## What is wrong Three documents under `docs/plans/` quote the in-flight cap census of `docs/dev/storage-accounting.md`. Two are merged plans. The third is the plans index, `docs/plans/README.md`. All three state how many caps the census holds. One plan also claims that the census tables each cap by constant and `file:line`. Work item 1193 removes every stated count from the census. It also removes the `:NNN` coordinate from every census table row, and keeps the constant name and the file path. Until the merge request for work item 1193 merges, both plans agree with the census. Once it merges, both plans quote wording that the census no longer carries. A reader who opens a plan and then opens the census finds no match, and nothing marks that the source moved. Work item 1193 excludes both plans on purpose. Its merge request stays at the three files that the issue lists. This work item records the residue. ## Where the falsified prose sits Every coordinate below comes from `origin/main` `7307bd94a2b8e65cb2ba0089da6d622e22ab679f`. Re-derive them before you edit, because both files move. ### `docs/plans/2026-09-03-repository-column-writers.md`, at `7307bd94a` | Lines | Short quote | What goes false | | --- | --- | --- | | `626` | "tables each cap by constant and `file:line`" | The census keeps the constant and the path, and drops the coordinate. | | `627` | "It counts **eleven caps** at" | The census states no count. | | `631` to `632` | "At `e590735c4` and on current main the row reads \"Eleven hand-rolled in-flight caps\" (`:1465`)" | The `e590735c4` half stays true. The "current main" half goes false. | | `634` to `637` | "Ten of the eleven bound a detached worker population" (`:1486`), "which sheds without spawning" (`:1487`) | Both quotes and both coordinates leave the census. | | `1617` | "in the Redis partition of the eleven-cap census" | The census carries no count for this phrase to name. | Two more lines in this file state a count, and a different change falsifies each one. Repair them against that change, not against work item 1193. | Lines | Short quote | Which change falsifies it | | --- | --- | --- | | `640` | "The container steps add **no twelfth cap, and no eleventh detach-bound dispatcher**" | `!2362` is Step 3 of this plan and it adds `repositoryStampMaxInFlight`. The sentence is then false about the plan's own step. | | `727` | "a separate semaphore would add a twelfth to a census that exists to shrink" | `!2324` and `!2362` each take the cap total past eleven. | One line was checked and left out. Line `630` reads "Eleven hand-rolled detach-bound-spawn-shed dispatchers" (`:1434`). That sentence pins the figure to the branch's own base, so it stays true after every change named here. ### `docs/plans/2026-09-03-upload-session-concurrency-cap.md`, at `7307bd94a` | Lines | Short quote | What goes false | | --- | --- | --- | | `1178` to `1181` | "covering eleven in-flight caps", "six hold a pooled Postgres connection, four hold a Redis connection, and the eleventh holds none" | The total and the three partition counts all leave the census. | | `1190` | "the eleventh member bounds row memory too" | The ordinal has no source to point at. | | `1196` to `1200` | "it says what it counts: eleven constants", "of which ten \"bound a detached worker population\"" | The census keeps the membership test and drops the count. | | `2166` | "**not** a twelfth row in it" | The ordinal has no source to point at. | | `2171` to `2177` | "eleven constants", "ten of the eleven", "the eleventh, npm's inline build" | Same as `1196` to `1200`, inside the Step 11 file bullet. | | `2179` to `2180` | "Adding a row would contradict the section's opening count" | This is the plainest falsification in either plan. After the change the section has no opening count. | | `2181` | "saying it is deliberately outside the eleven" | The Step 11 instruction names a count that the census drops. | | `2184` to `2186` | "where the eleven bound spawned workers", "census's eleventh member" | Same, inside the passage that warns against an inverted borrow. | This plan names no commit beside any of these figures. `repository-column-writers.md` names `e590735c4` beside one figure, and this file names none. ### `docs/plans/README.md`, at `7307bd94a` The plans index carries the same count in its own row for the upload-session plan. This file is the index rather than a plan, so a repairer reaches it last. | Lines | Short quote | What goes false | | --- | --- | --- | | `79` | "Also distinct from #559's **eleven caps**, whose census partitions them by the resource they draw on and whose **eleventh member** bounds row memory rather than a connection" | The census states no count, so both the total and the ordinal lose their source. | Two other lines of this file match a count search and are **not** in scope. Line `62` counts handler steps and line `71` counts plan steps. Neither counts caps. **Read this line's number again before you edit it.** Merge requests `!2361`, `!2357` and `!2353` each add a row to this index, so line `79` moves before this work item is worked. Find the row by its plan slug, `2026-09-03-upload-session-concurrency-cap`, rather than by its line number. ### A second group: partition counts outside the census passages Two more lines state a partition count of the same cap family. Work item 1193 does not falsify them on its own, because each states a fact about the world rather than a quote of the census. `!2324` and `!2362` are what make them wrong. A repairer can fold them into the same merge request, or leave them for the change that adds the twelfth cap. | File | Lines | Short quote | | --- | --- | --- | | `repository-column-writers.md` | `670` | "ruling that the four accounting-emit call sites keep the choice to detach, bound, spawn and shed" | | `repository-column-writers.md` | `1405` to `1406` | "the sentence naming OCI's copy as the one of the four that also holds a pooled connection" | ### One coordinate list to re-check after the work item 1193 merge request lands `repository-column-writers.md:1403` to `:1408` cites five lines of `docs/dev/storage-accounting.md` by number: `:1465`, `:1495`, `:1501`, `:1530` and `:1723`. The census rewrite touches that region. Read those five coordinates against the census after the merge request for work item 1193 lands, and correct any that moved. ## The command that re-derives the true state ```shell git grep -niE 'eleven|twelve|twelfth|ten of the|file:line|opening count' origin/main -- docs/plans/2026-09-03-repository-column-writers.md docs/plans/2026-09-03-upload-session-concurrency-cap.md ``` At `7307bd94a` this command returns 12 lines in `repository-column-writers.md` and 18 lines in `upload-session-concurrency-cap.md`. Six of the 18 are not about the caps. Lines `73`, `78`, `1122`, `2139`, `2633` and `2727` count plan steps, Status rows, test files or parameters. One of the 12 is line `630`, which stays true for the reason above. ## Work item 1194 quotes this plan by coordinate Work item 1194 cites `docs/plans/2026-09-03-upload-session-concurrency-cap.md` by line number at four places in its description. A repair of the plan moves that text and invalidates all four citations. **Repair the plan and edit work item 1194 in the same sitting.** A repairer who edits only the plan leaves work item 1194 pointing at coordinates that no longer hold. | Work item 1194 description lines | Plan coordinate cited | What work item 1194 carries there | | --- | --- | --- | | `88` to `91` | `:2165` to `:2166` | A block quote: "a statement of where this cap sits relative to the in-flight cap census, **not** a twelfth row in it." | | `93` to `99` | `:2179` to `:2183` | A block quote: "Adding a row would contradict the section's opening count, so Step 11 adds a short paragraph naming the cap, saying it is deliberately outside the eleven, and giving the reason that operates". | | `105` | `:2171` to `:2174` | "It records at `:2171` to `:2174` that the census command therefore does not find it." | | `112` | `:2206` to `:2208` | "The author of Step 11 re-reads the section before writing the paragraph, which the plan asks for at `:2206` to `:2208`." | One sentence in work item 1194 already survives the change. Its line 110 reads "If the count leaves the two documents, that clause loses its subject, and the parks-versus-sheds reason is what remains." The two block quotes are the part that breaks. Work item 1194 says nothing about `docs/plans/2026-09-03-repository-column-writers.md`. That half of the defect has no other record. ## How the repair lands The repair lands as a separate `docs(plans):` merge request. It does not go into the merge request for work item 1193. Work item 1088 and merge request `!2352` set this precedent on 2026-09-04, for stale figures in a merged plan. Two open merge requests move the cap arithmetic under this prose: - `!2324` `chore(lifecycle): bound the purger's counter emit behind a shedding cap` adds `counterEmitMaxInFlight` in `internal/lifecycle/emit_dispatch.go`. - `!2362` `feat(oci): stamp repositories.last_updated_at on the container writes` adds `repositoryStampMaxInFlight` in `internal/format/oci/emit_dispatch.go`. Each one takes the cap set past eleven. The repair is cheapest after both merge requests settle. `repository-column-writers.md:640` and `:727` are claims about that plan's own steps, and their correct wording depends on `!2362`. A repair written before `!2362` settles has to be written a second time. ## Related - https://gitlab.com/gitlab-org/ops/artifact-registry/-/work_items/1194 — quotes the upload-session plan by coordinate, and needs the same sitting. - https://gitlab.com/gitlab-org/ops/artifact-registry/-/work_items/559 — owns the in-flight cap census. - https://gitlab.com/gitlab-org/ops/artifact-registry/-/work_items/1088 — the precedent for this shape, already complete. Related to https://gitlab.com/gitlab-org/ops/artifact-registry/-/work_items/1193 _This is a bot message 🤖 — /smurfit_