AugEngine Flagship Game: Research Brief

Published 2026-05-23 · last verified 2026-05-24 · source of truth in repo: AUGENGINE-FLAGSHIP-BRIEF.md

Question

What should be built to demonstrate AugEngine (formerly Forge) as a serious game engine, and what has to be fixed before that demonstration is believable?

Answer

Build a premium, 12–20 minute third-person narrative-action vertical slice, not a creator form and not another mechanics harness. It should make the player immediately feel a designed lead character, meet a fully authored NPC with persistent consequence, fight through one polished encounter, and complete one cinematic climax in a single dense location.

That is achievable as a direction. It is not ready to enter content-heavy production today because AugEngine's current repository still has foundational proof gaps: the 3D example required repair during this review, high-fidelity textured/material and facial rendering are incomplete or unproven, measured runtime performance is missing, and the engine rename migration is broadly uncommitted at the time of writing (since committed: blackpig-forge@3986b9f).

Verified Local Findings

Area	Evidence	Assessment
Engine package	`npm run build --workspace=@blackpig/augengine` completed successfully.	Implemented baseline
Engine tests	42 passing files / 442 passing tests. Generated `.js` alongside TS, so this is not necessarily 442 distinct behaviors.	Useful but qualify coverage
Native renderer	`WebGpuLiveRenderer` implementation exists in `packages/augengine/src/runtime/live/webgpu/renderer.ts`.	Implemented in source
3D example	Initial build failed due to TSX configuration and undefined identifiers. Browser inspection uncovered stale generated JS shadowing current TS, renamed asset URLs pointing at absent files, and unhandled pointer-lock rejection. Repaired, rebuilt successfully, clean deployed run captured.	Functioning harness; visible primitive/blockout quality fails flagship bar
Facial fidelity	Renderer source says blendshape weights are staged while morph-target shader consumption ships in a follow-up.	Blocking gap for finished talking NPC
Materials/assets	Fragment shaders do not sample textures yet; texture support waits for first PBR-with-albedo asset.	Blocking gap for premium look
Draw/perf tooling	One `drawIndexed` per entity, no GPU timestamp queries, perf results table pending.	Performance not demonstrated

What AAA Practice Actually Requires

High-end game quality is a pipeline and review discipline, not a label applied to a renderer. The minimum bar pulled from current authoritative practice:

Character movement needs authored animation density and runtime selection — motion capture, Motion Matching, trajectory queries, IK, debugging (Epic's Game Animation Sample; Naughty Dog's GDC talk on TLOU2).
A lead NPC who performs on camera needs facial performance output, not just a text brain (Epic's MetaHuman Animator).
NPC intelligence separates authored intent from spatial decision tools — cover, line of sight, attack positions queried from behavior logic (Epic's Environment Query System). Persistent memory enriches relationships but cannot replace encounter design.
Environments require content organization and streaming strategy. Even a browser slice needs loaded zones, visibility/LOD decisions, clean encounter ownership (Epic's World Partition).
Performance is designed, instrumented, and budgeted. At 60 FPS, the game receives 16.67 ms per frame — flagship cannot claim smoothness without recorded CPU/GPU measurements (Epic's Animation Optimization).
Browser delivery has serious potential but imposes proof obligations. WebGPU and glTF 2.0 are the standards; AugEngine must actually exercise textured PBR assets, animation/facial deformation, compression/streaming, and GPU profiling to benefit from them.

Flagship Game Recommendation

Format

Genre: third-person narrative action thriller.
Scope: one 12–20 minute vertical slice, replayable through a relationship choice.
Location: one rain-soaked elevated transit terminal and maintenance district — dense enough for traversal, stealth/combat, dialogue, and a cinematic escape.
Why this format: the camera constantly exposes protagonist animation, NPC face and body performance, materials, lighting, combat readability, sound, and memory-driven story consequence. Much stronger engine showcase than a form generator or top-down stress test.

Core Experience

The player enters an abandoned transit hub looking for a data core. A central NPC, trapped behind security systems and not yet trusted, can guide, misdirect, or betray the player. An earlier decision in dialogue or rescue changes the combat route and the final extraction scene. Memory must change a designed gameplay outcome, not just produce different chatter.

Fully Designed Main Character Bar

Defined silhouette, clothing/material sheet, face reference, color/material language, biography, motive, combat vocabulary, camera framing rules.
A production character asset: textured materials, skeleton, skinning, facial shapes, hair/cloth treatment, LODs, collisions, damage/readability states.
Authored movement: idle, walk/run/sprint, starts/stops/pivots, aim/strafe, crouch/cover, climb/vault, hit/death, interaction, combat actions — with IK and transition evaluation.
One polished introduction shot and one close cinematic conversation in which face, body, eyes, and audio are inspected at final camera distance.

Fully Designed NPC Bar

Motivation, secrets, relationship states, trust rules, vocal/performance direction, visual design, authored scene arc.
Same visible asset fidelity required of the MC for the close-up scene.
Navigation/perception/cover or interaction logic that works independent of an LLM.
Memory bounded to approved facts/events, mapped to explicit encounter branches: trusted NPC opens a safer route; distrusted NPC locks a door and forces an exposed fight.
Dialogue authored for critical beats; generative behavior may add optional flavor or recall but cannot control mission completion or produce unreviewed story claims.

Engine Gap Matrix

Priority	Gate	Why it matters	Done when
P0	Resolve and commit the Forge→AugEngine migration	New work on a 450-file dirty rename risks losing or confusing the engine baseline.	DONE 2026-05-24 (`blackpig-forge@3986b9f`)
P0	Repair and visually verify `examples/smoke-game`	It was the nearest live proof and exposed compile, route, stale-module, and deploy-handling defects.	DONE harness only — still a low-fidelity blockout, not target content
P0	Implement textured PBR asset rendering	Premium characters and locations cannot be shown with procedural-only materials.	glTF base color/normal/metallic-roughness asset renders correctly in WebGPU and is captured.
P0	Complete morph-target/facial render path	A serious NPC shown in dialogue requires visible facial animation.	Facial shapes driven from approved animation/audio input render in the cinematic close shot.
P1	Establish animation target system	State-machine primitives alone do not equal premium locomotion.	Authored movement set and evaluation scene meet motion/transition quality review.
P1	Establish NPC authored behavior layer	A think scheduler and memory adapter are not an encounter-capable companion.	NPC handles navigation/perception/interaction/combat with deterministic fallback.
P1	Run measured performance baseline	No FPS or frame budget claim is presently supported.	Real-GPU CPU/GPU frame captures recorded on named target devices and browser versions.
P2	Asset compression/streaming/LOD plan	Browser delivery must remain loadable and stable as fidelity rises.	Budgeted glTF/texture assets, load strategy, LOD and memory acceptance tests.

Production Sequence

Phase 0: Truth Pass and Engine Stabilization

Commit or deliberately isolate the existing rename migration. (done)
Keep the now-repaired 3D example as a regression harness, not as the target game.
Prove one textured, skinned, morph-capable character in the native WebGPU renderer.
Create real frame measurements; do not authorize game content scale before this pass.

Phase 1: Character Proof Room

Build a tiny neutral-lit test level, not the game location.
Import the finished-look target MC and NPC stand-in/first-pass assets.
Implement camera distances, locomotion review lane, interaction pose, and dialogue close-up proof.
Review material quality, deformation, facial performance, animation transitions, silhouette readability, and performance.

Phase 2: Encounter Slice

Build the terminal location as a modular, authored environment.
Implement traversal and one combat archetype.
Implement the NPC trust branch with deterministic mission outcomes and memory write/recall proof.
Add audio, VFX, HUD, subtitles, checkpoint/save/reload, and cinematic.

Phase 3: Showcase Candidate

Run golden-path and alternate-path capture.
Run real-GPU performance and compatibility testing.
Polish only what is visible in the slice; do not add empty levels or content volume until the slice meets the acceptance bar.

First Executable Milestone

The first build milestone should be called Character Proof Room, not a game title. It is complete when a native AugEngine browser build loads a textured lead character and one NPC, shows responsive traversal and a facially animated dialogue beat, records frame timing on real GPU hardware, and produces capture evidence. Until then, narrative levels and mass content are a distraction.

Sources

Local source evidence under blackpig-forge/: AUGENGINE-ARENA-ALPHA.md, AUGENGINE-GAME-BUILD-PLAN.md, SHIP-CHECKLIST.md, docs/WEBGPU-COMPAT-MATRIX.md, tools/perf-harness/RESULTS-2026-05-18.md, examples/smoke-game/src/main.ts, packages/augengine/src/runtime/live/webgpu/renderer.ts, packages/augengine/src/ai/persona.ts, packages/augengine/src/ai/npc-think-scheduler.ts.

External authoritative sources cited inline above (Epic Games / GDC Vault / Khronos / W3C).

Methodology

This brief is the output of a verified-evidence research pass on 2026-05-24. Methodology, verification log (timestamped build/test/browser-inspection results), and decision log are preserved in blackpig-forge/AUGENGINE-FLAGSHIP-RESEARCH-PLAN.md alongside the working-judgment draft (AUGENGINE-FLAGSHIP-DRAFT.md). The verification log records: @blackpig/augengine build + 442 tests passing; mayas-hive and arena-alpha builds passing; smoke-game repair (JSX config + undefined identifiers + stale-JS shadowing + pointer-lock SecurityError) and final clean browser-deploy capture; renderer morph + sampled-texture path documented incomplete; FPS table pending. Three artifacts (plan → draft → brief) ship together so any future reviewer can audit how this was researched.