Complete 26.3 parity acceptance and dual-version release workflow

docs/DUAL_VERSION_RELEASE.md

@@ -43,11 +43,62 @@ JARs; use new worlds and remove them regardless of pass/fail. Stop test processe
 If generation stalls, hand the test to the user.
 
 Before closing a port umbrella issue, compare its acceptance criteria and comments
-to actual evidence. Current expanded-catalog two-client create/travel/reconnect/
-restart/stale-ack regression remains in #100. #103 retains dual-line release
-automation/regression completion and #104 retains the full parity gate.
+to actual evidence. The expanded-catalog matrix and manual restart/return handoff completed in #100.
+See [the final parity review](MC263_PARITY_REVIEW.md) for evidence and limits.
 Shared issues need both versions resolved before closure.
 
 For each release, document tested optional versions and no-optional fallbacks,
 match client/server JARs, preserve the independent 1.21.1 support line, update wiki
 and platform descriptions, and keep known bugs linked in the release notes.
+
+## Changed-target validation policy
+
+Only test a Minecraft target when its code/resources changed. Select targets with
+`python3 scripts/changed_targets.py --base <commit>`. Shared assets can select
+both; temporary servers must run sequentially. Earlier gameplay sign-off remains
+valid for unchanged 1.21.1 code.
+
+Repeatable local checks:
+
+- Java 21 / 1.21.1: `./gradlew build --console=plain`.
+- Java 25 / 26.3: `./gradlew -p ports/mc263 build --console=plain`.
+- Smoke one changed target: `python3 scripts/smoke_server.py --target mc1211`
+  or `--target mc263`.
+- Two-client 26.3 acceptance: `python3 scripts/regression_multiplayer_263.py`.
+  Fixture sources remain outside the shipping JAR. If a test stalls, stop it,
+  preserve logs, delete temporary worlds and hand the remaining check to the user.
+  Do not automatically retry.
+
+GitHub and Gitea workflow definitions select changed targets and use
+`max-parallel: 1`. Hosted execution is not yet verified. Repository runner
+inventory alone does not establish availability of instance-wide runners.
+The publisher checks embedded version/Minecraft metadata, Java selection,
+filename identity, absence of test fixtures and required 1.21.1 dependencies.
+
+Apply relevant fixes to both supported lines, with separate implementation and
+validation per target. Keep version-specific fixes on the affected line. Advancing
+the newer Minecraft/NeoForge target requires a compatibility audit and must not
+silently replace 1.21.1 support or widen an existing JAR dependency range.
+
+## Current acceptance evidence (2026-11-14)
+
+26.3.0.3 builds and the user passed book-window mouse-look, right-click return
+and re-entry. Beta branding is removed. The release collects these changes,
+the workflow tools and the final parity review.
+
+The expanded-catalog dedicated matrix ran with two real clients. Both received
+runtime-created Ages and authored environment data, traveled using normal books,
+and passed stale-ack rejection/recovery, occupied-unload refusal, empty-Age
+reload/reseed and retirement checks. After full server restart both reconnected in
+the surviving Age with its saved seed and presentation, without the retired Age.
+The run stalled at the final post-restart return Linking Book step (phase 22).
+It was stopped, temporary worlds deleted, and the last check handed off rather
+than retried. Singleplayer preview sign-off does not replace that dedicated check.
+
+The player completed the manual dedicated-server restart/return check on
+2026-11-14 and explicitly accepted #100. This was one player-operated client,
+complementing the automated two-client evidence; it is not described as a new
+automated two-client pass. The test server was stopped, logs retained under
+build/verification/issue100-manual, and its temporary world removed.
+See [the final parity review](MC263_PARITY_REVIEW.md). Renderer bug #116 remains
+separately tracked. #103/#104 and parent #96 close after release verification.