatai megmaradtak.” + button „Törlés 30 nap múlva” | "The off-site copies of this app were kept." + "Delete in 30 days" |
| A press that is refused (the app is installed; no hub tier) | „Ennek az alkalmazásnak a távoli másolatait most nem lehet törlésre kijelölni.” | "The off-site copies of this app cannot be marked for deletion now." |
| A record that could not be saved | „Nem sikerült elmenteni. Próbáld újra.” | "This could not be saved. Try again." |
| **The mail — subject** | „[Felhom] Információ: {app}: a távoli másolatok törlésre kijelölve” | "[Felhom] Information: {app}: the off-site copies are marked for deletion" |
| **The mail — body** (the HUB's own sentence, `mail.receipt.offsite_end_scheduled`, inside the hub's ordinary household mail frame: greeting, server, time, level, type, sign-off; the box's event carries the same sentence for the event list only) | „{app}: az alkalmazás távoli másolatai törlésre vannak kijelölve. Törlésük napja: {dátum} Addig a Mentések oldalon a „Mégis megtartom” gombbal visszavonhatod.” | "{app}: the off-site copies of this app are marked for deletion. They are deleted on: {date}. Until then you can take this back on the Backups page with the “Keep it after all” button." |
| **The mail — subject (316)** (`mail.receipt.offsite_end_scheduled.subject`; the receipt has its own subject, the frame's „[Felhom] Információ: …” is not used for it) | „[Felhom] {app}: a távoli másolatok törlésének napja: {dátum}” | "[Felhom] {app}: the off-site copies will be deleted on {date}" |
| **The mail — headline** (the frame's first line, `mail.event.offsite_end_scheduled`; unchanged) | „{app}: a távoli másolatok törlésre kijelölve” | "{app}: the off-site copies are marked for deletion" |
| **The mail — the message line** (the HUB's own sentence, `mail.receipt.offsite_end_scheduled`, inside the hub's ordinary household mail frame: greeting, server, time, level, type, sign-off; unchanged. The date is the hub's: the mail's day + 30) | „{app}: az alkalmazás távoli másolatai törlésre vannak kijelölve. Törlésük napja: {dátum} Addig a Mentések oldalon a „Mégis megtartom” gombbal visszavonhatod.” | "{app}: the off-site copies of this app are marked for deletion. They are deleted on: {date}. Until then you can take this back on the Backups page with the “Keep it after all” button." |
| **The mail — after the message line, a paragraph of its own (316)** (`….not_you`) | „Ha ezt nem te kérted, vond vissza a Mentések oldalon, és változtasd meg a jelszavadat.” | "If you did not ask for this, take it back on the Backups page and change your password." |
| The event list on the box (the box's own sentence, `event.offsite_end_scheduled`; unchanged, never mailed) | „{app}: az alkalmazás távoli másolatai törlésre vannak kijelölve. Törlésük napja: {dátum} Addig a Mentések oldalon a „Mégis megtartom” gombbal visszavonhatod.” | "{app}: the off-site copies of this app are marked for deletion. They are deleted on: {date}. Until then you can take this back on the Backups page with the “Keep it after all” button." |
No mail to the household at the deletion: it can do nothing with it (`08` §8.1). The operator's records are
`offsite_ended_app_removed` (info, per app, not mailed) and `offsite_ended_app_refused` (warning) — `08` §6.5.
`offsite_ended_app_removed` (info, per app, not mailed), `offsite_ended_app_refused` (warning),
`offsite_end_notice_operator` and `offsite_end_notice_stuck` (warnings, mailed) — `08` §6.5.
**Not measured, and what a box must show before a release:** (1) one whole run on a test box — press, record,
30 days by BOTH clocks (or a hub record back-dated by hand on a test household), window, the files gone from the
provider, the 07:10 audit silent; (2) how long `prune` takes after a whole app's history goes (the largest removal a
window has made is 209 snapshots, refused), and that restic 0.19.1 frees the space in the same window, inside the
window's 20 minutes; (3) the mail in a real inbox, in both languages; (4) the hub's listing times against the
provider's clock for a file older than six months (the listing then shows a date with no time).
provider's clock for a file older than six months (the listing then shows a date with no time); (5) the notice lock
on a box: a press, the mail in the inbox within one report, a window before day 30 refused with the reason, a
take-back and a second press mailed again; and the remove dialog in a browser on a box with and without the hub's
off-site tier (the suite reads the script's two branches; no browser has run it).
### 8.2c Decisions in code that no document recorded [DESIGN — confirmed by the operator 2026-10-25 (D44 = all picks, R-1265)]
+11 -6
View File
@@ -470,16 +470,20 @@ own records. What that adds here:
| Event | Severity | Who | When |
|---|---|---|---|
| `offsite_end_scheduled` (from the box) | info | **the household** — a receipt: one mail per press to its own notification address, whatever its group switches say; never the operator (`notify.householdReceiptEvents`) | the household pressed „delete backup data” at a removal, or „delete in 30 days” on the restore page |
| `offsite_end_scheduled` (from the box) | info | recorded, **mailed to nobody** (the notice lock, `09` §3 decision 315): the operator's record of the press | the household pressed „delete backup data” at a removal, or „delete in 30 days” on the restore page |
| the household's notice (hub; no event — logged in the sent log as `offsite_end_scheduled`) | — | **the household** — one mail per press to its own notification address, whatever its group switches say; never the operator (`notify.SendEndNotice`). Sent by the hub from the box's REPORT, at every report until the mail service accepts it; at most 5 per household in 24 hours (the rest go later) | a report first names the app in a pending end record. The hub agrees to the deletion no earlier than 30 days after this mail |
| `offsite_end_notice_operator` (hub; decision 315) | warning, **mailed** (sent at once, the answer of the mail service is awaited) | operator-only, once per app; at most 5 per household in 24 hours, the same app once in 7 days | the household has no notification address, or is blocked: this mail is the notice in its place and starts the 30 days. An address added later: the household is mailed and the 30 days start again |
| `offsite_end_notice_stuck` (hub; decision 315) | warning, **mailed** | operator-only, once per app (details `subject`); at most 5 per household in 24 hours; not while the mail hold is on | a notice has told nobody for 3 days (the mail hold, a failing send, the daily limit). Nothing of that app is removed meanwhile |
| `offsite_ended_app_removed` (hub) | info | operator-only, recorded, not mailed | a window removed an app's agreed snapshots — one per app, with the count |
| `offsite_end_claim_unseen` (hub; review of 2026-11-15) | warning, **mailed** | operator-only, once per household and app, at most 8 per household | a box first names, in a pending end record, an app the hub never saw in its app list — 30 days before the hub could agree to anything. The text names the household, the app and the day |
| `offsite_ended_large_agreement` (hub; review of 2026-11-15) | warning, **mailed** | operator-only, once per app set | a window was asked to remove more than half of the store's snapshots for removed apps. That window removes none of them; a window a week later does, at most the cap's size |
| `offsite_ended_app_refused` (hub) | warning | operator-only, once per app (details `subject`) | a box asked for an app the hub's records clearly do not show ended: the app is in its app list, or absent for less than 29 days. A box that asks less than a day early is refused quietly |
| `offsite_ended_app_refused` (hub) | warning | operator-only, once per app (details `subject`) | a box asked for an app the hub's records clearly do not show ended: the app is in its app list, or absent for less than 29 days — or (decision 315) no report of the box ever named it in a pending end record. A box that asks less than a day early, or whose notice is not 30 days old yet, is refused quietly |
The count detector above and the 07:10 listing check (`offsite_snapshots_vanished`) both know the allowance: a window
explains its cap's part as before, and — by name — the ids the hub agreed to for ended apps. A file that vanished
for any other reason raises exactly as before. The household gets NO mail at the deletion (§8.1: it can act on none).
A household with no notification address gets no receipt: the page is all it has.
A household with no notification address gets no notice of its own: the operator is told in its place
(`offsite_end_notice_operator`). **No notice, no deletion** — the rule and its wrong cases are in `07` §8.2f.
## 7. The intent test [DESIGN, R-386 — CLOSED controller v0.223.0]
@@ -567,9 +571,10 @@ app-mail settings).*
| **2. „Az alkalmazásaim"** — your apps | `app_stopped_unhealthy`, `app_start_failed`, `app_update_held`, `app_update_undone` | mostly us; the household may want to know an app is down |
| **3. „Amit mi intézünk"** — what we handle | `backup_failed`, `db_dump_failed`, `backup_integrity_failed`, `crossdrive_failed`, `expected_backup_missed`, `expected_dbdump_missed`, `node_down`, `health_critical`, `health_recovered` | us (the operator always gets these, with the detail) |
**One mail outside the groups (D84, `09` §3 decision 309)** `[FACT — built, not released]`: the receipt
`offsite_end_scheduled` (§6.5) is not a detector and belongs to no group — it confirms the household's own press, and
a switch must not be able to silence it. It is the only such type; a second one needs its own ruling.
**One mail outside the groups (D84, `09` §3 decision 309)** `[FACT — built, not released]`: the notice about
„delete backup data” (§6.5; sent by the hub itself since decision 315, logged as `offsite_end_scheduled`) is not a
detector and belongs to no group — it tells the household about its own press, and a switch must not be able to
silence it. It is the only such type; a second one needs its own ruling.
**Options:**
@@ -0,0 +1,19 @@
== MUTATION: the older sentence is shown on every box again (beside the new line)
RED on the mutant, GREEN on the real file
--- FAIL: TestD72_TheRemoveDialogShowsTheOlderSentenceOnlyWithoutTheDate (0.12s)
offsite_ends_test.go:326: [hu] the older sentence is on the page 2 times, want once (inside the no-date branch)
offsite_ends_test.go:326: [en] the older sentence is on the page 2 times, want once (inside the no-date branch)
== MUTATION: the older sentence is shown on no box
RED on the mutant, GREEN on the real file
--- FAIL: TestD72_TheRemoveDialogShowsTheOlderSentenceOnlyWithoutTheDate (0.12s)
offsite_ends_test.go:322: [en] WITHOUT a date the dialog must show the older sentence and not the new line:
offsite_ends_test.go:326: [en] the older sentence is on the page 0 times, want once (inside the no-date branch)
offsite_ends_test.go:322: [hu] WITHOUT a date the dialog must show the older sentence and not the new line:
== MUTATION: the two branches are swapped
RED on the mutant, GREEN on the real file
--- FAIL: TestD72_TheRemoveDialogShowsTheOlderSentenceOnlyWithoutTheDate (0.16s)
offsite_ends_test.go:319: [hu] WITH a date the dialog must show the new line and not the older sentence:
offsite_ends_test.go:322: [hu] WITHOUT a date the dialog must show the older sentence and not the new line:
offsite_ends_test.go:326: [hu] the older sentence is on the page 2 times, want once (inside the no-date branch)
3 mutations, 0 not a red-proof