Add 26.3 modpacks and retire 26.2 from CI

[rubensworks]

Adds Minecraft 26.3 to the pack tests. 26.3 supersedes 26.2, so 26.2 is dropped from CI and from the Renovate rules; its pack pom.xml files stay in the repo for reference. 1.21.1 and 26.1.2 are untouched.

Modpacks

Five new packs, mirroring the 26.2 layout: cyclops-all on neoforge, forge and fabric, plus cyclops-all-deps and cyclops-id on neoforge. Mod versions are the current 26.3 releases:

Mod	26.3
cyclopscore	1.30.0-1162
commoncapabilities	2.11.8-401
flopper	1.1.9-256
structuredcrafting	0.2.14-245
capabilityproxy	2.2.3-244
everlastingabilities	2.5.10-409
energeticsheep	1.2.0-278
colossalchests	1.8.16-321
iconexporter	1.4.2-234
integrateddynamics	1.33.4-2169
integratedtunnels	1.9.6-798
integratedcrafting	1.4.7-803
integratedterminals	1.7.0-1013
integratedscripting	1.0.22-495
integratedrest	1.1.15-220
integratednbt	1.6.1-86
evilcraft	1.2.98-1114

Third-party dependencies in cyclops-all-deps: JEI 26.3-neoforge:31.1.0.13, TerraBlender 26.3-26.3.0.0.0, Jade 8892782 (Jade-mc26.3-NeoForge-26.3.1).

theoneprobe and Curios have no 26.3 build, so both are commented out with a note. theoneprobe's latest is 26.2_neo-15.0.1-2 and Curios' is 16.0.0+26.2. Leaving either at its 26.2 artifact would break the pack at load: that is exactly how Jade's 26.2 build behaved when EvilCraft-Compat was ported, dying with NoSuchMethodError: VanillaRegistries.createLookup() before any test ran.

CI

The mc matrix entry 26.2 becomes 26.3, and the Fabric include entry moves to fabric-api: 0.161.0 with fabric-gametest-api: 4.0.32+3434d6d95d. That pairing is not guesswork: it is the fabric-gametest-api-v1 version that fabric-api:0.161.0+26.3 itself depends on, read from its POM.

Renovate

The per-MC-line allowedVersions rules for TerraBlender, theoneprobe, Architectury and Refined Storage move from modpacks/**/26.2*/** to modpacks/**/26.3*/**. Architectury's major goes 21 → 22, which is the major that declares minecraft [26.3,) in its neoforge.mods.toml.

Since the 26.2 packs are unmaintained, ignorePaths keeps Renovate away from them rather than having it keep opening auto-merging update PRs for packs nothing tests. Note this overrides the ignorePaths inherited from config:recommended, which is harmless here as the repo holds only POMs and workflows.

A pre-existing problem I did not touch

run/world/datapacks/cyclopsmc-gametests-26-1-2/pack.mcmeta declares min_format: 82, max_format: 88. Those are resource pack numbers (26.1.2 is 84, 26.2 is 88), not data pack formats: the data pack formats are 101 for 26.1.2 and 107 for 26.2. So that pack looks incompatible to both versions it is meant to serve, which would mean the two IntegratedCrafting recipes in it have not been loading. CI is green regardless, so either the recipes are no longer needed or incompatible world datapacks still load.

I did not change it, since it is outside this PR's scope and the answer decides whether it is a real bug or dead weight. The new cyclopsmc-gametests-26-3 pack declares 121, which is 26.3's actual data pack format, read from version.json in the 26.3 client jar and cross-checked against misode/mcmeta.

Validation

All five new packs resolve completely with the repo's own ./fetch.sh, against the same settings.xml CI uses:

modpacks/cyclops-all/26.3/neoforge       OK (17 jars)
modpacks/cyclops-all/26.3/forge          OK (8 jars)
modpacks/cyclops-all/26.3/fabric         OK (8 jars)
modpacks/cyclops-all-deps/26.3/neoforge  OK (28 jars)
modpacks/cyclops-id/26.3/neoforge        OK (9 jars)

renovate.json is still valid JSON.

🤖 Generated with Claude Code

https://claude.ai/code/session_01V35MFZ7JSgoLe79J4d1uxf

---

Generated by Claude Code

[rubensworks]

Build (server, cyclops-all-deps, 26.1.2, neoforge) is failing on this PR, and I do not believe it belongs to it.

evilcraft:gametestsbloodinfuser_testbloodinfuseremptyfluidcontainercreativeblooddrop failed at 30, -60, 2071!
Expected Blood infuser tank contents to be 10000: was 0 on tick 202
1 required test(s) failed :(

Two reasons it is not this PR's:

• git diff origin/master...HEAD -- modpacks/cyclops-all-deps/26.1.2 run/ is empty. This PR adds 26.3 packs, swaps 26.2 for 26.3 in the matrix and the Renovate rules, and adds a datapack-removal step gated on matrix.loader != 'neoforge', which is a no-op for this job. Its inputs are identical to master's.
• That exact test is already called out as fragile in this repo, in the cyclops-all-deps POMs themselves: "While we have support, the gametest evilcraft:gametestsbloodinfuser_testbloodinfuseremptyfluidcontainercreativeblooddrop fails for some reason when architectury is present." Architectury is commented out here, so that specific trigger is absent, but the test is evidently environment-sensitive.

No fix exists for me to port, so I am not widening this PR with one. I will re-run the job once when the run finishes and report what comes back. If it fails again on an unchanged 26.1.2 pack, it is a real pre-existing failure on master and deserves its own issue rather than blocking this.

The 26.3 jobs are what this PR is responsible for, and those are covered by the two fixes already pushed.

---

Generated by Claude Code

[rubensworks]

Correction to my previous comment: I was wrong that no fix exists. One does, and it is already in flight here.

CyclopsMC/EvilCraft#1263, "Fix random failures of blood infuser game tests" (19 Sep), rewrote GameTestsBloodInfuser to re-fetch the block entity inside succeedWhen instead of holding a stale reference from before the block was placed. It is on both master-26 and master-26-lts.

The packs were simply pinned below it. I checked the published sources jars directly:

Build	getInfuser(helper) present
evilcraft-26.1.2-neoforge:1.2.105-1085 (pinned)	no
evilcraft-26.1.2-neoforge:1.2.105-1107 (latest)	yes

That also explains why 26.3 was unaffected: it pins 1.2.98-1114, built from master-26 well after the fix.

#88 already bumps evilcraft-26.1.2-neoforge from 1.2.105-1085 to 1.2.105-1107 in both cyclops-all and cyclops-all-deps, and it has rebased onto this PR's merge. So the failure resolves when that automerges, and no separate change is needed. Apologies for the wrong call.

---

Generated by Claude Code