Eve library

Lilith V1.6 — Living Worlds: delivery, benchmarks and acceptance contract (2026-10-03)

Spec

Revision 6

On this page

Companion PRD: Lilith V1.6 Living Worlds.

Date: 2026-10-03. Status: proposed with the companion V1.6 PRD. All numeric values below are proposed release targets, not measured achievements or vendor performance claims. They become binding when the scope is approved. This contract covers all 126 PRD requirements across 24 workstreams.

1. What the final release contains#

  1. A real Maya–Neith native engine, headless server, full creator editor and typed human/agent operation system.
  2. A documented public SDK, working templates, asset/package/plugin/mod workflows, build/cook/patch tools and supported platform distributions.
  3. Maya Living World, a complete high-quality open-world game proving the integrated technology.
  4. A finished second game of a materially different genre, produced through the public tools by a separately operated creator/team.
  5. Every research family named in the PRD implemented as a real selectable capability, measured and supplied with a supported conventional fallback.
  6. Independent verification and an authorized maintained pilot. Candidate-ready and fully delivered are separate states; public publication requires Elikem’s separate instruction.

A milestone can be useful and demonstrable before release, but cannot redefine V1.6 to exclude unfinished requirements. Platform profiles describe legitimate fidelity/scalability differences; they cannot disguise an absent engine subsystem.

2. Milestones and dependency order#

Milestone Required working result Exit evidence
M0 — contract and ownership Current code/board audit; frozen benchmark corpus; architecture decisions; exact hardware/SDK access and cost plan; cross-release ownership and dependency map Approved scope and required ADRs, reproducible baseline manifests, complete traceability and acyclic task graph; unresolved access/cost decisions visible
M1 — executable spine Native player/editor host/server composition, world/asset identity contracts, lifecycle, scheduling, initial instrumentation and real device output Clean-checkout boot/load/play/save/exit, real GPU capture, component/contract tests and recovery failures exercised
M2 — production paths Real asset import/cook/cache, unified operations, shader pipelines, basic UI-to-runtime workflows and automated packaged test harness B01 and targeted B02/B09/B10 evidence; working operations and real files, no mock substitution
M3 — complete tools and presentation Production renderer, streamed world, animation/physics/audio, visual/text scripting and real creator workflows; first coherent district B02/B03/B04/B07/B11 on an integrated district with measured budgets; this district is an intermediate milestone
M4 — living multiplayer game Persistent life/economy/law/mission systems, real model cognition, complete gameplay framework, network authority and saves B05/B06/B08/B10 with 32 clients, real active NPC simulation and end-to-end consequences
M5 — supported engine product Full platform coverage, public SDK/templates, advanced methods, diagnostics and creator independence All physical platform gates; real 128-client run; complete specialist tools; independent creator can package a game
M6 — complete production candidates Full flagship content, independent second game, final art/audio/gameplay quality and every requirement at integrated/hardware level All B01–B12 pre-pilot cases passing, full coverage audit, no release-blocking defects, independent technical and creative reviews
M7 — maintained release proof Authorized pilot, long soaks, support/update/recovery evidence and final benchmark refresh 30-day record, every applicable requirement at production level, Elikem’s release decision and separately authorized publication/promotions

Workstream phase numbers in the PRD are earliest integration targets, not promises that all work ends there. CORE and VERIFY start together after GOV. DATA and OPS then unblock graphics/tools. Geometry, lighting, look, world, animation, physics and audio build on the spine; the living simulation, gameplay and networking integrate after their prerequisites. NPC inference must not own authoritative world state. SDK consumes stable editor/scripting/platform interfaces. SHIP consumes all streams. Every stream also feeds the final VERIFY and SHIP evidence bundle without making early instrumentation depend on the finished game.

Prototype findings can refine implementation choices inside the approved contract. Any weaker target, dropped profile or missing capability requires an explicit scope amendment. There is no automatic conversion of a mandatory feature to “research only” or “future release.”

3. Hardware and performance contract#

3.1 Proposed reference matrix#

These devices are proposed test fixtures, not an assertion that Oshun owns them. M0 must inventory availability and secure authorized access. Exact OS, API, driver, compiler, firmware, power mode and build hashes are recorded for every run. Substitution requires an approved equivalent-or-harder profile and preserved comparison results.

Profile Reference fixture Required mode and evidence
PC quality, NVIDIA Ryzen 7 7700, 32 GiB RAM, RTX 4070 12 GB, NVMe; Windows and Linux where supported DX12/Vulkan, 2560×1440 output, internal resolution at least 67% per axis when using quality upscaling; native-resolution results also reported
PC quality, AMD Same CPU/RAM/storage class, RX 7800 XT 16 GB Equivalent Vulkan/DX12 scene workload and output settings, independently tuned without reducing required content
PC compatibility, Intel Same CPU/RAM/storage class, Arc B580 12 GB Explicit performance profile at 1920×1080; functional conformance and real driver/device recovery
PC minimum Ryzen 5 5600, 16 GiB RAM, RX 6600 8 GB, SSD 1920×1080 performance profile with measured quality reductions and the same gameplay semantics
Mac M3 Pro class, at least 18 GB unified memory Native Metal player/editor at 1920×1080 performance profile; unified-memory pressure and long-session evidence
Browser WebGPU on current supported desktop browsers, including hardware/API diversity B01/B02/B05 compact conformance scenes at 1920×1080, declared browser limits and no fake native parity
Mobile iPhone 16 Pro class and Pixel 9 Pro class; a supported tablet from each ecosystem Physical iOS/Android installs, dynamic resolution with a frozen floor, 30-minute thermal soak and full lifecycle/input coverage
XR Quest 3, a PC OpenXR headset with the quality PC, and Apple Vision Pro Actual stereo device runs at the selected native refresh, input/tracking/audio/haptics and comfort evidence
Consoles Authorized PS5, Xbox Series X/S and Switch 2 devkits Actual packaged native runs and platform-required certification; exact quality/performance modes frozen with legitimate SDK access
Dedicated server 16 physical-core equivalent, 64 GiB RAM, NVMe, measured network; exact SKU and topology frozen at M0 128 real networked clients at 30 Hz in the declared region/world workload, with real persistence and fault recovery

Desktop and console targets run the complete flagship using appropriate explicit quality modes. Browser/mobile/XR ship complete engine profiles and the shared playable conformance experience derived from the flagship; they are not promises of identical desktop photorealism or identical campaign presentation. Their profile definitions must include gameplay, editing/authoring availability, storage, network, shader and asset limits. Desktop hosts carry full editor parity. Browser/mobile/XR authoring surfaces are scoped to their usable device workflows and must be specified at M0; no unlisted reduction can pass by assumption.

3.2 Frame and interaction gates#

  • PC quality and Mac/performance-desktop profiles: steady-play frame time p95 ≤16.67 ms and p99 ≤25 ms on every fixed route; no more than one frame over 50 ms per ten-minute route, and none over 100 ms after initial loading. Report CPU and GPU times separately, uncapped base rendering, presentation pacing and input-to-photon latency.
  • PC minimum and mobile/browser compact profiles: p95 ≤33.33 ms, p99 ≤50 ms, no more than one frame over 100 ms per ten minutes. A device-specific 60 Hz mode must meet the tighter profile.
  • XR: p95 at or below the selected native frame interval; target 90 Hz where supported (11.11 ms), otherwise the explicitly approved native profile. Record dropped frames, reprojection, motion-to-photon latency and thermal behaviour separately; synthetic frames do not replace base-render evidence.
  • Consoles: provide a frozen 60 Hz performance mode and quality mode on PS5/Series X, and a declared 30 Hz minimum on Series S/Switch 2 with equivalent gameplay. Any change in the proposed mode contract is reviewed explicitly.
  • Local input-to-photon p95 ≤80 ms at 60 Hz, measured with an external or validated end-to-end method. Audio command-to-output p95 ≤50 ms in the declared normal device path. Network-authoritative action response reports the RTT-dependent part separately.
  • No frames may be omitted from percentile/hitch reporting because streaming, shaders, AI, autosave or network correction made them slow. Loading screens are measured separately and must be clearly delimited. Frame generation stays off for qualification.
  • Freeze subsystem budgets before M3: initial quality-profile targets are GPU ≤14 ms p95, CPU critical path ≤12 ms p95, including measured AI/physics/streaming overhead. These overlap in a pipelined frame and must not be added as fake total latency. Final end-to-end gates take precedence.

3.3 Memory, streaming and reliability#

Quality PC client resident set ≤12 GiB and GPU-resident working set ≤10 GiB on the 12 GB fixture. The 8 GB GPU profile reserves operating-system/driver headroom and caps game GPU use at 6.5 GiB. Unified-memory/mobile/XR profiles receive explicit caps before implementation, tied to measured OS termination limits. No cap can be inferred from nominal hardware capacity alone.

After warmup, an eight-hour fixed-loop session must stay within 5% or 256 MiB (whichever is smaller) of its stable resident-memory baseline, allowing documented bounded caches. No unbounded entity, audio, job, event, inference or network queue is permitted. Track peak allocation, residency churn and paging.

Streaming routes include 240 km/h ground/aerial traversal, camera reversal, teleport and deliberately slowed I/O. Geometry visibility, collision and mission readiness are synchronized; a temporary loading gate is acceptable only where the authored UX contract permits it and its latency is reported. Cold launch to controllable gameplay target ≤30 seconds and warm launch ≤10 seconds on the quality NVMe fixture; scene transitions and teleport loading targets ≤5 seconds where a transition gate is used.

Run three cold-cache and five warm-cache repetitions per benchmark configuration; report distributions, all failures and run-to-run variation. Capture at least ten minutes per route and separately run eight-hour sessions. A failed run cannot be discarded without a documented invalid-measurement reason reviewed independently.

4. World, content and population contract#

The final flagship contains a continuous 64 km² authored playable world (8×8 km equivalent), including at least 4 km² of dense urban play, three distinct settlements, wilderness/biomes, navigable water and meaningful vertical/interior space. Include at least 80 distinct usable interiors, 24 complete authored missions spanning traversal, combat, stealth, investigation, social and systemic play, and a coherent campaign with a target of at least eight hours for the median first-time playtester. Empty terrain, repeated filler and unconnected minigames do not satisfy these floors.

The production/art plan includes named characters, rigs, animation and capture datasets, environments, props, UI, voiced/captioned dialogue, music, foley, VFX and mission scripting. It specifies reuse, generation, manual finishing, rights and review. Content quantity is a workload and completeness floor; craft and playability remain separate gates.

Single-player quality-profile population: 10,000 persistent NPC identities; in the fixed dense scene, at least 64 simultaneously fully simulated agents including the player’s immediate interactions, 512 reduced-fidelity agents and the remaining persistent aggregate population. Also include at least 100 active traffic vehicles in the region and a declared wildlife population. Each tier’s perception, animation, pathfinding, planning and update cadence is documented; off-screen bookkeeping is not advertised as 10,000 fully embodied agents. These are floor workloads, not an architectural maximum.

Measure actual wall-clock CPU time and memory. Initial target: full population simulation ≤4 ms p95 on the quality PC critical path using the declared worker allocation; model inference runs asynchronously with separately measured compute contention. The final frame gate still applies with graphics, physics, audio, streaming and networking enabled.

World continuity suite: at least 1,000 generated-and-reviewed consequence cases spanning inventory, social memory, mission state, law/economy, damage, ecology and LOD transitions. Zero lost/duplicated ownership or corrupted saves; at least 99.5% success on noncritical behavioural continuity assertions, with every failure triaged. Critical mission solvability and security invariants require 100%.

The second game must differ materially in genre and scene/workflow needs, have a complete start-to-finish loop of at least 30 minutes, and use the public SDK/editor without a private engine fork. Its purpose is to expose generality, not inflate flagship content.

5. Multiplayer, collaboration and cognition gates#

Multiplayer#

M4 requires 32 real clients; the final gate requires 128 real clients with active scripted player behaviours, spread across at least three physical or separately provisioned machines, communicating through the actual shipping transport. Bots must run the real client protocol, not mutate server memory. Include a rendered cohort to measure client impact; report headless versus rendered clients explicitly.

At the final 30 Hz server tick: p95 tick ≤33.33 ms, p99 ≤50 ms and ≤0.1% missed ticks over the steady test interval. Exercise 80 ms RTT, 20 ms jitter and 1% packet loss as the normal impaired profile; also test a severe 150 ms/3% loss profile for correct degradation and recovery. Position correction thresholds are scenario-specific, frozen at M0 using body/vehicle dimensions; no silent rubber-banding or impossible hits may be accepted merely because average bandwidth is low.

Provide an explicit per-client and aggregate bandwidth budget, initially ≤256 kbit/s average gameplay downstream per active client excluding voice/content downloads, with bounded bursts. If the actual game needs more, revise the budget transparently before qualification. Cross-region handoff and reconnect preserve transactions and canonical identity. Run a 72-hour server soak with zero corruption or unauthorized ownership changes. Recovery target: process failure returns the region to service within five minutes with no lost committed economic transactions; nontransactional motion recovery point ≤30 seconds.

Collaboration#

Eight separately authenticated editor clients, at least two working on conflicting shared assets, under 100 ms RTT and 1% loss; include offline edits and server restart. Committed changes become visible within two seconds p95 outside explicitly long-running operations. Conflicts are visible, attributable and recoverable. At least 100 conflict/reconnect permutations pass with no lost accepted edits.

Model-backed cognition and creation#

Real local inference must provide first usable NPC reaction through the conventional response layer within 100 ms; model-produced conversational first audio target ≤1.5 seconds p95 and complete constrained action-plan response ≤2 seconds p95 for the frozen model/context/hardware profile. Late inference cannot block gameplay, change already-rejected actions or exceed the frame budget. Measure quality, latency, context length, memory, energy and cost together. Hosted optional paths report network/cost separately.

The 500-case NPC corpus covers factual grounding, private knowledge, persona, memory, impossible requests, action preconditions, injection attempts and interrupted plans. Critical authorization/world-integrity tests require 100%; grounded factual/character consistency target ≥95%, with an independent sampled review and no reliance on a model-only judge.

The 30 held-out creator briefs include scene composition, assets/materials, missions/scripts, animation, audio, optimization and debugging. Compare the same brief and quality rubric against the configured Unreal human-only and AI-assisted workflows. Target ≥90% complete, valid outcomes within the declared budget, all lock/permission invariants passing, and at least 20% lower median hands-on creator time at noninferior quality. Failure to beat this target means no creation-efficiency leadership claim; the requirement remains open until improved or explicitly amended.

6. Quality, usability and competitive claims#

Render correctness uses a versioned reference corpus, fixed exposure/colour transforms and region-specific checks. Before implementation, M0 freezes quantitative thresholds calibrated against correct and deliberately broken outputs: static structural similarity, perceptual error, radiometric error where meaningful, temporal ghost/disocclusion error and silhouette/detail preservation. No single average image metric can pass all materials, lighting and motion. The current broad GI-conformance tolerance must not be used uncritically.

The minimum independent craft rubric is 4/5 median with no dimension below 3/5 for scene fidelity, temporal stability, animation/contact, believable behaviour, sound, controls and mission coherence. At least five independent reviewers perform blinded paired reviews against the frozen reference brief. Noninferiority requires no more than 0.25/5 mean deficit on any required dimension, with confidence intervals reported; ambiguous results require more samples. These are proposed project gates, not scientific proof of universal engine superiority.

Creator usability: at least eight independent participants complete installation, new project, import/reimport, scene editing, scripting, debugging, play, save/collaboration and packaging briefs. Core task success ≥90%; no unrecoverable data loss; median task times within 10% of the configured Unreal baseline before claiming parity. Warm play/reload ≤5 seconds for the conformance project, incremental simple material/gameplay changes visible ≤2 seconds excluding explicitly disclosed shader work, and a one-asset incremental cook ≤30 seconds on the quality workstation. Cold full-project cook targets are frozen from the declared asset corpus and machine budget at M0.

Accessibility includes automated web checks, keyboard-only completion, visible focus, readable scaling to 200%, captions and nonaudio cues, controller rebinding, reduced motion and a documented mobile/XR interaction alternative. Automated checks do not replace assistive-device and human review.

Industry-leading claims must be limited to dimensions that demonstrate a material advantage: target at least 20% lower creator effort at matched quality; coherent persistent consequences across all required systems without critical continuity errors; and stronger measured user control/recovery for mixed human/agent production. General feature parity, performance, image quality, usability and reliability must all independently pass; a single outstanding metric cannot compensate for a missing platform or broken workflow. RAGE comparison is to observed shipped-game outcomes and public presentations, not guessed internal architecture or fabricated benchmark numbers.

7. Integrated scenarios#

B01 — Boot, create, cook and play#

Workstreams: CORE, GPU, DATA, EDIT, SCRIPT, OPS, PLAT, SDK.

A clean installation creates a project, imports a real rigged asset, edits a material and mission, plays, debugs, cooks, installs a separate package, saves and updates it. Repeat through UI and typed operations; inject failed imports, shaders, permissions and interrupted cooks.

Evidence: exact build and content manifest, inputs/assets/seeds, machine and network profiles, automated assertions, complete timings/traces, captured output and independent review where applicable. Both success and injected-failure routes must pass.

B02 — Rendering and atmosphere corpus#

Workstreams: GPU, GEO, LIGHT, LOOK, RND, VERIFY.

Static and moving-camera scenes cover fine geometry, foliage, hair, skin/eyes/cloth, glass/water, layered materials, many lights, interiors, night, fog/clouds, rain and destruction. Include path-traced/reference captures, native/upscaled modes, disocclusion and camera cuts.

Evidence: exact build and content manifest, inputs/assets/seeds, machine and network profiles, automated assertions, complete timings/traces, captured output and independent review where applicable. Both success and injected-failure routes must pass.

B03 — Whole-world traversal#

Workstreams: WORLD, GEO, LIGHT, LOOK, PHYS, LIFE, AUDIO, SAVE, VERIFY.

Repeat a fixed route by foot, horse, vehicle, boat and aircraft through city, wilderness, interiors and water. Include 240 km/h travel, teleport, reversal, cold/warm caches, storage throttling and memory pressure. Measure all residency and gameplay readiness.

Evidence: exact build and content manifest, inputs/assets/seeds, machine and network profiles, automated assertions, complete timings/traces, captured output and independent review where applicable. Both success and injected-failure routes must pass.

B04 — Performance and physical acting#

Workstreams: ANIM, PHYS, LOOK, AUDIO, NPC.

Run held-out human and animal clips/paths, retargeted body proportions, slopes/stairs, paired interactions, vehicle entry, impacts and recovery, expressive dialogue and interruptions. Review full motion sequences and audio, not only still images.

Evidence: exact build and content manifest, inputs/assets/seeds, machine and network profiles, automated assertions, complete timings/traces, captured output and independent review where applicable. Both success and injected-failure routes must pass.

B05 — Market-stall consequence#

Workstreams: CORE, WORLD, ANIM, PHYS, LIFE, NPC, GAME, AUDIO, NET, SAVE.

During rain, a vehicle damages a market stall. Goods move, surfaces change sound, a witness sees or misses it according to occlusion, workers react, traffic/navigation detours, property/economic and law consequences follow, dialogue reflects legitimate knowledge, and a mission branches. Stream away, save, restart, rejoin and verify the same consequences.

Evidence: exact build and content manifest, inputs/assets/seeds, machine and network profiles, automated assertions, complete timings/traces, captured output and independent review where applicable. Both success and injected-failure routes must pass.

B06 — Living population and history#

Workstreams: LIFE, NPC, WORLD, SAVE, VERIFY.

Run 10,000 persistent NPC identities with the declared active/reduced/aggregate tiers, real paths/decisions, traffic, wildlife, economy and social propagation. Exercise repeated tier promotion/demotion and acceleration of long history separately from wall-clock CPU measurements.

Evidence: exact build and content manifest, inputs/assets/seeds, machine and network profiles, automated assertions, complete timings/traces, captured output and independent review where applicable. Both success and injected-failure routes must pass.

B07 — Creator production and collaboration#

Workstreams: DATA, EDIT, SCRIPT, OPS, SDK.

Eight separate editor clients author adjacent regions and shared assets, including offline edits, conflicting deletes, locks, reimport, rollback and reconnect under latency/loss. Independent creators complete identical measured briefs with human-only and assisted workflows.

Evidence: exact build and content manifest, inputs/assets/seeds, machine and network profiles, automated assertions, complete timings/traces, captured output and independent review where applicable. Both success and injected-failure routes must pass.

B08 — Networked play and failure#

Workstreams: NET, CORE, WORLD, PHYS, LIFE, GAME, SAVE, AUDIO.

Run 32 then 128 real clients in separate OS processes across at least three machines; exercise authority, prediction, vehicles, missions, voice, late join, region handoff, malicious RPCs and server/client failure under network impairment.

Evidence: exact build and content manifest, inputs/assets/seeds, machine and network profiles, automated assertions, complete timings/traces, captured output and independent review where applicable. Both success and injected-failure routes must pass.

B09 — Packaging, recovery and compatibility#

Workstreams: DATA, SAVE, PLAT, SCRIPT, SDK.

Install, update, interrupt, corrupt, roll back, migrate saves/plugins and restore backups on every physical target profile. Test offline start, denied permissions, device loss, suspension and storage exhaustion.

Evidence: exact build and content manifest, inputs/assets/seeds, machine and network profiles, automated assertions, complete timings/traces, captured output and independent review where applicable. Both success and injected-failure routes must pass.

B10 — Agent creation and NPC cognition#

Workstreams: OPS, NPC, EDIT, SCRIPT, GAME, RND.

Execute 30 held-out creative briefs and 500 held-out grounded NPC decision/conversation cases with real models, locks, budgets, injection attempts, cancellation, hosted failure and offline fallback. Preserve actual prompts, operations, outputs and ratings under approved data policy.

Evidence: exact build and content manifest, inputs/assets/seeds, machine and network profiles, automated assertions, complete timings/traces, captured output and independent review where applicable. Both success and injected-failure routes must pass.

B11 — Accessibility and player journey#

Workstreams: GAME, EDIT, AUDIO, PLAT, SDK.

Complete installation, onboarding, settings, creation/play, captioned dialogue, mission completion, save/recovery and quit using keyboard, controller, touch and appropriate assistive modes. Cover focus, text scaling, contrast, reduced motion, caption timing, hearing alternatives and XR comfort.

Evidence: exact build and content manifest, inputs/assets/seeds, machine and network profiles, automated assertions, complete timings/traces, captured output and independent review where applicable. Both success and injected-failure routes must pass.

B12 — Finished games and maintained service#

Workstreams: SHIP, SDK, VERIFY, PLAT, NET, SAVE, GAME.

Complete the flagship campaign and second game, 8-hour player sessions, 72-hour servers, real update/rollback drills and an authorized 30-day pilot. Maintain defect, support, crash, performance and compatibility records tied to exact builds.

Evidence: exact build and content manifest, inputs/assets/seeds, machine and network profiles, automated assertions, complete timings/traces, captured output and independent review where applicable. Both success and injected-failure routes must pass.

8. Planned work packages and requirement coverage#

These are scope-level work packages and task seeds, not records of created or completed board tasks. After scope approval, materialize them in the live Eve board under the proposed family lilith-v1-6, following the family’s current authoring policy. Read the current policy and existing tasks first. Keep platform infrastructure work in platform-ops with product-side dependency links. Do not create Markdown tracker checkboxes or falsely mark these rows shipped.

Each requirement below is a mandatory delivery item. It usually needs several claimable tasks: implementation, authoring/operations, integration, failure/recovery, platform verification and documentation/evidence. Split by independently reviewable behaviour, not arbitrary file counts. Preserve the requirement ID on every task and one explicit final acceptance task per requirement. A reuse task must demonstrate the reused implementation; a bridge cannot finish with a type map or mock.

Every task body must contain: problem and working outcome; exact requirement IDs; inspected starting paths; dependency keys and document revision pins; production implementation and integration; affected UI/typed operations; failure/migration/security considerations; executable verification commands and real fixtures; hardware and performance gates; evidence location; and exact done criteria. A task whose acceptance relies on hardware or approval that is absent remains open/parked with that actual reason.

The universal acceptance checklist applies to every requirement: code/content exists; integrated caller exercised; applicable UI/API/agent operation works; failure/recovery handled; target profiles verified; tests and documentation complete; evidence bound to the merged origin/main commit and Eve revisions; independent review done. Non-code requirements use approved decisions, inventory or review artefacts instead of a manufactured code commit.

WP-GOV — Programme, ownership and honest capability accounting#

Owner discipline: Product direction / architecture. Earliest phase: M0. Direct prerequisite packages: none. Integrated scenarios: M0 contract audit and final B12 release adjudication. Starting anchors: Eve release records; libs/oshun/navigation/src/release-scope.ts.

Requirement / task seed Required component evidence Integrated acceptance
LW-GOV-01 — Release contract TC-LW-GOV-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs M0 audit + B12 plus universal acceptance checklist
LW-GOV-02 — Repository evidence audit TC-LW-GOV-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs M0 audit + B12 plus universal acceptance checklist
LW-GOV-03 — Architecture and responsibility decisions TC-LW-GOV-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs M0 audit + B12 plus universal acceptance checklist
LW-GOV-04 — Resource and production plan TC-LW-GOV-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs M0 audit + B12 plus universal acceptance checklist
LW-GOV-05 — Scope and decision control TC-LW-GOV-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs M0 audit + B12 plus universal acceptance checklist

WP-CORE — One shipping native runtime#

Owner discipline: Maya runtime / Neith systems. Earliest phase: M1. Direct prerequisite packages: WP-GOV. Integrated scenarios: B01, B05, B08. Starting anchors: libs/maya/engine-core; libs/neith/core; libs/neith/hal; libs/neith/integration-maya.

Requirement / task seed Required component evidence Integrated acceptance
LW-CORE-01 — Executable composition TC-LW-CORE-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B05, B08 plus universal acceptance checklist
LW-CORE-02 — Canonical world contracts TC-LW-CORE-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B05, B08 plus universal acceptance checklist
LW-CORE-03 — Scheduling and memory TC-LW-CORE-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B05, B08 plus universal acceptance checklist
LW-CORE-04 — Lifecycle and recovery TC-LW-CORE-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B05, B08 plus universal acceptance checklist
LW-CORE-05 — Shared semantic state TC-LW-CORE-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B05, B08 plus universal acceptance checklist

WP-GPU — Real graphics devices and render infrastructure#

Owner discipline: Neith graphics. Earliest phase: M2. Direct prerequisite packages: WP-CORE. Integrated scenarios: B01, B02. Starting anchors: libs/neith/hal/crates/neith-gpu; libs/neith/forge-core/crates/forge-render-wgpu; libs/maya/engine-core/crates/maya-renderer.

Requirement / task seed Required component evidence Integrated acceptance
LW-GPU-01 — Native backend execution TC-LW-GPU-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B02 plus universal acceptance checklist
LW-GPU-02 — Render graph and resource safety TC-LW-GPU-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B02 plus universal acceptance checklist
LW-GPU-03 — Shader and pipeline toolchain TC-LW-GPU-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B02 plus universal acceptance checklist
LW-GPU-04 — GPU observability and robustness TC-LW-GPU-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B02 plus universal acceptance checklist
LW-GPU-05 — Feature profiles and parity TC-LW-GPU-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B02 plus universal acceptance checklist

WP-GEO — Virtualized geometry and visibility#

Owner discipline: Neith rendering / Maya scene. Earliest phase: M3. Direct prerequisite packages: WP-GPU, WP-DATA. Integrated scenarios: B02, B03. Starting anchors: libs/maya/engine-core/crates/maya-meshlet; libs/maya/engine-core/crates/maya-vegetation; libs/neith.

Requirement / task seed Required component evidence Integrated acceptance
LW-GEO-01 — Virtual geometry pipeline TC-LW-GEO-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03 plus universal acceptance checklist
LW-GEO-02 — Skinned, foliage and translucent geometry TC-LW-GEO-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03 plus universal acceptance checklist
LW-GEO-03 — Visibility and occlusion TC-LW-GEO-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03 plus universal acceptance checklist
LW-GEO-04 — Authoring and diagnostics TC-LW-GEO-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03 plus universal acceptance checklist

WP-LIGHT — Lighting, shadows and reconstruction#

Owner discipline: Neith rendering. Earliest phase: M3. Direct prerequisite packages: WP-GPU, WP-DATA. Integrated scenarios: B02, B03. Starting anchors: libs/neith/gi-conformance; libs/neith; libs/maya/engine-core/crates/maya-renderer.

Requirement / task seed Required component evidence Integrated acceptance
LW-LIGHT-01 — Dynamic global illumination TC-LW-LIGHT-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03 plus universal acceptance checklist
LW-LIGHT-02 — Many lights and shadows TC-LW-LIGHT-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03 plus universal acceptance checklist
LW-LIGHT-03 — Temporal reconstruction TC-LW-LIGHT-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03 plus universal acceptance checklist
LW-LIGHT-04 — Reference rendering and look development TC-LW-LIGHT-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03 plus universal acceptance checklist
LW-LIGHT-05 — Scalable quality and power TC-LW-LIGHT-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03 plus universal acceptance checklist

WP-LOOK — Materials, atmosphere and effects#

Owner discipline: Rendering / technical art. Earliest phase: M3. Direct prerequisite packages: WP-GPU, WP-DATA. Integrated scenarios: B02, B03, B04. Starting anchors: libs/maya/engine-core/crates/maya-character-render; libs/maya/engine-core/crates/maya-atmosphere; libs/maya/engine-core/crates/maya-vfx.

Requirement / task seed Required component evidence Integrated acceptance
LW-LOOK-01 — Material authoring and shading TC-LW-LOOK-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03, B04 plus universal acceptance checklist
LW-LOOK-02 — Atmosphere and weather TC-LW-LOOK-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03, B04 plus universal acceptance checklist
LW-LOOK-03 — VFX and environmental interaction TC-LW-LOOK-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03, B04 plus universal acceptance checklist
LW-LOOK-04 — HDR and colour management TC-LW-LOOK-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03, B04 plus universal acceptance checklist
LW-LOOK-05 — Production art quality TC-LW-LOOK-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03, B04 plus universal acceptance checklist

WP-WORLD — Large worlds and coordinated streaming#

Owner discipline: Maya world / Neith I/O. Earliest phase: M3. Direct prerequisite packages: WP-CORE, WP-GPU, WP-DATA. Integrated scenarios: B03, B05, B06, B08. Starting anchors: libs/maya/engine-core/crates/maya-streaming; libs/maya/engine-core/crates/maya-world; libs/maya/engine-core/crates/maya-vram-budget.

Requirement / task seed Required component evidence Integrated acceptance
LW-WORLD-01 — World partition and precision TC-LW-WORLD-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08 plus universal acceptance checklist
LW-WORLD-02 — Coordinated resource streaming TC-LW-WORLD-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08 plus universal acceptance checklist
LW-WORLD-03 — World creation and procedural systems TC-LW-WORLD-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08 plus universal acceptance checklist
LW-WORLD-04 — World layers and collaboration TC-LW-WORLD-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08 plus universal acceptance checklist
LW-WORLD-05 — Streaming correctness under failure TC-LW-WORLD-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08 plus universal acceptance checklist

WP-ANIM — Character movement, acting and animation tools#

Owner discipline: Maya animation / technical animation. Earliest phase: M3. Direct prerequisite packages: WP-CORE, WP-DATA, WP-GPU. Integrated scenarios: B04, B05. Starting anchors: libs/maya/engine-core/crates/maya-motion-matching; libs/maya/engine-core/crates/maya-ml-locomotion; libs/maya/engine-core/crates/maya-facial-capture.

Requirement / task seed Required component evidence Integrated acceptance
LW-ANIM-01 — Animation graph and authoring TC-LW-ANIM-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B04, B05 plus universal acceptance checklist
LW-ANIM-02 — Motion matching and learned locomotion TC-LW-ANIM-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B04, B05 plus universal acceptance checklist
LW-ANIM-03 — Retargeting and contact TC-LW-ANIM-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B04, B05 plus universal acceptance checklist
LW-ANIM-04 — Facial performance and dialogue TC-LW-ANIM-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B04, B05 plus universal acceptance checklist
LW-ANIM-05 — Physical animation and recovery TC-LW-ANIM-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B04, B05 plus universal acceptance checklist
LW-ANIM-06 — Animals, mounts and interactions TC-LW-ANIM-06: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B04, B05 plus universal acceptance checklist

WP-PHYS — Physics, vehicles and material simulation#

Owner discipline: Maya simulation / Neith physics substrate. Earliest phase: M3. Direct prerequisite packages: WP-CORE, WP-DATA. Integrated scenarios: B03, B04, B05, B08. Starting anchors: libs/maya/engine-core/crates/maya-physics; libs/maya/engine-core/crates/maya-vehicles; libs/maya/engine-core/crates/maya-chemistry.

Requirement / task seed Required component evidence Integrated acceptance
LW-PHYS-01 — Rigid bodies and collision TC-LW-PHYS-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B04, B05, B08 plus universal acceptance checklist
LW-PHYS-02 — Destruction and deformable terrain TC-LW-PHYS-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B04, B05, B08 plus universal acceptance checklist
LW-PHYS-03 — Cloth, hair and soft simulation TC-LW-PHYS-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B04, B05, B08 plus universal acceptance checklist
LW-PHYS-04 — Water, fluids and reactive materials TC-LW-PHYS-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B04, B05, B08 plus universal acceptance checklist
LW-PHYS-05 — Vehicles and transport TC-LW-PHYS-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B04, B05, B08 plus universal acceptance checklist
LW-PHYS-06 — Simulation contracts and debug tools TC-LW-PHYS-06: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B04, B05, B08 plus universal acceptance checklist

WP-LIFE — Persistent living-world simulation#

Owner discipline: Maya AI / world design. Earliest phase: M4. Direct prerequisite packages: WP-WORLD, WP-PHYS, WP-ANIM. Integrated scenarios: B03, B05, B06, B08. Starting anchors: libs/maya/engine-core/crates/maya-npc-life; libs/maya/engine-core/crates/maya-city-sim; libs/maya/engine-core/crates/maya-sim-lod.

Requirement / task seed Required component evidence Integrated acceptance
LW-LIFE-01 — Perception and navigation TC-LW-LIFE-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08 plus universal acceptance checklist
LW-LIFE-02 — Routines and social life TC-LW-LIFE-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08 plus universal acceptance checklist
LW-LIFE-03 — Economy, justice and missions TC-LW-LIFE-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08 plus universal acceptance checklist
LW-LIFE-04 — Ecology and traffic TC-LW-LIFE-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08 plus universal acceptance checklist
LW-LIFE-05 — Simulation LOD continuity TC-LW-LIFE-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08 plus universal acceptance checklist
LW-LIFE-06 — World history and causal debugging TC-LW-LIFE-06: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08 plus universal acceptance checklist

WP-NPC — Real model-backed cognition#

Owner discipline: Maya cognition / shared model services. Earliest phase: M4. Direct prerequisite packages: WP-LIFE, WP-OPS. Integrated scenarios: B04, B05, B06, B10. Starting anchors: libs/maya/engine-core/crates/maya-slm-runtime; libs/maya/engine-core/crates/maya-cognition; libs/maya/engine-core/crates/maya-npc-planning.

Requirement / task seed Required component evidence Integrated acceptance
LW-NPC-01 — Production inference runtime TC-LW-NPC-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B04, B05, B06, B10 plus universal acceptance checklist
LW-NPC-02 — Grounded planning and constrained action TC-LW-NPC-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B04, B05, B06, B10 plus universal acceptance checklist
LW-NPC-03 — Memory and character consistency TC-LW-NPC-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B04, B05, B06, B10 plus universal acceptance checklist
LW-NPC-04 — Latency and graceful fallback TC-LW-NPC-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B04, B05, B06, B10 plus universal acceptance checklist
LW-NPC-05 — Evaluation and voice integration TC-LW-NPC-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B04, B05, B06, B10 plus universal acceptance checklist

WP-GAME — Reusable gameplay and narrative framework#

Owner discipline: Maya gameplay / game design. Earliest phase: M4. Direct prerequisite packages: WP-LIFE, WP-ANIM, WP-PHYS, WP-AUDIO. Integrated scenarios: B05, B08, B10, B11, B12. Starting anchors: libs/maya/engine-core/crates/maya-gameplay-systems; libs/maya/engine-core/crates/maya-combat; libs/maya/engine-core/crates/maya-investigation.

Requirement / task seed Required component evidence Integrated acceptance
LW-GAME-01 — Gameplay framework TC-LW-GAME-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B05, B08, B10, B11, B12 plus universal acceptance checklist
LW-GAME-02 — Player systems TC-LW-GAME-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B05, B08, B10, B11, B12 plus universal acceptance checklist
LW-GAME-03 — Narrative and mission authoring TC-LW-GAME-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B05, B08, B10, B11, B12 plus universal acceptance checklist
LW-GAME-04 — UI, input and accessibility TC-LW-GAME-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B05, B08, B10, B11, B12 plus universal acceptance checklist
LW-GAME-05 — Content quality and audience TC-LW-GAME-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B05, B08, B10, B11, B12 plus universal acceptance checklist

WP-AUDIO — Spatial audio, dialogue and music#

Owner discipline: Maya audio / shared voice and music. Earliest phase: M3. Direct prerequisite packages: WP-CORE, WP-DATA. Integrated scenarios: B03, B04, B05, B08, B11. Starting anchors: libs/maya/engine-core/crates/maya-audio; libs/maya/engine-core/crates/maya-acoustics; libs/maya/engine-core/crates/maya-music-engine.

Requirement / task seed Required component evidence Integrated acceptance
LW-AUDIO-01 — Production audio runtime TC-LW-AUDIO-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B04, B05, B08, B11 plus universal acceptance checklist
LW-AUDIO-02 — Spatial acoustics TC-LW-AUDIO-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B04, B05, B08, B11 plus universal acceptance checklist
LW-AUDIO-03 — Dialogue and voice TC-LW-AUDIO-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B04, B05, B08, B11 plus universal acceptance checklist
LW-AUDIO-04 — Interactive music and ambience TC-LW-AUDIO-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B04, B05, B08, B11 plus universal acceptance checklist
LW-AUDIO-05 — Audio tools and accessibility TC-LW-AUDIO-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B04, B05, B08, B11 plus universal acceptance checklist

WP-DATA — Assets, builds and interchange#

Owner discipline: Asset pipeline / Bellona. Earliest phase: M2. Direct prerequisite packages: WP-CORE. Integrated scenarios: B01, B07, B09. Starting anchors: libs/neith/engine-interchange; libs/neith/integration-maya; libs/bellona; libs/isis/3d-post-pipeline.

Requirement / task seed Required component evidence Integrated acceptance
LW-DATA-01 — Canonical asset graph TC-LW-DATA-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09 plus universal acceptance checklist
LW-DATA-02 — Real import and export TC-LW-DATA-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09 plus universal acceptance checklist
LW-DATA-03 — Reimport and artistic control TC-LW-DATA-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09 plus universal acceptance checklist
LW-DATA-04 — Cook, cache and packaging TC-LW-DATA-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09 plus universal acceptance checklist
LW-DATA-05 — Patch and compatibility TC-LW-DATA-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09 plus universal acceptance checklist
LW-DATA-06 — Migration from existing engines TC-LW-DATA-06: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09 plus universal acceptance checklist

WP-EDIT — A complete creator editor#

Owner discipline: Maya tools / Lilith creator experience. Earliest phase: M3. Direct prerequisite packages: WP-CORE, WP-GPU, WP-DATA, WP-OPS. Integrated scenarios: B01, B07, B10, B11. Starting anchors: libs/maya; libs/neith/integration-maya; Existing approved studio shell; final host chosen by architecture ADR.

Requirement / task seed Required component evidence Integrated acceptance
LW-EDIT-01 — Workspace and visual design TC-LW-EDIT-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B10, B11 plus universal acceptance checklist
LW-EDIT-02 — Scene manipulation and play mode TC-LW-EDIT-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B10, B11 plus universal acceptance checklist
LW-EDIT-03 — Specialist authoring tools TC-LW-EDIT-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B10, B11 plus universal acceptance checklist
LW-EDIT-04 — Transactions and recovery TC-LW-EDIT-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B10, B11 plus universal acceptance checklist
LW-EDIT-05 — Real collaboration TC-LW-EDIT-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B10, B11 plus universal acceptance checklist
LW-EDIT-06 — Creator journeys and help TC-LW-EDIT-06: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B10, B11 plus universal acceptance checklist

WP-SCRIPT — Programming and visual scripting#

Owner discipline: Maya developer experience. Earliest phase: M3. Direct prerequisite packages: WP-CORE, WP-DATA, WP-OPS. Integrated scenarios: B01, B07, B09, B10. Starting anchors: libs/maya/engine-core/crates/maya-reflect; libs/maya/engine-core/crates/maya-plugin-api; Editor scripting implementation selected by ADR.

Requirement / task seed Required component evidence Integrated acceptance
LW-SCRIPT-01 — Supported programming surfaces TC-LW-SCRIPT-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09, B10 plus universal acceptance checklist
LW-SCRIPT-02 — Visual gameplay scripting TC-LW-SCRIPT-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09, B10 plus universal acceptance checklist
LW-SCRIPT-03 — Debugger and live iteration TC-LW-SCRIPT-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09, B10 plus universal acceptance checklist
LW-SCRIPT-04 — Plugin compatibility and sandbox TC-LW-SCRIPT-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09, B10 plus universal acceptance checklist
LW-SCRIPT-05 — Developer tooling quality TC-LW-SCRIPT-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09, B10 plus universal acceptance checklist

WP-OPS — One typed operation system for humans and agents#

Owner discipline: Shared agent platform / Maya tools. Earliest phase: M2. Direct prerequisite packages: WP-CORE. Integrated scenarios: B01, B07, B10. Starting anchors: libs/bellona/mcp-gateway; Existing V1.3 agent and V1.5 studio operation services; Maya editor operation adapters.

Requirement / task seed Required component evidence Integrated acceptance
LW-OPS-01 — Unified editable operations TC-LW-OPS-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B10 plus universal acceptance checklist
LW-OPS-02 — Granular artistic control TC-LW-OPS-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B10 plus universal acceptance checklist
LW-OPS-03 — Isolation and authority TC-LW-OPS-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B10 plus universal acceptance checklist
LW-OPS-04 — Plans, previews and checkpoints TC-LW-OPS-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B10 plus universal acceptance checklist
LW-OPS-05 — Measured autonomous creation TC-LW-OPS-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B10 plus universal acceptance checklist

WP-NET — Production multiplayer and online worlds#

Owner discipline: Maya networking / online services. Earliest phase: M4. Direct prerequisite packages: WP-CORE, WP-WORLD, WP-PHYS, WP-GAME. Integrated scenarios: B05, B08, B12. Starting anchors: libs/maya/engine-core/crates/maya-nexus; libs/neith/integration-maya/src/metaverse-platform.

Requirement / task seed Required component evidence Integrated acceptance
LW-NET-01 — Authoritative networking TC-LW-NET-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B05, B08, B12 plus universal acceptance checklist
LW-NET-02 — Responsive simulation TC-LW-NET-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B05, B08, B12 plus universal acceptance checklist
LW-NET-03 — World lifecycle and scale TC-LW-NET-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B05, B08, B12 plus universal acceptance checklist
LW-NET-04 — Player services TC-LW-NET-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B05, B08, B12 plus universal acceptance checklist
LW-NET-05 — Security and abuse resistance TC-LW-NET-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B05, B08, B12 plus universal acceptance checklist
LW-NET-06 — True network measurement TC-LW-NET-06: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B05, B08, B12 plus universal acceptance checklist

WP-SAVE — Persistence and world recovery#

Owner discipline: Maya persistence / shared storage. Earliest phase: M4. Direct prerequisite packages: WP-CORE, WP-DATA, WP-WORLD. Integrated scenarios: B03, B05, B06, B08, B09, B12. Starting anchors: libs/maya/engine-core/crates/maya-save-system; libs/maya/engine-core/crates/maya-world-history.

Requirement / task seed Required component evidence Integrated acceptance
LW-SAVE-01 — Versioned saves and checkpoints TC-LW-SAVE-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08, B09, B12 plus universal acceptance checklist
LW-SAVE-02 — Crash consistency TC-LW-SAVE-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08, B09, B12 plus universal acceptance checklist
LW-SAVE-03 — Offline and cloud continuity TC-LW-SAVE-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08, B09, B12 plus universal acceptance checklist
LW-SAVE-04 — Server persistence and replay TC-LW-SAVE-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B03, B05, B06, B08, B09, B12 plus universal acceptance checklist

WP-PLAT — Shipping platform coverage#

Owner discipline: Platform product integration / platform-ops. Earliest phase: M5. Direct prerequisite packages: WP-GPU, WP-DATA, WP-CORE. Integrated scenarios: B01, B09, B11, B12. Starting anchors: Maya player/server packaging; Existing platform build and release systems.

Requirement / task seed Required component evidence Integrated acceptance
LW-PLAT-01 — Desktop clients and servers TC-LW-PLAT-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B09, B11, B12 plus universal acceptance checklist
LW-PLAT-02 — Browser and mobile TC-LW-PLAT-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B09, B11, B12 plus universal acceptance checklist
LW-PLAT-03 — XR TC-LW-PLAT-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B09, B11, B12 plus universal acceptance checklist
LW-PLAT-04 — Consoles TC-LW-PLAT-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B09, B11, B12 plus universal acceptance checklist
LW-PLAT-05 — Release engineering and operations TC-LW-PLAT-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B09, B11, B12 plus universal acceptance checklist
LW-PLAT-06 — Compatibility lifecycle TC-LW-PLAT-06: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B09, B11, B12 plus universal acceptance checklist

WP-SDK — Engine product, ecosystem and adoption#

Owner discipline: Developer experience / product. Earliest phase: M5. Direct prerequisite packages: WP-EDIT, WP-SCRIPT, WP-DATA, WP-PLAT. Integrated scenarios: B01, B07, B09, B11, B12. Starting anchors: Maya public SDK; Eve documentation and examples; Existing package registry and rights services.

Requirement / task seed Required component evidence Integrated acceptance
LW-SDK-01 — Public SDK and documentation TC-LW-SDK-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09, B11, B12 plus universal acceptance checklist
LW-SDK-02 — Templates and sample projects TC-LW-SDK-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09, B11, B12 plus universal acceptance checklist
LW-SDK-03 — Packages, plugins and mods TC-LW-SDK-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09, B11, B12 plus universal acceptance checklist
LW-SDK-04 — Creator distribution and business readiness TC-LW-SDK-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09, B11, B12 plus universal acceptance checklist
LW-SDK-05 — External validation and support TC-LW-SDK-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B01, B07, B09, B11, B12 plus universal acceptance checklist

WP-RND — Cutting-edge methods that earn production use#

Owner discipline: Rendering / animation research. Earliest phase: M5. Direct prerequisite packages: WP-LIGHT, WP-GEO, WP-ANIM, WP-LOOK. Integrated scenarios: B02, B10. Starting anchors: libs/neith research/rendering crates; libs/maya/engine-core/crates/maya-ml-locomotion.

Requirement / task seed Required component evidence Integrated acceptance
LW-RND-01 — Neural texture compression TC-LW-RND-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B10 plus universal acceptance checklist
LW-RND-02 — Neural radiance caching TC-LW-RND-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B10 plus universal acceptance checklist
LW-RND-03 — Learned animation and reconstruction TC-LW-RND-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B10 plus universal acceptance checklist
LW-RND-04 — Capture and novel representations TC-LW-RND-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B10 plus universal acceptance checklist
LW-RND-05 — Research-to-product discipline TC-LW-RND-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B10 plus universal acceptance checklist

WP-VERIFY — Proof, diagnostics and release quality#

Owner discipline: QA / performance / independent review. Earliest phase: M1. Direct prerequisite packages: WP-GOV. Integrated scenarios: B02, B03, B06, B12. Starting anchors: libs/maya/engine-core/project.json; libs/maya/engine-core/crates/maya-integration-tests; libs/neith/gi-conformance; Existing repository automation.

Requirement / task seed Required component evidence Integrated acceptance
LW-VERIFY-01 — Full verification topology TC-LW-VERIFY-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03, B06, B12 plus universal acceptance checklist
LW-VERIFY-02 — Correlated observability TC-LW-VERIFY-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03, B06, B12 plus universal acceptance checklist
LW-VERIFY-03 — Realistic test scenes and failure injection TC-LW-VERIFY-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03, B06, B12 plus universal acceptance checklist
LW-VERIFY-04 — Fair competitive assessment TC-LW-VERIFY-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03, B06, B12 plus universal acceptance checklist
LW-VERIFY-05 — Usability, accessibility and craft review TC-LW-VERIFY-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03, B06, B12 plus universal acceptance checklist
LW-VERIFY-06 — Sustained reliability and release evidence TC-LW-VERIFY-06: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B02, B03, B06, B12 plus universal acceptance checklist

WP-SHIP — Flagship game and final platform demonstration#

Owner discipline: Game production / programme. Earliest phase: M6. Direct prerequisite packages: WP-GAME, WP-NPC, WP-NET, WP-SAVE, WP-SDK, WP-RND, WP-PLAT, WP-VERIFY. Integrated scenarios: B12. Starting anchors: New flagship project using public Maya SDK; Second independent project; Eve release evidence.

Requirement / task seed Required component evidence Integrated acceptance
LW-SHIP-01 — Maya Living World flagship TC-LW-SHIP-01: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B12 plus universal acceptance checklist
LW-SHIP-02 — Cross-system consequence proof TC-LW-SHIP-02: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B12 plus universal acceptance checklist
LW-SHIP-03 — Independent second game TC-LW-SHIP-03: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B12 plus universal acceptance checklist
LW-SHIP-04 — Full production delivery TC-LW-SHIP-04: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B12 plus universal acceptance checklist
LW-SHIP-05 — Final release adjudication TC-LW-SHIP-05: exercise every implementation and acceptance clause in the PRD, including failures and real inputs B12 plus universal acceptance checklist

9. Verification execution and evidence schema#

Product workers must extend and run every relevant existing automated avenue: Rust formatting/clippy/check/tests, TypeScript lint/typecheck/unit tests, contract/schema/migration checks, asset-corpus tests, integration/build/package tests, GPU validation, visual/temporal tests, fuzz/security checks, performance, accessibility and full end-to-end journeys. Deep Playwright is required for user-facing web/editor flows where automatable; use suitable native UI, mobile and XR automation for their equivalents. A changed route includes the required code-owned route matrix entry.

First replace the narrow Maya default-target assumption with an accurate shipping-component matrix. Do not hide excluded crates or feature combinations behind a green default target. Hardware jobs must identify the real backend and reject software/mocks for hardware gates. Full CPU/GPU/network/world scenarios run on dedicated appropriate hosts, not unbounded parallel commands on Elikem’s Mac. Respect the repository resource checks, concurrency limits and platform-ops ownership; missing test setup is implementation work.

Each evidence record contains: requirement IDs; scenario/test IDs; task keys; full commit SHA and proof it is on origin/main; build and content hashes; Eve document revisions; model/dataset versions if relevant; exact commands; compiler/OS/driver/hardware; power/thermal settings; seeds and input logs; cold/warm state; quality/resolution/profile; real population/client counts; raw samples and failures; aggregate p50/p95/p99 and hitches; memory/network/power measurements; screenshots/video/traces; expected thresholds; result and limitations; independent reviewer identity and decision. Retain raw artefacts, not screenshots of a success badge.

No code is implemented by this scoping task. Validation of the scope itself includes unique requirement IDs, complete work-package coverage, nonempty implementation/acceptance clauses, acyclic known dependencies, all workstreams connected to integrated scenarios or governance gates, and exact Eve readback. This does not claim to have run the future engine tests.

10. Risks, decisions and ownership of blockers#

  • Architecture choices: scripting runtime, renderer/backend reuse, ECS/job ownership, editor host, physics substrate and plugin ABI are M0 ADR decisions with executable spikes; do not add competing implementations by default.
  • Hardware/platform access: identify every missing GPU, phone, XR headset, console SDK/devkit and certification route. Elikem decides purchases/agreements; product workers never infer permission or credentials. Requirements stay visible until satisfied or explicitly amended.
  • Content and quality: capture/rigging/animation, environment art, mission design, writing and sound need production work and independent review. Agent roles help execute but do not remove disciplines or make quality automatic.
  • Model/asset rights: validate provenance and licence conditions for real datasets, models, voices, plugins and sample projects. A successful technical integration alone does not authorize distribution.
  • Shared services: inventory live V1.3–V1.5/Isis/content-workbenches dependencies and assign exactly one owner. Preserve prior approvals and publish migration/compatibility evidence.
  • Performance: schedule combined scenes early. Individual subsystem wins do not guarantee final frame time, thermal limits, memory, population or network scale.
  • External validation: recruit independent creators/reviewers only through authorized arrangements. If unavailable, the evidence gate remains pending; self-review is not a substitute.
  • Planning: estimate optimistic/likely/conservative effort after measured spikes, then resource-level the dependency graph. No unsupported deadline or assumption that 126 requirements are 126 small tasks. This is a substantial engine-and-content programme, not a short point-release patch.

11. Release decision and content workflow#

Review the PRD and this contract together. The PRD is the normative product commitment; this specification operationalizes it. Approve the PRD first, then pin this document to its independently approved revision before approving the specification; this prevents a circular approval dependency. Required architecture ADRs are separately approved before dependent implementation.

Once approved, create the live board tasks, verify cross-family prerequisites, record an actual estimate and update the release feature/brand records through revision-checked Eve edits. Ready tasks must have concrete dependencies and acceptance, not an unapproved architecture assumption. Existing automatic-worker/reviewer switches remain under their owners; scope creation changes none of them.

A component task can close when its full scoped behaviour and verification are complete. The release acceptance tasks stay open until every component is integrated, all profiles and final workloads pass, both projects are finished, and the authorized pilot has supplied production evidence. Final review explicitly reports unmet requirements, exceptions and claim limits. Elikem owns final product/architecture approval and publication decisions; the author does not approve its own work.

The final state ledger distinguishes draft scope, approved scope, implementation underway, integrated candidate, verified candidate, authorized pilot, production-proven release and published/promoted versions. No transition is inferred from document creation, task count, model confidence or a successful unit-test run.

12. Coverage of the preceding Unreal/RAGE gap analysis#

This crosswalk keeps every prior recommendation inside the V1.6 scope. The detailed PRD clauses remain authoritative; a row is not satisfied merely because a similarly named crate exists.

Prior analysis area Mandatory requirement families Final proof
Runtime ownership and real Maya–Neith composition GOV, CORE Executable composition and B01/B05
Native graphics, real devices and scheduling GPU, CORE GPU captures, multi-vendor runs, B02
Virtualized geometry and production lighting GEO, LIGHT Dense/moving corpus and frame/image gates
Materials, atmosphere, weather, HDR and VFX LOOK, LIGHT Day/night/weather/material sequences in B02/B03
Coordinated large-world streaming WORLD, DATA, SAVE Fast traversal, teleport and faults in B03
Animation, motion matching, acting and physical characters ANIM, PHYS, NPC Real assets and held-out B04 sequences
Physics, vehicles, destruction and reactive materials PHYS, WORLD Gameplay-coupled B03/B05 and recovery
Believable living society, ecology and simulation LOD LIFE, CORE Real population B06 and persistent B05
Editor, authoring, collaboration and creative control EDIT, SCRIPT, OPS Independent creator and eight-client B07
Real asset interchange, reimport, cooking and patches DATA, SAVE File corpus B01 and interruption B09
Authoritative multiplayer and measured server scale NET 32/128 real-client B08 and 72-hour soak
Spatial audio, dialogue and adaptive music AUDIO, NPC Listening, timing and scene coherence B04/B05
CPU/GPU/network/I/O diagnostics and full verification VERIFY, GPU, CORE Correlated raw traces for every benchmark
Desktop, mobile, web, XR, console and release support PLAT, DATA Physical-device matrix and B09/B11
SDK, public APIs, packages, plugins, mods and support SDK, SCRIPT Second game, extension and B12 pilot
Measured AI-assisted and autonomous creation OPS, EDIT, SCRIPT Held-out real-operation B10 briefs
Real NPC inference, memory and constrained decisions NPC, LIFE Offline model runs and 500-case B10 suite
Shared semantic world model and coherent consequences CORE, LIFE, SAVE Market-stall B05, histories and migrations
Neural compression/radiance, learned methods and capture RND, LIGHT, ANIM Working selectable modes and comparative results
Fair benchmarks and maturity/evidence distinctions GOV, VERIFY Frozen corpus and reproducible report
Full flagship, second project and maintained production SHIP, SDK B12 plus full content and pilot contract
Art, narrative, sound and player-facing craft LOOK, GAME, ANIM, AUDIO, SHIP Independent motion/play/creator reviews
Discipline coverage, realistic resources and timing GOV Named accountable roles and measured delivery plan
Cross-release reuse and current canonical documentation GOV, DATA, OPS Eve ownership map and dependency audit
Independent approval, incremental milestones and final gates GOV, VERIFY, SHIP Full requirement ledger and separate publication decisions