Files
aevolution/docs/architecture/DECISION_REGISTER.md
T

35 KiB
Raw Blame History

Decision Register

This is the working record of decisions made while reviewing the architecture decision records (ADRs) and OPEN_QUESTIONS.md. A [Reviewed] tag means the record was reviewed in this process; it does not mean every deferred question inside it is resolved. Pending items remain here so the same ADR need not be reopened to find them.

Authority remains: founding brief → architecture → scoped accepted ADRs → public contracts → tests → implementation. Existing accepted ADRs are recorded as the current baseline, not silently re-ratified as new game rules.

Confirmed decisions from the interactive reassessment

ID Date Reference Maintainer decision Follow-up
D-001 2026-10-08 ADR-0087: session-window delivery A accepted. Workflow commits Open/Extend/Close configuration; operator manually syncs aevolution-dev. Four-hour default/maximum and absolute expiry remain. Commit success is pending application, not live success. Align workflow/status wording and verify application through operator observations. No automatic sync or CI deployment credentials authorized.
D-002 2026-10-08 ADR-0087: session-window clock behavior A accepted. Close at the earlier of absolute UTC expiry and a suspend-aware elapsed deadline captured from remaining time at admission. Clock corrections cannot extend that elapsed allowance; closure is latched. Restart retains original UTC expiry; synchronized hosts remain required. Implement and test both deadlines, suspend/resume, clock jumps and restart. Host clock-health tolerance remains a separate engineering detail; this decision does not certify current runtime behavior.
D-003 2026-10-08 ADR-0087: session hostname B accepted. Retain void.aeternalisgames.com for dev sessions. Maintainer confirms the deprecated HTTP Ingress was removed and no HTTP site remains on that hostname. Preserve the existing session hostname/VIP/certificate contract. Infrastructure removal is maintainer-confirmed; no new cluster verification or deletion was performed.
D-004 2026-10-08 ADR-0097: capacity and recovery qualification A accepted with environment constraint. Qualify steady-state capacity after bounded warm-up and cold-start/restart recovery separately; both must pass on the intended evolution-dev Kubernetes/reference-worker deployment. WSL/developer-host runs cannot qualify these gates. Record deployed image/resources/network/stores/hardware; retain startup evidence; define bounded warm-up and separate reports. Workstation-generated clients may exercise the real cluster. No tests executed or gate closed by this decision.

The inventory below is the earlier document survey. Its Reviewed labels do not certify a completed critical reassessment; the confirmed decisions above track the current one-question-at-a-time review. Other recommendations remain proposals.

Inventory review status

Reference Review status Current decision or unresolved item
ADR-0001 Reviewed Retain C++23 subset; verify compiler/library feature support in each toolchain lane.
ADR-0002 Reviewed Retain target-oriented CMake, Ninja, and checked-in presets.
ADR-0003 Reviewed Pending: decide whether to compare/select a package manager now or defer selection until a dependency needs managed acquisition. Recommendation: defer; when that gate arrives compare vcpkg and Conan against the ADR's reproducibility, licensing, offline, and maintenance criteria.
ADR-0004 Reviewed Retain the bounded owned harness for now. Pending: decide whether the current test volume justifies adopting Catch2 (or another framework) and Google Benchmark; recommendation: avoid a migration absent concrete maintenance or measurement pain, and set a decision gate before the next major test-suite expansion.
ADR-0005 Reviewed Retain sorted component rows for the bounded proof; production ECS remains open. Re-evaluate with variable components, capability churn, long-lived entity churn, interacting systems, and external-ID mapping before biological entity expansion.
ADR-0006 Reviewed Fixed logical ticks retained. The old 50 ms value is superseded as an experiment by ratified M10 timing in ADR-0089: 50 ms. Rate changes require a versioned review.
ADR-0007 Reviewed SplitMix64 remains the synthetic proof generator. Production biology requires named subsystem streams and versioned distributions; no world-wide production stream choice is needed before those systems are specified.
ADR-0008 Reviewed Keep the local proof codec local. Durable checkpoints and replay use the later accepted formats/contracts in ADRs 0061–0065; M10 protocol is versioned separately. A universal serialization library remains unselected pending a concrete schema/workload comparison.
ADR-0009 Reviewed Keep explicit typed integer units and local frames. Dimensional library and planetary coordinate mapping remain deferred to biology/planet prototypes.
ADR-0010 Reviewed Retain std::expected for recoverable failures, optional only for legitimate absence, and process-boundary exception handling.
ADR-0011 Reviewed Retain single-writer simulation ownership; no general parallel scheduler until profiled immutable work and stable merge contracts justify it.
ADR-0012 Reviewed Keep bounded structured application logging at the current boundary. spdlog remains unadopted; reconsider only with measured sink/queue needs.
ADR-0013 Reviewed Historical recommendation superseded by ADR-0059, which adopts the scoped direct-libpq adapter. No decision needed here.
ADR-0014 Reviewed No broker currently: ADR-0072 uses the PostgreSQL outbox. Reconsider NATS only if measured fan-out/backpressure warrants an independent broker.
ADR-0015 Reviewed Keep simulation metrics separate from authoritative state. Later M7–M8 ADRs establish bounded server exposition and the selected observability stack; no metrics-client library is required now.
ADR-0016 Reviewed The initial persistence/ownership proposal is now scoped by accepted M7 contracts (ADRs 0058–0068), M8 durability (0069–0076), and M10 probe handoff experiment (0096). General production cross-region ownership remains a later-world-scale gate.
ADR-0017 Reviewed Server authority and no custom cryptography remain locked. M10 transport is QUIC with DATAGRAM plus reliable streams (ADR-0090); session identity/actions are scoped by ADRs 0091–0092. Sensory interest and production threat review remain later work.
ADR-0018 Reviewed Keep modular backend, PostgreSQL, durable object/snapshot boundaries, no Redis by default, and no broker absent measured need. M8 ADRs scope current dev/stage deployment and backup. MinIO must not be adopted as the default while its upstream distribution is archived; choose a maintained S3-compatible backend only when storage deployment is scheduled.
ADR-0019 Reviewed Sparse procedural base and persisted modifications remain locked. M5 local chunk representation is accepted (ADRs 0042–0049); planetary topology, global coordinates, LOD, mesher and hydrology remain experiments for M13/R&D.
ADR-0020 Reviewed Gitea Actions and the finite dev verification Job were accepted for foundation delivery; persistent region deployment is governed by later M7/M8 ADRs.
ADR-0021 Reviewed EnTT is an opt-in measurement dependency only; benchmark evidence retains sorted rows for the bounded proof and does not decide production ECS.
ADR-0022 Reviewed Sequential bounded atomic lifecycle-command scheduling is a synthetic M1.6 contract only; it does not authorize client/gameplay commands or concurrency.
ADR-0023 Reviewed Exact linear periodic-neighborhood query is the M1.7 baseline; no production spatial index or planetary geometry is selected.
ADR-0024 Reviewed Exact bounded integer inventory ledger is adopted for synthetic accounting; it does not establish production biochemical units or rates.
ADR-0025 Reviewed Finite-source producer capture/growth is an M2.2 fixture; synthetic parameters are not game biology.
ADR-0026 Reviewed Shared-owner feeding fixture is adopted for M2.3; intake/priority behavior remains a proof contract, not a final feeding rule.
ADR-0027 Reviewed Terminal/decomposition accounting is a bounded M2.4 fixture; it does not settle real corpse or nutrient-cycle rates.
ADR-0028 Reviewed Reproduction is fully funded and retains ancestry in M2 fixtures; duplication and construction overhead are explicitly not hereditary/gameplay rules.
ADR-0029 Reviewed Seeded short-episode driver is measurement tooling only; lifespan, targeting and contention restrictions are not ecology or AI decisions.
ADR-0030 Reviewed Immutable, validated, canonical genome v1 record is accepted for M3.1. Capability set/ranges are schema scope, not a complete mutation/development/gameplay decision.
ADR-0031 Reviewed Deterministic development and phenotype derivation are accepted for M3.2; v1 cube geometry/rule constants are a prototype, not the final body/development model.
ADR-0032 Reviewed Derived compartment composition and physical-resource budgets are accepted for M3.3; constants are explicitly not balance, viability, starvation or death rules.
ADR-0033 Reviewed Uniform seeded candidates over a bounded operator catalog are accepted as an M3.4 experiment. Environmental weighting, mutation count/cost rules, NPC mutation, recombination and inherited/acquired expression remain open gameplay decisions.
ADR-0034 Reviewed Immutable ancestry archive admission/birth/death rules are accepted; archive durability is later ADR-0062. This does not decide mutation by NPCs or species.
ADR-0035 Reviewed Seeded genome-derived ensemble evidence closes M3; no mutation happens during episodes and the results are not gameplay fitness/balance.
ADR-0036 Reviewed M4 cohort value/registry ownership contract is accepted with later M4.3 amendments; it is an abstraction representation, not proof of ecological fidelity.
ADR-0037 Reviewed Atomic identified-individual to cohort collapse is accepted for M4.2; no production persistence or behavior equivalence follows.
ADR-0038 Reviewed Bounded interval accounting is accepted without selecting a birth/mortality/ecology model.
ADR-0039 Reviewed Conservative bounded rematerialization is accepted; lossy cohort smoothing is explicitly exposed and cannot be read as an inverse.
ADR-0040 Reviewed M4.5 reports resolution-dependent error and a non-equivalence result for its fixture. Production tier thresholds require new multi-genome ecological evidence.
ADR-0041–0049 Reviewed M4.6 measures abstraction cost without enabling a runtime tier selector. M5 accepts a local 16³ chunk experiment, deterministic abiotic base, bounded edit ownership, codecs, and file publication/restart proof. These do not settle planet topology, world-scale resolution, meshing, chemistry, or the M7 durable store.
ADR-0050–0057 Reviewed M6 accepts typed bounded habitat/resource units, sparse field pages, views, atomic local transfers, coupled checkpointing, and synthetic process/decay/interval rules. These are versioned fixture rules; they do not settle planetary chemistry, hydrology, atmosphere, or final biological rates.
ADR-0058 Reviewed Region authority must commit complete checkpoints, including resource/source/retry state, under fenced ownership. This is the M7 recovery boundary, with implementation details delegated to later accepted ADRs.
ADR-0059–0068 Reviewed M7 contracts select the scoped direct-libpq adapter, database-time fencing, immutable checkpoint publication, durable outcomes, single-owner region host, recovery-before-readiness, operational endpoints, isolated clones, and build-bound upgrade barrier. Rules/format conversion still fails closed until a separately reviewed converter exists.
ADR-0069–0073 Reviewed M8 selects protected environment namespaces, provisional dev/stage store layout, single-replica host deployment, bounded Prometheus/Alertmanager/logging, DB-outbox background delivery, and operator-run deterministic fault drills. No World0 deployment is authorized by these records.
ADR-0074 Reviewed Verified snapshot-consistent backup/restore bundle and audit are accepted. Open: independent permanent destination, encryption, schedule/retention and PostgreSQL WAL archiving.
ADR-0075 Reviewed Dev/stage recovery target is worker-loss RPO 0/RTO ≤60s; host-loss restoration is an explicit rewind. World0 remains blocked until acknowledged outcomes survive host loss in an independent failure domain and promotion controls are satisfied.
ADR-0076 Reviewed Clone-based accelerated upgrade comparison, guarded simulation-version barrier, failure rehearsal, and policy-checked storage/compute node placement are accepted. Hardware and storage inventory still need validation before expansion.
ADR-0077–0086 Reviewed M9 accepts the scoped SDL engineering client, lifecycle/presentation/input boundaries, engineering-only ImGui and rebindable controls, plus a local reproducible Windows ZIP. These do not choose final player UI, 3D renderer, audio or physics. The absent Windows runner remains a packaging/CI limitation rather than a design choice.
ADR-0087 Reassessment in progress Route is included in the dev overlay and permitted by policy; absolute session-window code exists. VIP 172.10.1.202, UDP 7400, subnet 172.10.1.0/24 and four-hour windows remain the recorded choices. D-001 confirms manual Argo application. Live state and qualification evidence must be assessed separately.
ADR-0088 Reviewed Render all overlays with pinned checksum-verified kustomize and validate effective resources; route activation remains fail-closed until its full contract is implemented and amended.
ADR-0089–0096 Reviewed M10 has ratified 50 ms ticks and budgets, QUIC/msquic+OpenSSL transport, signed provider-independent grants, durable no-effect actions, bounded datagram snapshots, probe-only motion/interpolation, and a durable two-process handoff experiment. Probe movement/handoff do not establish organism rules.
ADR-0097 Reviewed Qualification must run on evolution-dev reference hardware; overload policy is refuse then shed; the first reference run was invalidated by cold-start governor behavior and grant retry consumption. The 10.9d corrected reference rerun remains an operational gate.

Open questions

Reference Question Options to consider Current recommendation Decision
ADR-0003 When to select a general C++ package manager? A. Select and compare now. B. Defer until a dependency requires managed acquisition. B; keep explicit, pinned dependency acquisition until a real need gives us a representative workload. Pending user response
ADR-0004 Should test/benchmark frameworks be standardized now? A. Adopt a framework now and migrate suitable suites. B. Keep existing harnesses until specific diagnostic/statistical gaps justify a migration. B; preserve low-friction tests and avoid churn unless the current harness is measurably limiting. Pending user response
ADR-0005 When to decide production ECS? A. Benchmark/adopt before the first biological entity systems. B. Keep the bounded rows through the next gate and decide when realistic component churn exists. B; defer to representative biological workloads, while explicitly preventing the proof storage from becoming an unreviewed production commitment. Pending user response
ADR-0006 Does the ratified 50 ms M10 tick need revisiting now? A. Retain it until evidence requires change. B. Reopen timing before M11.1. A; ADR-0089 ratified it and requires future rate changes to be versioned and evidenced. No change requested; retain current contract
ADR-0007 When to specify production biological PRNG streams? A. Specify before genome/mutation implementation. B. Defer exact stream partitioning until the M11 biology contracts, preserving explicit seeded streams in the meantime. B, with the M11 gate required to freeze stream names, draw ownership, versioning and regression vectors before gameplay use. Pending at M11 biology contract
ADR-0008 Which production serialization library/schema should be used? A. Choose a cross-project library now. B. Keep explicit versioned codecs per domain until a shared workload justifies a library comparison. B; the repo already has concrete persistence and wire contracts, and a library should not replace them without evidence. Pending until schema consolidation need
ADR-0009 Adopt a dimensional-units library? A. Evaluate/adopt before M11 biology expands. B. Keep typed integer quantity wrappers and evaluate only if conversion complexity or dimensional mistakes justify it. B for current integer accounting; revisit before introducing broad continuous physical quantities. Pending at biology/physics boundary
ADR-0012 Adopt a general logging library? A. Adopt spdlog now. B. Retain current bounded sink boundary until operational needs are measured. B; no concrete shortfall is recorded, and logging cannot become simulation truth. No change requested; retain current boundary
ADR-0014 Add NATS JetStream? A. Introduce for asynchronous fan-out. B. Keep PostgreSQL outbox/direct calls unless measured fan-out requires a broker. B; later ADR-0072 selected the outbox and did not adopt a broker. No broker now; revisit only on measured need
ADR-0015 Add a Prometheus/Otel client library? A. Adopt a client library. B. Keep the tested bounded exposition/adapters. B; later ADRs already define the implemented observability boundary. Resolved by later scoped ADRs
ADR-0016 What durability/ownership guarantees apply beyond the M7–M8 single-region contracts and M10 probe transfer experiment? A. Extend the accepted fenced/checkpoint model to real cross-region ownership only when gameplay requires it. B. Specify global multi-region authority now. A; current process has no real organism migration requirement, and M10.6 is explicitly experimental. Defer to real cross-region gameplay gate
ADR-0017 Is ADR-0090's QUIC choice sufficient for current M10, and when revisit perception-aware interest/security? A. Keep QUIC for M10 and schedule a separate security/perception gate before real gameplay. B. Reopen transport now. A; M10 includes measured channel behavior, while sensory interest is not implemented. Current transport settled; future security/perception review pending
ADR-0018 Which S3-compatible object store and backup destination? A. Select a maintained backend and independent destination before persistent-store rollout. B. Defer store selection until dedicated storage rebuild/backup gate. B for now; ADR-0074 records destination, encryption, schedule and WAL archiving as pending. Avoid archived MinIO upstream by default. Pending storage/backup gate
ADR-0019 Which planetary topology, voxel LOD and surface mesher? A. Choose a candidate now. B. Prototype cubed sphere/topology and compare meshers with shared fixtures in M13. B; topology and geometry are explicitly experiments and no renderer is needed yet. Pending M13 R&D
ADR-0020 Is the foundation delivery model still appropriate? A. Keep Gitea CI plus manual reviewed digest promotion. B. Reopen delivery architecture. A; later GitOps/Argo ADRs preserve separation between image build and cluster sync. No change requested; retain
ADR-0030–0035 Are M3 genome/development/mutation rules final gameplay rules? A. Treat these contracts as the game rules. B. Keep them as an initial schema/prototype and decide omitted heredity/ecological pressures before gameplay. B; every M3 ADR scopes parameters and candidate generation as experiments and defers the major biological questions to later gates. Keep prototype scope; reopen before gameplay integration
ADR-0036–0040 / OPEN_QUESTIONS #16 What fidelity/error envelope is acceptable for production population abstraction? A. Adopt the M4 fixture thresholds as production gates. B. Use M4 only as proof of measurable loss, then validate thresholds against representative multi-genome cellular ecology. B; M4 evidence explicitly says its one-genome fixture does not generalize. Pending representative population-model gate
ADR-0070, 0074 / OPEN_QUESTIONS additional blockers Where are durable object snapshots and backups stored, encrypted, retained and verified? A. Finalize after storage/RAID rebuild with an independent failure domain. B. Treat current node-local dev/stage volumes as sufficient for World0. A; ADR-0070 calls its storage provisional and ADR-0075 blocks World0 until independent durability is proven. Pending storage rebuild and backup plan
ADR-0072 Where should actionable alerts be delivered? A. Configure a destination once operational ownership/contact path is chosen. B. Keep alerts local-only. A before persistent unattended operation; destination is intentionally pending. Pending operations/contact decision
ADR-0075 / OPEN_QUESTIONS #38 What World0 RPO/RTO and promotion approvers are acceptable? A. Use the dev/stage limits and explicitly accept host-loss rewind. B. Define stricter independent durability and named human approvers before World0. B; current ADR blocks World0 until independent-failure-domain ACK durability and promotion authority exist. Pending before World0, not before current development
ADR-0087 Does route documentation match repository and live evidence? Reconcile existing configuration and qualification evidence; this is factual verification, not a new enable/disable choice. The dev overlay already includes the route; the earlier disabled assertion was stale. Repository status corrected; live qualification evidence still requires reconciliation.
ADR-0087 How should workflow-created windows reach Argo? A. Workflow prepares the change; operator manually syncs. B. Restricted workflow-driven Argo sync. A. A accepted, 2026-10-08; D-001. Live Status read path is separate follow-up work.
ADR-0087 How should clock changes affect a running session window? A. Close at either UTC expiry or a suspend-aware elapsed deadline, whichever comes first. B. Rely on UTC expiry and host synchronization alone. A. A accepted, 2026-10-08; D-002. Implementation/tests and numerical clock-health tolerances remain follow-up work.
ADR-0097 When should the corrected reference load drill run? A. After the new image is promoted and a grant bundle is prepared, run the 10.9d reference test on evolution-dev. B. Defer the qualification and leave M10.9 open. A, because the first run did not qualify capacity and the corrected harness is now in user changes. Awaiting operator window/image promotion

Reviewed open-question register

The crosswalk below preserves the original numbering in OPEN_QUESTIONS.md. “Pending” means the architecture needs a future design decision; it does not necessarily block the current engineering milestone. User decisions can be submitted in plain text by question number (for example, “Q3 B; Q14 A, with…”); the register will record the choice and any limits.

# Topic Options (A / B) Recommendation Disposition / source
1 Tick frequency and biological rates A. Keep 50 ms. B. Reopen the tick quantum. A; time-based biological rates convert deterministically at the tick boundary. Ratified in ADR-0089; rate migration/precision is M11.1.
2 Fixed/variable boundaries A. Fixed authoritative ticks, host wall-clock pacing. B. Variable authoritative dt. A; variable dt would weaken replay and rate accounting. Settled by ADR-0006/0089; ecology uses integer tick multiples.
3 ECS A. Keep sorted rows until realistic component churn exists. B. Adopt a production ECS now. A; benchmark against biological workloads before exposing library handles or changing IDs. Pending production ECS gate; ADR-0005/0021.
4 Dependency manager A. Compare/select now. B. Defer until managed acquisition solves a concrete dependency problem. B; compare vcpkg and Conan on a real compiled dependency and include offline/reproducibility/license evidence. Pending; ADR-0003.
5 Units library A. Keep typed integer wrappers for current exact accounting. B. Evaluate a dimensional library before continuous physics expands. A now, with a review gate before continuous biological/physical equations. Pending at that gate; ADR-0009 and ADR-0024.
6 Determinism guarantees A. Require bit-exact determinism for all systems/platforms. B. Define exactness per integer system and explicit tolerances for floating-point/physics. B; keep authoritative decisions reproducible and versioned, but do not promise universal floating-point identity. Pending a per-system contract; ADR-0007/0089.
7 Mutation generation A. Keep uniform seeded valid candidates as a baseline. B. Add bounded environmental/behavioral context without a ranked “best build.” B for gameplay, with explicit and testable context inputs; preserve the unweighted catalog as a baseline. Pending before player-facing mutation; ADR-0033.
8 Genome representation A. Extend the immutable developmental-capability record. B. Replace it with literal DNA or a new representation. A; maintain versioned capability schemas, not literal DNA. v1 accepted; determine capability expansion/versioning before gameplay; ADR-0030.
9 Developmental body representation A. Typed developmental/body graph. B. Direct arbitrary mesh editing. A; geometry must derive from capabilities and validated attachments. Pending M13 morphology contract; ADR-0031 and OPEN_QUESTIONS.
10 Heredity and reproductive mutation A. Funded offspring with explicit seeded mutation and parent ancestry. B. Copy the genome unchanged. A as a design direction; define mutation timing, costs, count, and inheritance before M11 gameplay. Pending gameplay rule; ADR-0028/0033.
11 NPC mutation A. Descendants can diverge under the same bounded mutation rules. B. Retain M11 exact-copy budding pending qualification. B for the current scope; autonomous divergence requires funded biological rules and ensemble evidence. Explicitly deferred by M12.1; ADR-0133 preserves ADR-0125, not a permanent prohibition.
12 Species/speciation A. Optional derived, versioned biological classification with no class mechanics. B. Player-selected species/class labels. A; taxonomy must not gate capabilities or progression. ADR-0133 distinguishes organism/genome/lineage/species; classifier and threshold explicitly deferred.
13 Lineage extinction A. Extinct only with no living members and no viable reserve at any simulation tier. B. Mark extinct when no active individuals remain. A; suspended, dormant, reproductive and in-flight evidence must be reconciled; uncertainty prevents terminal extinction. Evidence semantics in ADR-0133; global reconciliation remains M12.7, local archive status unchanged.
14 Offline player state A. Remove/park the controlled organism with zero progression and explicit reconnect policy. B. Leave it active in-world and vulnerable. A; never allow offline farming, and specify whether logout safely removes or leaves a non-progressing body before persistence. Pending player-persistence design; founding requirement.
15 Cellular-to-multicellular transition A. Require physiological capability and ecological viability, without a linear ladder. B. Use a fixed progression threshold. A; transition should emerge from development and available resources. Pending M14 criteria; biology/world design.
16 Population abstraction math A. Use the M4 measured fixture bounds. B. Calibrate cohorts against representative multi-genome cellular ecology and declare error envelopes. B; M4 proves loss can be measured but its numbers do not generalize. Pending population model; ADR-0036–0041.
17 Active region dimensions A. Benchmark against interest, entity count and migration overhead. B. Pick a fixed size now. A; dimensions are operational performance choices, not permanent world identity. Pending M7/M10 production workload; ADR-0060/0096.
18 Planet coordinates A. Keep bounded local integer frames with stable global addresses. B. Commit to a full planetary projection now. A now; prototype global mapping before the voxel planet. Local contract accepted; planetary mapping pending M13; ADR-0009/0042.
19 Spherical voxel strategy A. Prototype cubed sphere and compare alternatives. B. Use one global Cartesian grid. A; test poles, seams, deep edits and LOD. Pending M13; ADR-0019.
20 Voxel LOD A. Sparse procedural base, aggregate far fields, active detailed chunks. B. Uniform meter voxels everywhere. A; it is required by the sparse-world vision. Principle locked; thresholds/refinement policies pending prototypes; ADR-0019.
21 Surface meshing A. Compare Marching Cubes, Surface Nets and Dual Contouring on shared fixtures. B. Select a familiar mesher now. A; measure seams, topology, collision, edits and cost. Pending M13; ADR-0019.
22 Hydrology A. Start with conserved regional reservoirs/flux. B. Build full fluid dynamics first. A; constrain water accounting before more detailed flow. Pending environmental R&D; ADR-0050/0051.
23 Atmosphere A. Aggregate mass/species budgets. B. Global computational fluid dynamics. A; no global CFD promise without a demonstrated need. Architecture direction recorded; detailed model pending world/ecology design.
24 Cross-region ownership A. Fenced lease and explicit prepare/transfer/commit recovery. B. Duplicate state during handoff. A; exactly one authoritative owner at every step. Probe experiment accepted; production organism semantics pending gameplay; ADR-0060/0096.
25 Reliability layer A. Use library-backed QUIC streams plus datagrams. B. Implement reliability over UDP. A. Settled for M10 by ADR-0090; revisit only with new platform or workload evidence.
26 Serialization A. Keep explicit bounded versioned codecs per domain. B. Adopt one universal schema library now. A until a shared workload and migration need justify a measured comparison. Pending only if consolidation is needed; ADR-0008/0061/0090.
27 Snapshot/events A. Snapshots + bounded replay batches + durable semantic history. B. Persist every tick/event as universal event sourcing. A; keep full recovery state and history records distinct. M7 contract accepted; ADR-0058/0061–0063.
28 Recovery A. Replay a committed contiguous range under a unique fenced epoch. B. Resume from latest snapshot without full verification. A. Settled for M7 region host; ADR-0060/0065.
29 Cross-version migration A. Clone, migrate, accelerated compare, then explicit barrier/approval. B. Allow mixed simulation versions in adjacent live regions. A; no live-world rule drift. Framework accepted; actual converters and approval policy pending per migration; ADR-0068/0076.
30 Steam identity A. Provider subject maps through an adapter to internal AccountId. B. Make Steam ID the account key. A; preserve provider independence. Architecture direction accepted; integration deferred until Steam milestone; ADR-0091.
31 Friend recognition A. Separate authenticated social signal; no player/NPC tags. B. Show ownership/player labels. A; this preserves the shared biological presentation requirement. Principle locked; authorization and presentation details pending social feature.
32 Anti-cheat A. Server authority, bounded authenticated intents, rate limits and validation. B. Trust client state and reconcile abuses later. A. Core policy locked; threat review/abuse coverage continues at M10 and before gameplay; ADR-0017/0091/0097.
33 AI behavior A. Shared bounded intention interface and deterministic controller. B. Separate AI-only command path. A; prevent privileged AI outcomes and keep control source separate from biology. Pending behavior architecture before M11 AI; ADR-0022/0079.
34 Evolvable behavior A. Optional heritable behavior capabilities with costs. B. Retain the fixed bounded M11 wildlife controller for now. B for the current scope; A remains an experiment and intelligence remains optional. Explicitly deferred by M12.1/ADR-0133 pending funded heritable-state rules and evidence.
35 Morphology limits A. Body graph validates material, attachment and genetic capability. B. Allow arbitrary player geometry. A. Pending M13; ADR-0030/0031 and body-graph proposal.
36 Procedural animation A. Prototype IK/gaits/locomotion from morphology. B. Depend on hand-authored species animations. A; hand-authored animation cannot define the open-ended body system. R&D pending M13; no early implementation required.
37 Retention A. Keep genome ancestry permanently and tier older replay/snapshots/backups. B. Retain every raw event and snapshot forever at full resolution. A; preserve history while defining separate replay/archive retention and verified backup policies. Ancestry permanence locked; concrete retention periods pending backup design; ADR-0034/0062/0074.
38 Disaster recovery objectives A. Dev/stage worker-loss RPO 0, RTO ≤60 s; host-loss restore is explicit rewind. B. World0 requires acknowledged outcomes in an independent failure domain before ACK. Both, applied to their respective environments; do not relax the World0 gate. Dev/stage ratified; World0 remains blocked; ADR-0075.