Release 3.10.0: last changelog splice and README sync

[erikdarlingdata]

The last 3.10.0 changelog splice, plus the README sync. Part A added the entries and the README text. Part B merged dev (with #5628, #5636 and #5637), re-cited the #5628 entry, and ran the archive split and census last. Branch release/v3.10.0-splice5, based on dev 3954666.

What each commit did

1. Darling's hourly deadlock maintenance reads stored findings and alerts in bounded pages (#5625) #5626 entry (Fixed). The buffer line is title only, so the full-entry text in docs/changelog/3.10.md is written from the merged PR body (600,000-row test table: 35,698 buffers and 202 ms before, 2,048 buffers and 22 to 43 ms after). Index line and link definition in both files.
2. The pre-lock check trusts the install's own service SID and an existing install gets exact steps (#5627) #5628 entry (Fixed). Text comes from the buffer. Part A cited the issues [QUESTION] Is there changes to installation for Darling headless since version 3.7.1? #5627 and The compose quickstart pulls the nightly image on a release #5629. Part B re-cited it as [The pre-lock check trusts the install's own service SID and an existing install gets exact steps (#5627) #5628] (see Part B).
3. Date: 3.10.0 is dated 2026-10-09 in both files (was 2026-10-08).
4. README sync (checklist step 2b): see below.
5. Important note: it said migrations V158 through V170, but dev is at schema 173. V171 and V172 (Query Store interval tables are partitioned by day #5573) and V173 (A one-day Custom Views panel on the Query Store wide table reads 8.5 M rows and takes 48 s on a large store #5582) are in PgMigrations.cs. The title (in both files) now says V158 through V173, and the note in docs/changelog/3.10.md describes the three. The facts come from the Query Store interval tables are partitioned by day #5573 entry and the migration comments: V171 and V172 attach the interval tables as legacy partitions without a scan (1.1 s and 20 ms on a large store), and V173 creates its tables empty.
6. Second Important entry: adds a 3.10.0 Important entry: on Windows, upgrading an install that ordinary users can write to stops once and prints steps to move it ([[QUESTION] Is there changes to installation for Darling headless since version 3.7.1? #5627]).

Files are CRLF in the tree; the byte census after every write showed CRLF count equal to LF count and no bare CR.

Duplicate section headers (brief step 4)

None found, so nothing was consolidated. [Unreleased] has no section headers. 3.10.0 in CHANGELOG.md has Important, Added, Changed and Fixed once each, and so does docs/changelog/3.10.md. Entry-line multiset of the 3.10.0 block (count, SHA-256 of the sorted non-header lines), unchanged by the check:

• CHANGELOG.md 3.10.0 after commit 3: 293 lines, 03ad6c90adad7b3f17e25b8879fd9543e6959b12a6eb62f9a46881e404e84cad
• docs/changelog/3.10.md after commit 3: 609 lines (entries plus link definitions), 9206139f11583a8078da9328d0444d395db2ea20f2bd3c5afed9edd2a721518a

Reference links (brief step 5)

Nothing to fix. 297 distinct [#N] are used in the 3.10.0 index lines and 297 in docs/changelog/3.10.md. Every one has a definition, there are no duplicate definitions, and 3.10.md has no orphan definitions. All 3.10 URLs are /pull/N. Outside this lane: CHANGELOG.md has 5 definitions no line uses (#85, #86, #4325, #4400, #4401), all in older versions. I left them.

README sync (step 2b)

README.md already covered everything up to #5491 (yesterday's sync). I checked the 3.10.0 entries merged after it and added three user-facing items:

• The shared time range picker (Time range picker: more presets, calendar periods and typed ranges #5570) and the text column filters (Column filters list a column's values to tick, like Excel, and stay set across refreshes and restarts #5577), as two sentences in the Tabs section.
• The DuckDB memory limit setting (Lite's Job History reads only the selected time range, and tab loads read off the UI thread (#5457) #5553), as one bullet under Lite Data Storage.

Checked and unchanged: Lite collector table (42 rows, no collector was added in 3.10.0), tab lists, edition comparison, managed platform support, permissions section, deprecation callout. The plain-English checker on the added text gave one hit, a pre-existing long sentence that I made shorter by 3 words.

Release checks (reported, nothing changed)

• Step 4a. The latest dev push build run at 3954666 was green. No job builds PerformanceMonitor.sln. The build job builds Lite.Tests, Darling.Tests and the Darling Viewer in Release: each shows 0 Warning(s) and 0 Error(s). The Linux job publishes the service (no counts printed). The deprecated Dashboard and Installer builds do not run on a dev push.
• Step 4b. DARLING_RELEASE_CUT=1 ... -method "*StatementCensus_PendingEntries_BlockARelease*": Total 1, Failed 0, prints Statement-column census pending entries: 0.
• ChangelogIndexAndArchiveTests: Total 5, Failed 1. EveryArchive_StillHoldsTheProseThatLeftTheIndex fails on a digest mismatch, as expected: the archive census (archive-census.txt) has not been regenerated for the new 3.10.0 entries. Part B's census fixes it.
• ProductVersionDeclarationTests: Total 26, Failed 0.
• *Census* (every class, one wildcard run): Total 500, Failed 0, Skipped 1.
• Darling.Tests Debug build: 0 Warning(s), 0 Error(s). No full suite, as briefed.

Release PR body

Draft at handoffs/release/release-3.10.0-pr-body.md. It has a HARDENING: placeholder line for the coordinator.

For the coordinator

• Note the Important-note fix in commit 5: the release PR body says V158 through V173.
• Part B hand-off: handoffs/release/splice5-partB.md.

Part B

Part B started from branch head b2f4b57. Dev already had #5628, #5637 and #5636 (24971b8). Three commits, pushed in order as fast-forwards:

1. Merge of dev (633fd72). Four conflict hunks, all in the 3.10.0 Fixed list and the link-definition list of each changelog file. Both sides kept: this branch's Darling's hourly deadlock maintenance reads stored findings and alerts in bounded pages (#5625) #5626 and [QUESTION] Is there changes to installation for Darling headless since version 3.7.1? #5627 lines, and dev's Darling web: response security headers; clearer consent text; connection-test errors carry less detail; token honored on loopback listeners #5636 and Darling's deadlock maintenance logs a line when a stage recovers after a failed pass (#5634) #5637 lines. archive-census.txt merged without a conflict and step 3 regenerated it.
2. The pre-lock check trusts the install's own service SID and an existing install gets exact steps (#5627) #5628 entry re-cited (4c458f9). ([#5627], [#5629]) became ([#5628]) on the index line and on the full entry. The rest of the text is unchanged and matches the buffer. The [#5628] link definition goes at the end of both link lists. The two [#5629] definitions are gone, because nothing else cites The compose quickstart pulls the nightly image on a release #5629. [#5627] stays for the Important entry.
3. Split and census (9f13e7b). Last commit. It changed docs/changelog/3.10.md (the split put the [#5628] definition in numeric order) and tools/changelog/archive-census.txt (3.10.0 row: 295 entries).

Byte census of both changelog files after each write: all CRLF in the tree, with no bare LF and no bare CR. After the merge that was 5,199 lines in CHANGELOG.md and 633 in 3.10.md.

Entry multiset proof (3.10.0 section, lines starting with - **)

Expected is the branch side plus the dev side minus the shared base (3954666). Merged equals expected in both files.

File	base	branch	dev	expected	merged	SHA-256 of sorted lines (expected = merged)
CHANGELOG.md	290	293	292	295	295	7e7c024f1cb6bd2abc7e7e6d32992d5a879c74ac55de26cc4ccb40e11f46468d
docs/changelog/3.10.md	290	293	292	295	295	6dc0524a03d9ec957127768b96989f1c37efa9431967c4b281b881b16b1ec853

Link check (3.10.0 section)

299 distinct [#N] cited in each file. Every one has a definition. No duplicate definitions. 3.10.md has no orphans. CHANGELOG.md still has the 5 old orphans from older versions (#85, #86, #4325, #4400, #4401), left alone.

Archive script results

• split: wrote 12 files, every file under its ceiling.
• census: wrote tools/changelog/archive-census.txt. The 3.10.0 row is 295 entries.
• verify: exit 0, 94 PASS lines (last PASS line: PASS 3.9.0: 760 entries, prose sha256 3b0001be58413502 matches tools/changelog/archive-census.txt).
• roundtrip: exit 0, PROSE INTACT every version 3653e8d63a08 held is byte-identical in prose, and every link definition still resolves.

Entries per section in docs/changelog/3.10.md after the split

Important 2, Added 109, Changed 42, Fixed 142 (295 in all). The release PR body draft now says 109 Added, 42 Changed and 142 Fixed.

Tests (Darling.Tests Debug build: 0 Warning(s), 0 Error(s))

Class filter	Total	Failed	Skipped
ChangelogIndexAndArchiveTests	5	0	0
ProductVersionDeclarationTests	26	0	0
*Census*	500	0	1
*Ratchet*	15	0	0

The PR diff against dev is four files: CHANGELOG.md, docs/changelog/3.10.md, README.md and archive-census.txt, so it stays on the docs fast path.

For the coordinator

• No store migration came in with Darling web: response security headers; clearer consent text; connection-test errors carry less detail; token honored on loopback listeners #5636 or Darling's deadlock maintenance logs a line when a stage recovers after a failed pass (#5634) #5637, so the Important note's V158 through V173 still holds.
• Not run: the full suite (CI decides) and the solution build.