What every thing a customer can end leaves behind, and who is told. - documentation/audits/endings-list-2026-11-10/: README (one page), TABLE (67 endings x 8 questions, each answer with a mark and a pointer), MATRIX, the search list with controls, the helpers' raw answers, and the ageing drill on scratch guest 9202 (five rounds, put back). - Marks: A 499, D 0, K 15, N 22. 9 endings are not named in the architecture folder. - Register: six new rows, one per class (R-1373 … R-1378); 230 -> 236 open rows. Nine existing rows get a dated pointer line; none is closed, merged or re-ranked. - STATUS.md: four decisions for the operator (D71 … D74). No code change. No release. Machines: scratch guest 9202 only, put back. Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS
The Endings list — 2026-11-10
The second list of the gap review (the first was the Space list). One block for each way in which a thing of the
product stops, goes away or is replaced, with eight questions each. Built from the source of four repositories and from
one drill on scratch guest 9202. No architecture document covers endings as a whole; each ending is named, if at
all, in the document of its own area. Files: TABLE.md (the answers), MATRIX.md (the marks), evidence/, tools/.
Five lines
- Endings found: 74. 67 have a code path and are in the table; 7 seed endings have no code path (listed in the table's head); 20 more are tiny and are named in one line.
- Endings no architecture document names: 9 of 67 (24 more are only mentioned).
- Marks (536 answers):
A499 ·D0 ·K15 ·N22, plus oneNamong the endings with no code path. - The worst
N: the operator unticks the off-site tier of a household, and later resets or deletes that household. The account and the data at the provider stay, we go on paying, and the hub holds no record that they exist (EH06, EH01). - The operator decides the four questions at the end of this page.
The N answers, by class (one register row each)
| Class | Row | What | Members |
|---|---|---|---|
| E1 | R-1373 | A second door that does less than the first | orphan delete (EC03-4) · move of one app (EC13-4) · signed retire of a drive (EA08-4) |
| E2 | R-1374 | Kept with no end named | off-site snapshots of a removed app (EC02-4, EC23-4) · files set beside a whole restore (EC18-4) · exported files (EC27-4) |
| E3 | R-1375 | The outside part is by hand or not done | off-site account after an untick (EH06-4, EH01-4) · second off-site tier switched off (EA21-4) · a tester ends (no code path) |
| E4 | R-1376 | Records of a thing that is gone | hub tables with no delete (EH00-4, EH02-4, EH21-4) · host mounts after uninstall (EA00-4) · a wiped drive's record (EA11-4) · a destroyed guest's token (EA27-4) |
| E5 | R-1377 | Nobody is told | orphan delete (EC03-6) · wipe of a drive (EC15-6) |
| E6 | R-1378 | Half-way with no finish, or no way back nobody chose | wipe: drive unregistered before the format (EC15-4, EC15-7) · orphan delete (EC03-7) · import with no undo copy (EC27-5) |
Kept on purpose (each is an A, with its ruling in the table)
Kept app data, until the household deletes it (EC09) · user data on every remove (decision 67) · the restore copy, 30 days (EC17) · the database conversion copy (EC08) · a set-aside backup unit (EC20) · kept second-drive copies (EC19) · moved-aside off-site history (EC25; no end date, by ruling R-245) · retained key packages (EH09) · an ejected drive's mount unit (EA09) · the tunnel marker (EA29) · Cloudflare objects after a delete (EH05, accepted in closed R-688) · both addresses of a box (EH13) · hold logs (R-651) · the template files of a removed app.
The ageing drill on 9202 (a spike)
| Round | Times | What stayed after the ending |
|---|---|---|
| Install a small app, remove it with data and backups | 10 | Nothing. Only live logs and the controller's own data grew. |
| Remove and keep everything, install again, remove all | 1 | Nothing. The named volume goes on every remove (by design). |
| A drive app: remove all, install, remove keeping data, install "start fresh", remove | 1 | One empty folder for the app on the drive, 4,096 bytes; it does not grow (R-1276). |
| Update, restore, remove with backups | 1 | Nothing: old image, restore copy, both backup units all went. |
| Delete kept data on the kept-data page | 1 | The same empty folder again (R-1276, a second door). |
9202 is back as found (evidence/drill/putback-1.txt, putback-2.txt, snap 9-putback against 0-before). Not
covered by the drill: every agent ending (9202 has no agent door), every hub ending, off-site endings, orphan delete,
a cut install, a whole restore, export and import, and the real "use my kept data" choice (the kept folder was empty).
Decisions for the operator (also in STATUS.md)
- Shall the hub refuse to forget an off-site account that still exists? Yes (CC's pick): the hub keeps the account record after an untick, and RESET and delete take it down; one hub change. No: write the hand step in a runbook; the cost is that it depends on memory. If nothing happens: an account can stay and cost money with no record.
- Shall the off-site snapshots of a removed app have an end? Yes, after a fixed time with a notice (CC's pick; needs your number of days). No: keep for ever by ruling; the cost is paid space that only grows. If nothing: they stay.
- Shall the hub delete every record of a household or box that is deleted? Yes (CC's pick): add the missing tables to the purge. No: keep them as history and say so in the privacy text. If nothing: 13 tables keep rows of gone things.
- Shall CC write the "household leaves" and "tester ends" runbook now? Yes (CC's pick): from the five code doors that exist; one evening, documents only. No: wait for the first leaver; the cost is a first run by memory (R-810).
Method verdict (second list)
- It worked, with a new weak point. Three read-only helpers produced 96 raw endings fast. Their pointers were all
opened by a tool, but only 86 of 536 cells (every
K, everyNand the mainAin questions 4 to 7) were read by eye. The otherAcells are a map, not proof. Several helper "candidates" died when read: the tool check is needed. - The ninth line was useful: 8 endings have a hand step that no runbook holds.
- Is the ageing drill worth repeating? Yes, but not the cheap round. Ten cheap rounds found nothing. The drive round
and the update round found or proved things in one pass each. Repeat it only with drive apps, and on a box with an
agent, or it cannot reach the endings where the
Nanswers are.