aetherbound-guild/docs/presentation/PHASE6_PRODUCTION_CHECKPOINT_2026-08-12.md

Phase 6 Production Checkpoint

Date: 2026-08-12 Repository closure: f8f7bbb17595807fd67e7c25949e56817dd3f7f3 Stage: BOUNDED_PIPELINES_PROVEN_BULK_FROZEN

What Is Now Proven

Layer	Strongest evidence	Result
Visual direction	Owner review of UI06-05	Japanese 2D thin-line/cel-color actors, soft painted landscape and 10-14% ordinary-unit height accepted.
Canonical actor motion	Processed/agent reviewed	Recruit, ordinary enemy and boss identity/action rows exist; 52-frame boss set passed fixed R2.
Crowded battle composition	Owner direction accepted; repaired developer candidate	10 allies, 12 ordinary enemies, one boss, 16 frames/8 FPS/2 seconds; non-loop scheduling repaired in proof-02; independent R3 remains open.
Detached VFX	Owner direction accepted; independent fixed PASS at 8a52af8	THREAT, DAMAGE, PROTECT and TEMPEST_PHASE, 16 states at 128/256px; deterministic, green-clean and battle-scale reviewed.
Equipment inventory art	Owner direction accepted; processed/agent reviewed at 7ea5728	EQ-001..EQ-012, 24 assets; explicit component manifest, 70% normalization, 96/64px light/dark review.
Formal UI masters	Source/concept direction	Eight landscape page-family masters exist; final engine text, responsive behavior and Owner page-by-page acceptance remain open.

No engine runtime, gameplay implementation, packaging or release work was opened. Human understanding, fun, final visual acceptance and device evidence remain separate.

Owner Direction Review

On 2026-08-12, after reviewing the displayed repaired battle composition, VFX contact/actual-size material and equipment contact/96px/64px material, the Owner accepted the presented content and direction. This closes the direction decision needed before the next bounded proofs. It does not approve undisplayed page families, bulk asset production, final runtime art, gameplay understanding/fun, audio, device behavior, packaging or release.

Production Methods Locked By Evidence

Characters And Enemies

1. Define a stable identity request: silhouette, facing, camera, height, anchor, baseline, materials and prohibited drift.
2. Generate and accept one idle identity before action work.
3. Generate one compatible action family at a time, normally a 2x2 or 4x2 keyframe sheet. Never mix unrelated actions into a final atlas.
4. Extract frames deterministically, preserve source/task/cost, and inspect the moving sequence for anatomy, crop, pivot, baseline and transition continuity.
5. Record loop explicitly. Looping states may phase-offset; non-looping states run once, transition by a declared state table, and hold only a legal terminal pose. They must never use unconditional modulo playback.
6. Keep damage, targeting and phase VFX detached from body sprites.
7. Test accepted actors at real crowd scale before expanding the identity family.

Final runtime sprites may be packed into atlases only after curation. The raw AI sheet is a source, not the runtime sprite sheet.

Small Objects And Equipment

1. Batch only objects sharing camera, light, material scale and runtime size.
2. Keep a stable row/column ID map and removable color background.
3. Treat model grid placement as approximate. Fixed cells establish identity; a tracked component manifest establishes exact content bounds.
4. Normalize complete item sets to a fixed transparent envelope after component review. Never crop by guessed equal rectangles when straps, cloth or rope sit near cell boundaries.
5. Review at master size and at actual 96/64px on light and dark surfaces.

This method is appropriate for category batches of the 320 equipment items. It does not justify generating all categories before their behavior/VFX and UI bindings are tested.

VFX

• Semantic families use shape before hue and remain independent of actor art.
• Pure green is preferable when magenta would collide with purple/black line work; premultiplied resize and post-resize despill are mandatory.
• Ground-contact, contested-gap and boss-attached placement each have separate anchors. Effects may not cover faces, order marks, forecasts or HUD facts.
• Standard, reduced motion, reduced flash and low-power timing remain future manifest fields; static masters do not prove those modes.

Why Bulk Production Is Still Frozen

The current proofs establish a viable pipeline, not a complete library. Bulk generation would still multiply these unresolved risks:

1. the 42 Profession silhouettes have not proven modular variety at crowd size;
2. the 100-enemy regional families have not proven body-plan diversity;
3. eight environment masters have not proven editable parallax extraction;
4. equipment trigger VFX and worn/held overlay rules are not connected to the first twelve inventory masters;
5. formal UI masters still need Owner page-family review and real Chinese/English layout; and
6. battle repair proof-02 still needs a fresh independent fixed review.

Next Bounded Order

1. Run fixed R3 on battle proof-02 and fixed review on equipment recovery 06R.
2. Prove one Windglass Verge 16:9 environment master and deterministic sky, far, midground, playfield and foreground layer package, with no actors/UI/VFX.
3. Prove three materially different base-Profession identities at the accepted 10-14% crowd scale using shared body grammar without recolor-only variation.
4. Bind four of EQ-001..012 to detached behavior-read VFX in a preparation compare state and one battle consequence state.
5. Only after those gates pass, authorize category production waves with a manifest/checkpoint after every 12-20 objects or one compatible actor family.

Music, SFX, engine integration, runtime input, save behavior, packaging and release remain separate future Goals.