Research: garden planting, pet care state, chemistry timers, story phases, summons

Session 2026-09-28. Scope: the ROADMAP section "Deployables, pets, and progression" plus "Summoned companions". Evidence is only what the checked-in fixtures and the repo's own verified data show; the installed game paks were not available and nothing was tested in game. Fixture tests: tests/AbioticEditor.Tests/Features/GardenPlotPlantingTests.cs, PetSavedCareStateTests.cs, ProgressionCensusFixtureTests.cs.

1. Garden plots: planting and clearing

Census (all fixture worlds, DeployedObjectMap, class /Farming/GardenPlot_*)

34 plot entries (some are the same plots seen in two copies of one world).

Class	Entries	State observed
GardenPlot_Large (8 spots)	14	all spots planted
GardenPlot_Medium (4 spots)	7	all spots planted
GardenPlot_SmallRound (1 spot)	5	planted (3 grown, 2 dead)
GardenPlot_Small (1 spot)	8	1 planted (Legacy Cascade, Nyxshade), 7 EMPTY (client world "Chrissie", ItemProxies_ has zero elements)

Only GardenPlot_Small has both an empty and a planted example. Comparing an empty small plot with the planted one, the two entries have the same top-level tag names in the same order and differ only in the actor's own id and transform and in the ItemProxies_ array. So:

• Empty spot = no proxy element. Planted = one ItemProxies_ element with SpotIndex_, ChangeableData_ (an item struct with its own AssetID_, durability 350/350, stack 1, and a DynamicProperties_ array of Portions, GrowthStage, GrowthProgress), ItemRow_ (RowName = Plant_*), CookingData_ (all defaults, CookState Raw) and an empty TimerData_.
• The proxy layout is identical across every class and crop in every fixture (12 crop rows seen).
• AssetID_ is a 32 char uppercase hex GUID, unique per proxy, referenced nowhere else in the file (a raw text scan finds each once), so a fresh GUID is a legal value.

Observed growth states

Only three GrowthStage values ever occur: 4 (Grown, GrowthProgress 0, by far the most common), 5 (one Greyeb spot, progress 7900) and 7 (Dead, 2 spots, no GrowthProgress tag and durability 100 instead of 350). The stage-name list in GardenPlotsFeature also names 0 to 3 (Sprout, Budding, Juvenile, Flowering) and 6 (Regrowing), but no fixture shows any of them, so the values a fresh planting takes are UNVERIFIED. This also means the existing "changing a crop resets the spot to Sprout/0" behaviour rests on the enum's first entry, not on an observed freshly planted spot. Watered and fertilized transitions: water (LiquidLevel_) and per-spot fertilizer (PlayerMadeString_, 0,|,0,...) are stored at plot level; every fixture plot has fertilizer 0, and no fixture shows a fertilized spot, so fertilizer above 0 has no confirmed example either.

What was implemented

GardenPlotPlanting (Core, Services/WorldMapFeatures) plants into or clears the single spot of a GardenPlot_Small, and GardenPlotsFeature exposes it as the crop:0 choice with an (Empty) option. The write is the minimum the fixtures prove:

• Planting clones a game-authored fully grown proxy (stage 4, progress 0) from the same save (the same never-fabricate rule AddDroppedItem follows) and replaces only the spot, the crop row and a new AssetID_ GUID. The result is therefore the exact observed "grown crop" state.
• Clearing removes the element, giving the observed empty shape.
• Test proof (legacy Cascade, which holds the planted small plot and other planted plots to copy from): clear then re-plant restores the identical serialized length, reloads cleanly and reads back as Grown/0; the clear step shrinks the file and reloads as an empty array.
• Planting is refused, with no change to the bytes, when the save has no planted crop to copy the layout from. This is the case for the "Chrissie" world that holds the 7 empty plots: it has no planted proxy of any kind, so nothing can be written there today. Hand-building the struct from scratch is not done because a fabricated struct array fails on reload (see the PetDynamicProperties remarks). A cross-save graft (template taken from another world) was not attempted because the two worlds come from different game versions.

Still unknown, capture needed

• Medium, Large and SmallRound plots: no plot of those classes is empty or partly planted, so whether an unplanted spot is a missing element is unproven. Capture: a save with a medium/large plot where some spots are empty and others planted, plus the same plot before and after planting one spot.
• Fresh states: a plot within seconds of planting (expect stages 0 to 3 and a progress value), one after harvest (stage 5 or 6), and a fertilized spot.
• Whether the game accepts a hand-planted spot on load (never run in game).

2. Pet feeding and mutation

What a saved pet stores

PetSavedCareState (read-only, Core Services/World) lists every dynamic int on a saved pet. Across all fixtures the only keys on pets are XP, TimerState, CurrentAmmo, Generic2, Portions, and on carried pets MutationProgress and PetMutation. There is:

• No saved food identity. No field on a PetNPC entry or a carried pet item names a food row, a last-fed item or a taming food. Food lists exist only in game tables (DT_PetsTamingFood_/Mutations_[].MutationItems_), loaded at runtime by PetCareCatalog.
• No named cooldown or progress field for feeding. The candidates are two opaque ints:
  • TimerState: on every pet (world pets 270,500 to 448,945; carried 436,894 and 448,843). It is 0 on non-pet items that carry the same key (healing kits, pet beds, nets). Two snapshots of one world (dedicated vs Steam copy, different MinutesPassed) hold identical values, so it is not derived from the world minute counter at save time. Pets stored as items in containers keep one too. Plausibly a clock stamp of the pet's last timed event; unproven.
  • Generic2: only on the 10 Peccary Sows, always greater than the same pet's TimerState (differences 2,248 to 80,163). Meaning unknown.
• MutationProgress is 3 on both carried pets and PetMutation is 1 + the pet's identity index in its owner row's Mutations_ list (Leyak Pest 6, crafted Magma Skink 1); see section B of research-garden-crops-and-pet-mutation.md. World PetNPC entries carry neither key.

Mutation graph

The graph lives only in the pak tables (DT_Pets Mutations_), so it is available at runtime through PetCareCatalog.MutationOptionsFor, not offline here. The one owner list documented from real data is the base pest row: Pest, Volatile, Snow, Magma, Enlightened, Leyak, Carbonated (values 1 to 7), and the crafted skink lineage owns a separate list (Magma at index 0, per that note). Nothing new could be added without the paks.

Decision

No feeding or cooldown editor and no write path: the timers are opaque and there is no food identity to model. Capture needed: the same pet before and after being fed (each valid food, and a wrong food), with the game clock noted, to see which of TimerState/Generic2/MutationProgress change and by how much.

3. Offline chemistry production

A chemistry bench is a generic deployed-object entry: its top-level tags (Class_, ActorPath_, ChangableData_, Transform_, ContainerInventories_, ItemProxies_, ...) are exactly the same 18 as a garden plot's, with no bench-specific tag. The words ProcessingActive and ProcessingTimestamp (the blueprint's runtime properties named in the earlier audit) occur in none of the Facility region saves, so no processing state is serialized. There are three bench entries in the fixtures (two distinct benches) and every one of their 4 slots (3 inputs, 1 output) is Empty. The only saved flasks are 4 flask_pheromone stacks, in ordinary container and player inventories, with LiquidLevel_ 0: a flask's content is its item row, not a liquid level. Consequences: batch start, remaining time and elapsed time are runtime-only; no elapsed time is fabricated and none should be. What could be captured to go further: a bench mid-mix and a bench with a finished output flask (to see whether the game moves items between slots on its own).

4. Character story phases

NarrativeNPCMap.NarrativeState_ is the only saved phase. Joining all 534 non-pet entries in all fixture worlds to the bundled NarrativeNpcNames registry (verified per-actor character names, 509 of 534 resolve, the 25 rest are unnamed slots):

• Phases seen: 3 (514 alive plus 9 dead), 2 (7, all dead), 0 (4, all dead). Phases 1, 4 and 5 exist in the enum but never occur in a save.
• Per character: Ela (Abe's Electro-Pest) is the only phase 0 (dead). Abe and Dr. Jager reach phase 2 (dead) in the older worlds; Dr. Jager is phase 3 dead in the "Chrissie" world. Janet and the Unknown Militant are phase 3 dead. Every other named character is phase 3 alive.
• The enum names are compiler artifacts (NewEnumeratorN) and no game table maps a phase to a description, so no per-character stage names are added: the census supports only "3 is the settled state, 0 and 2 occur together with removal". Adding labels would be a guess.

5. Summoned companions (armor-set summons)

PetCatalog already curates two summon classes, NPC_Exor_Ally (Exor Spirit) and NPC_MageEye_Ally (Mystagogue Drone), as shown-but-not-editable. In the fixtures:

• Every PetNPC entry across all worlds is one of NPC_Peccary_Sow, NPC_Monster_Pest_Electro or NPC_Skink_Crafted; no summon class is ever in the pet map.
• The strings Exor_Ally, MageEye_Ally and Summon do not occur in any Facility region save (Destructible_Fracture_MageEye_* hits are unrelated map props). No player save lists a summon.
• Conclusion: the fixtures give no evidence that armor-set summons are persisted at all. Because saved pets are keyed by GUID in PetNPC and gated by the game's tamed-pet marker, the likeliest lifecycle is a runtime spawn that ends with the wearer's session or set (no owner field is saved for the regular pets either), but that is inference, not observation. Capture needed: a world saved while a summon is active (and while one is despawning), plus a save reloaded after logging out with the set equipped.