rivencraft/docs/MILESTONE_6_REGRESSION.md

Milestone 6 instability regression and migration check (#73)

This is the release-slice acceptance record for instability, not a promise of general old-save compatibility. The latter remains issue #60. Run the focused archive with ./gradlew build runGameTestServer -ParchivedGameTestSuites=instability; it includes the default smoke suite. For 0.5.8.10 on 2026-10-10, that dedicated GameTest server passed all 55 required tests with no optional content/client integrations installed (the required Dynamic Dimensions and Bad Packets dependencies were present). The full archive was not run; it has a previously documented storage stall, so do not count it as passing. The 0.5.8.11 run with live block profiling stalled in its initial batch and was stopped; it is not a passing result. The live profiling scenario is handed to the player in the block-profile checklist.

Automated coverage

Requirement	Existing default test or focused archive
Stable writing, defective writing, score explanations, repeated use of the same book	AgePersistenceGameTests.instabilitySeparatesStableAndDefectiveWriting
Frozen score, age-scoped writing cost, random baseline	AgePersistenceGameTests.randomBaselineAndBrightBalanceAreFrozen, auditedSymbolCostsAreCountedAndAgeScoped
Complete Age definition, score and provider-plan save/load	AgePersistenceGameTests.authoredPlanRetainsResolvedMeaningAndAgeIdentity; AgeInstabilityRuntimeGameTests.selectionIsDeterministicAndFrozen
Old pre-scoring Ages, score-only saves, removed symbols, unavailable handlers	AgePersistenceGameTests.missingRegistryEntryDoesNotReinterpretAnOldAge, instabilitySeparatesStableAndDefectiveWriting; AgeInstabilityRuntimeGameTests.savedPlanSurvivesAgeUpdatesAndMissingHandlers
Per-Age controls and changed config without score/plan reroll	AgeInstabilityRuntimeGameTests.perAgeControlsPersistWithoutChangingScoresOrOtherAges; Milestone6RegressionGameTests.disablingEffectsCannotRerollFrozenPlan
Hazard cooldown and attempt sequence across saved-state reload	AgeInstabilityRuntimeGameTests.hazardCooldownAndAttemptSequenceSurviveReload
Active-Age rotation, action/tick budget	AgeInstabilityRuntimeGameTests.ageScopeRotationAndActionBudgetAreBounded
Multiplayer loaded-chunk-only candidate selection, deduplication and cap	Milestone6RegressionGameTests.multiplayerCandidatesStayLoadedDeduplicatedAndCapped
Decay loaded-chunk target and spawn protection	AgeDecayGameTests.targetMustRemainInChosenLoadedChunk, savedSpawnAndPlatformAreProtected
Explosion policy gates and protected blocks	Milestone6RegressionGameTests.disabledGriefAndProtectedBlocksStayUntouched
No Overworld status or decay provider	AgeStatusEffectGameTests.overworldCannotRunAnAgeStatusProvider; AgeDecayGameTests.overworldCannotRunDecayProvider
Optional-mod material weights, block overrides, and old profile migration	Milestone6RegressionGameTests.commonModdedMaterialWeightsStayBounded, blockProfileSamplesPersistWithoutDoubleCounting
Ordered celestial modifier consumption, repeated modifiers, and dangling +100 per kind	WritingGrammarGameTests.orderedGrammarAcceptsPartialWritingAndDiagnosesInvalidSequences; EnvironmentWritingGameTests.celestialModifiersFollowWrittenOrder

These tests exercise saved-data serialization and deterministic policies. They do not simulate a complete process restart, an actual multiplayer server, optional-mod installation, or real terrain and hazard visuals. The three new checks live only in the instability archive to avoid duplicating default tests.

Migration outcome and known gaps

• Pre-scoring Milestone 5 Ages load with a neutral score and no effects; they are not retroactively rescored under current writing rules.
• Score-only Ages receive a deterministic provider plan once from their saved score and Age seed. An already-saved score and plan retain their meaning when pages, handlers, or runtime config change. Unknown but syntactically valid provider IDs are preserved and skipped if their handler is unavailable.
• Missing symbol definitions do not reinterpret a frozen instability score. That check is narrower than arbitrary missing biome, block, or worldgen registry entries: loading a world with removed data packs is not guaranteed. Broad save compatibility and migration remain tracked in #60.
• Version-3 plans keep their previous stage-sorted celestial interpretation and frozen writing cost. Only new version-4 plans consume period/phase modifiers in authored order and charge a trailing modifier once per pending kind.
• In-process roundtrips cannot prove that Age dimensions reload correctly after a full stop/start. Check that in a real client/server world. Do not add another GameTest for it if it hits the documented storage stall; hand the exact scenario to the player instead.
• The dedicated GameTest run establishes classloading with no optional runtime flags. It does not establish classloading with every optional integration installed, nor the reverse of removing one from an existing save.

Manual in-game matrix

Use a fresh test world, then keep one save for the restart and data-pack checks. Record the seed, mod list, age name, and /rivencraft age instability output before and after each change. Do not deliberately remove a required registry entry from a valued world; copy the test save first.

1. Bind a low-instability book and a deliberately unstable book. Reuse each book and verify the Age name, dimension ID, score, selected providers and provider levels remain unchanged. Exit the Ages, stop Minecraft completely, restart and re-enter; compare the same diagnostic values.
2. With two survival players in one unstable Age and a third player in the Overworld, verify effects stay in the Age, follow the Age players, and do not appear in the Overworld. Move both Age players away, let old chunks unload, and confirm hazards do not keep remote chunks loaded or operate there. Confirm controls on one Age do not alter a second Age.
3. Disable global instability, then world mutation, then one provider. Verify the suppressed effect stops and the frozen score/plan do not change after re-enabling or restarting. With block destruction disabled, trigger an explosion in an expendable Age; check that arrival platforms, portals, block entities and protected blocks survive. Check lightning, explosion, decay, status-effect, burning, meteor and erosion visuals where selected.
4. With Biomes O' Plenty or another TerraBlender biome mod, confirm recognized biome pages still author an Age and the instability diagnostics remain stable after restart. Repeat with Serene Seasons installed, then absent in a disposable copy; confirm its Age season behavior and that Rivencraft itself still loads. JEI should display the relevant pages and recipes when installed, while the mod works without it. Separately, install a mod with common-tagged ores/storage blocks or source fluids. Compare a material-rich Age with an ordinary Age after the profile thresholds are reached; the installed mod's recognized blocks should affect the result without a Rivencraft allowlist. Then remove that optional mod in a disposable copy and confirm Rivencraft still starts. Test a single block_weight_overrides entry and a zero-weight exclusion. This remains a player-run scenario, not an automated pass.
5. Check the book preview separately with its fallback animation, Ponder/Create, and Immersive Portals when available. Opening a book should not change player location, instability diagnostics, or ambient audio in the actual Age. Do not install both season mods together for this matrix.
6. On a copied save, remove or alter a nonessential data pack providing a writing symbol or optional handler, then reload. Existing saved scores and provider plans must not reroll. If a referenced worldgen registry entry prevents load, record it against issue #60 rather than treating #73's frozen-instability guarantee as general missing-registry support.

The matrix is guidance for a player-run pass; none of these six scenarios is claimed as completed by the automated result above. The player reported an in-game pass for the 0.5.8.18 candidate on 2026-10-12. That is a release-slice acceptance report, not a claim that every expanded matrix row below was individually exercised.

Regression run record

For each row, record pass/fail, world seed, Age name, Rivencraft version, optional-mod versions, and a short observation. "Not run" is intentional until a player completes the in-game check; a successful build is not a pass.

Area	Specific manual observation	Status
New vs reused book	Dimension ID, score, and providers do not reroll	Not run
Full process restart	Re-enter two Ages; saved traits, score, plan, and cooldowns persist	Not run
Multiplayer isolation	Two Age players receive effects; Overworld player does not	Not run
Loaded-chunk scope	Departing players permit unloading; hazards do not reach old chunks	Not run
Controls	Global, per-Age, mutation, and provider toggles suppress effects without reroll	Not run
Protected blocks	Explosion/decay/erosion spare arrival platforms, portals, and block entities	Not run
Hazard visuals	Lightning, explosions, decay, status, burning, meteor, and erosion each behave when selected	Not run
Third-party biomes	TerraBlender biome pages and cave classification survive Age creation/restart	Not run
Optional integrations absent	Mod launches without JEI, Create/Ponder, Immersive Portals, Serene Seasons, or biome mods	Not run
Optional integrations present	JEI pages/recipes, preview backends, seasonal behavior, and ambient sound work	Not run
Optional-mod materials	Tagged ore/storage/source-fluid samples change live profile automatically	Not run
Material overrides	Positive and zero block_weight_overrides values affect only named blocks	Not run
Removed optional mod	Disposable copy launches without the mod; unknown sampled IDs stay serialized	Not run
Removed symbol/handler	Frozen score and provider plan survive when nonessential data is removed	Not run
Ordered modifiers	Five book-order cases below produce expected time and dangling costs	Not run

Focused ordered-writing and optional-material checks for 0.5.8.18

Use /rivencraft age instability in each newly bound Age; old Ages retain their original plan and score. Record each row as pass/fail with the Age name, world seed, installed optional mods, and the command output. The ordered modifier assertions are archived, not a claim that the stalled GameTest suite ran.

Book pages after Link Panel, in exact order	Expected result	Result
Half Length, Double Length, Normal Time	Length combines to 1.25×, no dangling cost	Not run in game
Normal Time, Zero Length	Sun keeps normal motion; one dangling +100	Not run in game
Rising, Normal Time, Setting	Rising start phase; Setting dangles +100	Not run in game
Half Length, Double Length, no celestial page	One combined pending length dangles +100; seeded clock remains	Not run in game
Normal Time, Zero Length, Normal Time	Last sun consumes Zero Length; duplicate celestial cost remains separate	Not run in game

With a common-tagged ore/storage-block or source-fluid mod installed, bind comparable material-rich and ordinary Ages and wait for profiling thresholds. Verify the modded material contributes automatically, then restart and confirm the score/profile remains consistent. On a disposable copy, remove the mod and confirm Rivencraft launches without it; restore the mod before judging the world's registry-dependent terrain. Also check a zero-weight override and a positive override. These are optional-mod compatibility checks, not a new required dependency.