diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..4339028
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,15 @@
+# Unity generated folders
+/Library/
+/Temp/
+/Logs/
+/UserSettings/
+/MemoryCaptures/
+/Recordings/
+
+# Local OS/editor files
+.DS_Store
+*.swp
+*.swo
+
+# Local analysis outputs
+/local-evidence/
diff --git a/docs/character/2d-to-3d-character-workflow.md b/docs/character/2d-to-3d-character-workflow.md
new file mode 100644
index 0000000..2e85dc8
--- /dev/null
+++ b/docs/character/2d-to-3d-character-workflow.md
@@ -0,0 +1,187 @@
+# 2D To 3D Character Workflow
+
+Status: production workflow note for using 2D concept images as the starting
+point for Meshy/Maya/engine character creation.
+
+## Direct Answer
+
+Maya should not be treated as a one-click "2D character image to commercial 3D
+game character" converter.
+
+What Maya can do well:
+
+- place 2D concept art as image planes for modeling reference
+- trace silhouettes or hard-surface motifs into curves
+- extrude flat logos, shields, blades, reliefs, and symbols from curves
+- use height/displacement/normal-style images for surface relief
+- clean, retopo, UV, rig, skin, pose-test, and export the resulting model
+
+What should use AI/Meshy first:
+
+- rough 3D shape from one concept image
+- better rough 3D shape from multiple concept views
+- early comparison of several silhouettes before manual cleanup
+
+What must still be done by hand or at least human-reviewed:
+
+- top-down silhouette judgement
+- production topology around shoulders, elbows, knees, hands, cloth, and weapon
+- UV layout and material separation
+- skin weights
+- attack animation timing
+- game-camera readability
+- legal/IP similarity review
+
+## Practical Route For This Project
+
+Use 2D images as a gate before any paid 3D generation.
+
+1. Create 2D concept sheets for 3 hero candidates:
+   - `H07 White-Charcoal Judgement Axe`
+   - `H02 Black-Iron Pincer Knight`
+   - `H08 Navy-Lantern Threadbinder`
+2. Each sheet must include:
+   - black silhouette strip
+   - front view
+   - back view
+   - side or 3/4 view
+   - top-down gameplay readability crop
+   - weapon close-up
+   - material/value swatches
+   - explicit reject notes
+3. Run a 64-128 px thumbnail test before 3D generation.
+4. Use Meshy Image-to-3D or Multi-Image-to-3D for preview only.
+5. Bring the preview into Maya.
+6. Clean the mesh, split parts, fix weapon thickness, and verify top-down shape.
+7. Only then spend on refine, texture, rig, animation, or engine import.
+
+## 2D Sheet Requirements
+
+Every candidate sheet must answer these questions without text labels:
+
+- Which direction is the character facing?
+- What is the weapon?
+- What is the class fantasy?
+- Where is the head?
+- Where are the shoulders?
+- What one shape makes this character different from any other dark knight?
+- Does the silhouette survive when reduced to 96 px tall?
+
+If the answer is unclear, do not generate 3D.
+
+## Candidate 2D Prompts
+
+These prompts are for creating 2D concept sheets. They are intentionally not
+final game art prompts.
+
+### H07 White-Charcoal Judgement Axe
+
+```text
+dark fantasy top-down ARPG playable hero concept sheet, white-charcoal
+execution axe warrior, one oversized arm, one huge one-sided axe head wider
+than the shoulders, charcoal black body, white ash cracks, dark red cloth,
+clear black silhouette strip, front view, back view, side view, top-down
+gameplay readability crop, weapon close-up, material swatches, practical game
+character design, readable at small size
+```
+
+Reject:
+
+```text
+no generic barbarian, no viking helmet, no tiny axe, no realistic bodybuilder,
+no gore, no huge fur cloak, no unreadable black blob, no excessive ornamental
+detail
+```
+
+### H02 Black-Iron Pincer Knight
+
+```text
+dark fantasy top-down ARPG playable hero concept sheet, asymmetric black-iron
+pincer knight, one huge shoulder pincer silhouette, one-handed great shears
+weapon, short shield blade, rust red cloth, pale scar edge lines, clear black
+silhouette strip, front view, back view, side view, top-down gameplay
+readability crop, weapon close-up, material swatches, practical game character
+design, readable at small size
+```
+
+Reject:
+
+```text
+no literal crab body, no robot armor, no tiny scissors, no hidden weapon, no
+long cape covering legs, no excessive spikes, no merged hands, no extra arms,
+no gore
+```
+
+### H08 Navy-Lantern Threadbinder
+
+```text
+dark fantasy top-down ARPG playable hero concept sheet, navy-lantern
+threadbinder, small pale lantern floating above the head, black lacquer armor,
+dark navy cloth, visible thick magical thread loops around the hands, compact
+body silhouette, fingertip blade shapes, clear black silhouette strip, front
+view, back view, side view, top-down gameplay readability crop, material
+swatches, practical game character design, readable at small size
+```
+
+Reject:
+
+```text
+no puppet master cliche, no thin invisible threads, no dolls, no anime school
+outfit, no cyber wires, no wings, no face-detail focus, no huge robe blob
+```
+
+## Maya Use Cases By Asset Type
+
+| Asset Type | Best 2D Input | Maya Method | Meshy Useful? |
+| --- | --- | --- | --- |
+| Player hero | concept sheet with multiple views | image planes, blockout, cleanup, retopo, rig, skin | yes, preview only |
+| Enemy fodder | concept sheet or strong silhouette | model over reference, simplify forms | yes |
+| Boss | front/top silhouette plus attack-shape sheet | blockout in simple masses first | limited; cleanup heavy |
+| Weapon | side view and top view | curve/mesh modeling, thickness pass | sometimes |
+| Shield/emblem | vector/black shape | curve/extrude/bevel | rarely needed |
+| Floor relief | grayscale height/ornament map | displacement/normal/mesh relief | not needed |
+| UI icon | flat concept/vector | keep 2D or vector; do not force 3D | no |
+
+## Preview Pass/Fail
+
+Pass:
+
+- 96 px grayscale still reads as the same role.
+- Weapon silhouette is obvious.
+- Character has one dominant hook, not five small hooks.
+- Head and shoulders are separable from cape/body.
+- The design can be modeled without relying on hair-thin details.
+
+Fail:
+
+- It only works as a full-resolution illustration.
+- The weapon disappears in top-down view.
+- The pose is doing all the work.
+- Materials are all the same dark value.
+- It depends on floating particles or VFX to identify the character.
+- It looks like a known commercial character, class skin, or franchise mascot.
+
+## First Batch Decision
+
+Run 2D concept exploration first, not 3D generation.
+
+Batch size:
+
+- 3 hero candidates
+- 2 variations each
+- 6 concept sheets total
+
+Pick one for Meshy preview only after this review:
+
+1. H07 if the goal is strongest melee impact.
+2. H02 if the goal is strongest dark-fantasy identity.
+3. H08 if the goal is novelty, accepting higher production risk.
+
+No refine/rig/animation spend until a 2D sheet passes the tiny-camera test.
+
+## Sources
+
+- Autodesk Maya Help: https://help.autodesk.com/view/MAYAUL/2026/ENU/
+- Meshy Image to 3D API: https://docs.meshy.ai/en/api/image-to-3d
+- Meshy Multi-Image to 3D API: https://docs.meshy.ai/en/api/multi-image-to-3d
+- Meshy for Maya plugin: https://docs.meshy.ai/en/maya-plugin/introduction
diff --git a/docs/character/character-pattern-cards.md b/docs/character/character-pattern-cards.md
new file mode 100644
index 0000000..a64928a
--- /dev/null
+++ b/docs/character/character-pattern-cards.md
@@ -0,0 +1,322 @@
+# Character Pattern Cards
+
+Status: paused as final protagonist direction. Use these concepts as part
+library and risk evidence before spending Meshy/API credits.
+
+## Why This Exists
+
+The current Phase 1 character is not worth polishing as the final direction.
+However, the first generated concept round skewed too rugged: armored bodies,
+large weapons, and grim combat silhouettes are not enough to sell the game.
+
+The current product direction is `Threadlight Pilgrimage`; see
+`docs/product/game-definition-threadlight.md`. Character work should now start
+from role, world rule, hub, and emotional promise. Weapons are only supporting
+props.
+
+Preserve useful parts:
+
+- H09: hooded field-worker mass and relic weight.
+- F01: pale veil/crest and polearm readability.
+- F03: bell-root ecology and occult support shape.
+- H08: lantern/thread action language.
+- H02/H07: enemy, guard, elite, or boss mass.
+
+Do not adopt any generated sheet wholesale as the first hero.
+
+Before generating or modeling, compare several readable silhouettes at gameplay
+camera distance. Reject weak patterns early.
+
+Judgement order:
+
+1. black silhouette at tiny size
+2. grayscale value separation
+3. job tool, facing, and action readability from top-down
+4. attack pose and telegraph
+5. material identity
+6. only then texture detail
+
+## Hero Pattern Cards
+
+| ID | Concept | Silhouette Hook | Weapon | Material/Value Bands | Animation Identity | VFX Role | Reject If |
+| --- | --- | --- | --- | --- | --- | --- | --- |
+| H01 | Lantern-Coffin Pilgrim | Tall coffin box on back, one-shoulder cloak | Hooked spear with candle head | Black cloth, dark wood, dull gold, bright face area | Heavy steps, flame jumps only on attack | Amber; cleanse/heal | Angel wings, cross overload, generic paladin |
+| H02 | Black-Iron Pincer Knight | Asymmetric huge shoulder pincers | One-handed great shears plus short shield blade | Black iron, rust red, white scar lines | Side-stepping, opens wide before clamping | Red-black; bind/bleed | Literal crab/insect, robot read |
+| H03 | Ash-Glass Scout | Slim body, broken glass blade-fan on back | Reverse-grip twin daggers | Gray leather, transparent glass, silver edges | Low sliding motion, afterimage shatters after attack | Pale blue; dodge/critical | Ninja costume, cyber glow |
+| H04 | Mud-Crown Hexer | Huge clay crown, round body mass | Bone staff and hanging jar | Dry clay, bone white, black cords | Swaying chant, jar lags behind body | Green-brown; poison/summon | Real-world tribal mimicry, tiny symbols |
+| H05 | Silver-Needle Execution Medic | Long beak mask, needle bundle on back | Telescoping injection spear | Black leather, dull silver, dirty white cloth | Small footwork, body becomes a straight thrust line | Sick yellow-green; weaken/drain | Direct plague doctor copy, medical clutter |
+| H06 | Star-Eater Chain Monk | Circular back-chain halo, small head | Weighted chain ring | Black cloth, bronze, sparse white star dots | Constant visible circular orbit | Deep purple; pull/area control | Space wizard, zodiac ornament spam |
+| H07 | White-Charcoal Judgement Axe | One giant arm, axe head wider than shoulders | One-sided execution axe | Charcoal body, white ash cracks, dark red cloth | Long charge, kneels after slam | White ash; stun/shatter | Barbarian cliche, muscle-only design |
+| H08 | Navy-Lantern Threadbinder | Floating small lantern overhead, threads outward | Thread spool ring and fingertip blades | Navy cloth, black lacquer, pale lantern white | Body stays calm; threads move first | Blue-white; traps/remote cuts | Puppet-master cliche, invisible thin threads |
+| H09 | Veiled Oath Relic Duelist | Hooded mantle, heavy beads, broad reliquary greatblade | Broken cathedral-slab greatblade | Charcoal hood, slate mantle, bone-white blade edge, brass beads | Guarded forward stance, heavy sweeping cuts | Cyan seam plus iron censer sparks | Key-shaped weapon, star charm, black zipper coat, recognizable franchise read |
+
+## Heroine Pattern Cards
+
+| ID | Concept | Silhouette Hook | Weapon | Material/Value Bands | Animation Identity | VFX Role | Reject If |
+| --- | --- | --- | --- | --- | --- | --- | --- |
+| F01 | Ash-Veil Halberd Matron | Tall ash veil crest, broad shoulders | Crescent halberd with thick blade | Charcoal armor, ash-white veil, oxblood sash, dull brass | Guarded spear stance, deliberate sweeping control | Pale ash plus warm steel sparks | Bikini armor, generic valkyrie, tiny spear, face-detail focus |
+| F02 | Glass-Moth Duelist | Broken glass moth-wing back shape | Paired broad crescent daggers | Dark leather, pale glass-blue blade edges, muted teal cloth | Low lateral cuts, glass afterimage shatters | Pale blue crit/dodge shards | Ninja costume, fairy wings, thin daggers, cyber glow |
+| F03 | Bell-Root Hex Huntress | Broken bell shoulder mantle, root cords | Crescent sickle-staff with bronze bells | Black-green leather, bone mask crest, tarnished bronze | Trap-setting stance, bell-root summon gestures | Green-black roots plus bronze pulse | Sexy witch, pointed hat, tiny charms, cultural stereotype symbols |
+
+## Enemy Pattern Cards
+
+| Role | Concept | Top-Down Silhouette Hook | Telegraph | Color/Value Role | Audio Cue | Reject If |
+| --- | --- | --- | --- | --- | --- | --- |
+| Fodder | Soot-Split Crawler | Small black teardrop, one long claw side | Claw side glows pale and half-steps | Low value body, claw marks facing | Dry claw scrape | Generic zombie, detail noise |
+| Charger | Spineback Rammer | Thick triangle head and straight back-spine | Drops head, red line shadow appears | Dark red mid-value, high-contrast head | Low inhale into burst | Literal bull/boar, unreadable curved charge |
+| Ranged | Lamp-Eye Shellgun | Round body plus one long arm cannon | Cannon tip point-light, impact circle appears | Cold blue-white, projectile brightest | Short glass pluck | Human archer, bullet same value as body |
+| Summoner | Womb-Bell Conductor | Bell robe with small satellite shards | Shards stop spinning, summon sigil appears | Purple mid-value, pale gold sigil | Distant bell, reversed whisper | Wizard hat, unclear summon point |
+| Tank | Black-Shield Gravekeeper | Huge half-moon front shield, tiny rear legs | Shield rim lights, front cone hardens | Black shield, bright metal rim | Dull metal scrape | Omnidirectional invincible read, unclear shield facing |
+| Elite | Twin-Blade Shadow Priest | Thin cross shape, long blade sleeves left/right | Sleeves open, cross slash range appears | Black plus green poison light, bright blade tips | Cloth snap into metal hit | Ninja/reaper copy, instant attack |
+| Miniboss | Furnace-Back Execution Bug | Large oval body, glowing back furnace, axe legs | Furnace brightens in three stages, axe legs lift | Black body, orange furnace danger meter | Furnace hum, steam vent | Just a bigger fodder, hidden weak point |
+| Boss | Abyssal Crown Tree | Central crown trunk, radial root arms | Each root pulses white, then tears floor | Low-value trunk, white crown, red-black attack roots | Heartbeat plus heavy wood split | Generic tentacle boss, full-screen unreadable telegraphs |
+
+## Historical Hero Candidate Notes
+
+These notes were useful before the world direction was defined. They are no
+longer direct Meshy/Maya targets.
+
+### H02 Black-Iron Pincer Knight
+
+Use as enemy, guard, or heavy alternate silhouette if the game needs brutal
+melee identity.
+
+Pass condition:
+
+- pincer shoulder and shears read in a 96 px thumbnail
+- attack telegraph can open/close clearly
+- body does not become a crab/robot
+
+### H07 White-Charcoal Judgement Axe
+
+Use as boss, elite, or backup combat readability reference if the game needs
+heavy hit-stop and clear impact.
+
+Pass condition:
+
+- giant arm and axe head read as one bold silhouette
+- white ash cracks separate hero from floor
+- attack anticipation is readable in a still frame
+
+### H08 Navy-Lantern Threadbinder
+
+Use as threadlight tool reference. Do not keep the thin magical loops without a
+physical survey device.
+
+Pass condition:
+
+- floating lantern and thread field remain visible without becoming noise
+- thread attacks are thick enough to read
+- body stays distinguishable from trap/VFX layer
+
+## First Enemy Pack
+
+Use these only as historical combat-role references. The first Threadlight
+enemy set is defined in `docs/product/game-definition-threadlight.md`.
+
+1. `Soot-Split Crawler`: fodder; teaches basic hit/death.
+2. `Spineback Rammer`: charger; creates dodge pressure.
+3. `Black-Shield Gravekeeper`: tank; teaches facing/positioning.
+4. `Womb-Bell Conductor`: summoner; creates priority target.
+5. `Furnace-Back Execution Bug`: miniboss; creates trailer frame.
+
+Do not add all eight enemy patterns at once. Five is already enough for the
+first enemy readability test.
+
+## Historical Meshy Preview Prompt Patterns
+
+These prompts are archived. Do not run them as-is. New prompts must start from
+the Threadlight Surveyor, route tool, light-moth, or loom-gate language in
+`docs/product/game-definition-threadlight.md`.
+
+### Hero Preview: Black-Iron Pincer Knight
+
+```text
+dark fantasy top-down ARPG hero character, full body, centered, neutral A-pose,
+asymmetric black iron pincer knight, one huge shoulder pincer silhouette,
+one-handed great shears weapon, short shield blade, oversized gauntlets and
+boots, rust red cloth accents, pale scar edge lines, stylized realistic game
+character, clean separate armor shapes, strong readable silhouette for
+isometric camera, Maya-ready 3D asset
+```
+
+Reject:
+
+```text
+no literal crab body, no robot armor, no tiny scissors, no hidden weapon, no
+long cape hiding legs, no excessive spikes, no merged hands, no extra arms, no
+realistic face focus, no gore
+```
+
+Maya cleanup expected:
+
+- separate shears, shield blade, shoulders, body, and cloth
+- thicken the shears so they read from top-down
+- reduce tiny armor shards
+- add clear top-facing value bands
+
+### Hero Preview: White-Charcoal Judgement Axe
+
+```text
+dark fantasy top-down ARPG hero character, full body, centered, neutral A-pose,
+white-charcoal execution axe warrior, one oversized arm, huge one-sided axe
+head wider than shoulders, charcoal black body, white ash cracks, dark red
+cloth, heavy readable silhouette, stylized realistic proportions, broad upper
+body, game-ready 3D character, Maya-ready asset
+```
+
+Reject:
+
+```text
+no generic barbarian, no bodybuilder focus, no tiny axe, no symmetrical normal
+arms, no horned viking look, no long fur cape, no fused weapon hand, no gore
+```
+
+Maya cleanup expected:
+
+- separate axe mesh and hand socket
+- exaggerate axe top plane and white cracks
+- simplify muscles into readable armor/body masses
+- pose-test charge, impact, and recovery silhouette
+
+### Hero Preview: Navy-Lantern Threadbinder
+
+```text
+dark fantasy top-down ARPG hero character, full body, centered, neutral A-pose,
+navy lantern threadbinder, small pale lantern floating above head, dark navy
+cloth, black lacquer armor pieces, visible thick magical thread loops around
+hands, compact body silhouette, finger blade shapes, stylized game character,
+readable from isometric camera, Maya-ready 3D asset
+```
+
+Reject:
+
+```text
+no puppet master cliche, no tiny invisible threads, no huge robe blob, no
+multiple dolls, no anime school outfit, no cyber wires, no wings, no face detail
+focus
+```
+
+Maya cleanup expected:
+
+- make thread loops separate, thick, and optional
+- lantern must stay above head without hiding body
+- split cloth layers so legs and facing remain clear
+- keep VFX/trap geometry separate from body mesh
+
+### Enemy Preview: Soot-Split Crawler
+
+```text
+dark top-down ARPG fodder enemy, small soot-black crawling creature, teardrop
+body silhouette, one side has one long pale claw, hunched low stance, simple
+large forms, readable facing direction from top-down, dark fantasy game enemy,
+Maya-ready 3D model
+```
+
+Reject:
+
+```text
+no generic zombie, no exposed gore, no many tiny fingers, no realistic human
+face, no fully black silhouette, no thin unreadable claw
+```
+
+### Enemy Preview: Spineback Rammer
+
+```text
+dark top-down ARPG charger enemy, thick triangular head, straight spine ridge
+on back, compact heavy body, dark red hide, high contrast head plate, low
+aggressive stance, readable charging silhouette, stylized dark fantasy creature,
+Maya-ready 3D asset
+```
+
+Reject:
+
+```text
+no literal bull, no boar copy, no realistic animal, no curved unreadable horns,
+no thin legs, no cluttered spikes
+```
+
+### Enemy Preview: Black-Shield Gravekeeper
+
+```text
+dark top-down ARPG shield tank enemy, huge half-moon black shield on front,
+small legs visible behind shield, gravekeeper armor, bright worn metal rim,
+clear front-facing silhouette, dark fantasy dungeon enemy, stylized game-ready
+3D character, Maya-ready asset
+```
+
+Reject:
+
+```text
+no full round turtle shell, no hero knight read, no shield covering entire body
+from all angles, no tiny shield, no ornate unreadable engraving
+```
+
+## Preview Spend Rules
+
+- One preview per pattern per review round.
+- No refine unless the preview passes 64-128 px readability.
+- No rig/animation until Maya cleanup scope is written.
+- Store task id, prompt, negative prompt, export format, credit cost, and
+  preview verdict in the asset ledger.
+- Kill patterns quickly. Weak silhouette is not fixed by texture quality.
+
+## Selection Matrix
+
+| Candidate | Differentiation | Gameplay Readability | Meshy Risk | Maya Cleanup Risk | First Slice Fit |
+| --- | --- | --- | --- | --- | --- |
+| H02 Pincer Knight | High | High | Medium | Medium | Strong melee identity |
+| H07 Judgement Axe | Medium | Very high | Low-medium | Low-medium | Strong hit-stop demo |
+| H08 Threadbinder | High | Medium | High | High | Unique but risky |
+| H09 Relic Duelist | High | High | Medium | Medium | Strong hooded relic-weapon identity |
+| F01 Halberd Matron | Medium-high | High | Low-medium | Medium | Strong female melee/control read |
+| F02 Glass Moth | Medium-high | Medium | High | Medium | Visually strong, risks thin-wing/ninja read |
+| F03 Bell Root | High | Medium-high | Medium | Medium-high | Strong caster/enemy-boss crossover |
+| H01 Lantern Pilgrim | Medium-high | Medium | Medium | Medium | Strong world identity |
+| H03 Glass Scout | Medium | Medium-low | High | Medium | Risks thin unreadable blades |
+| H04 Mud-Crown Hexer | Medium | Medium | Medium | Medium | Better as NPC/enemy first |
+| H05 Execution Medic | Medium | Medium | Medium | Medium | IP/cliche risk if too plague-doctor |
+| H06 Chain Monk | High | Medium | High | High | VFX-dependent, risky first hero |
+
+Historical recommendation after first concept-sheet generation round:
+
+1. H09 `Veiled Oath Relic Duelist`
+2. F01 `Ash-Veil Halberd Matron`
+3. H07 `White-Charcoal Judgement Axe`
+4. F03 `Bell-Root Hex Huntress`
+
+Keep H02, H08, and F02. Do not discard them:
+
+- H02 can become a heavy enemy elite, armor skin, or alternate melee class.
+- H08 can become a trap-focused class after thread/VFX readability is proven.
+- F02 can become a rogue class or enemy assassin, but avoid the ninja/fairy read.
+
+Do not generate Meshy previews from these sheets directly. First create the
+Threadlight world contact sheet and gameplay mock, then choose at most two
+assets that prove the new world rule.
+
+## Generated Concept Sheets
+
+Generated with built-in `image_gen` as pre-Meshy 2D concept exploration:
+
+- `docs/character/concepts/h07-white-charcoal-judgement-axe-sheet.png`
+- `docs/character/concepts/h02-black-iron-pincer-knight-sheet.png`
+- `docs/character/concepts/h08-navy-lantern-threadbinder-sheet.png`
+- `docs/character/concepts/h09-veiled-oath-relic-duelist-sheet.png`
+- `docs/character/concepts/f01-ash-veil-halberd-matron-sheet.png`
+- `docs/character/concepts/f02-glass-moth-duelist-sheet.png`
+- `docs/character/concepts/f03-bell-root-hex-huntress-sheet.png`
+- `docs/character/concepts/concept-comparison-contact-sheet.png`
+- `docs/character/concepts/concept-silhouette-strip.png`
+
+Current reuse verdict:
+
+| Candidate | Verdict | Reason |
+| --- | --- | --- |
+| H09 Relic Duelist | Part library | Keep hooded field-worker mass and relic weight; do not keep weapon-first identity. |
+| F01 Halberd Matron | Part library | Keep pale veil/crest and polearm readability; convert armor into field uniform. |
+| H07 Judgement Axe | Enemy/boss backup | Most readable impact/weight, but too rugged for first hero. |
+| F03 Bell Root | Strong region motif | Use bell-root ecology for NPC, elite enemy, or boss-adjacent caster. |
+| H02 Pincer Knight | Keep for reuse | Strong silhouette, but leans heavily armored/monster; better as elite or alternate class. |
+| H08 Threadbinder | Keep for later | Novel, but thread readability and production risk are high. |
+| F02 Glass Moth | Keep with caution | Good image, but wing/ninja/fairy read risks IP/generic drift and thin silhouette issues. |
diff --git a/docs/character/concepts/concept-comparison-contact-sheet.png b/docs/character/concepts/concept-comparison-contact-sheet.png
new file mode 100644
index 0000000..08414e3
Binary files /dev/null and b/docs/character/concepts/concept-comparison-contact-sheet.png differ
diff --git a/docs/character/concepts/concept-silhouette-strip.png b/docs/character/concepts/concept-silhouette-strip.png
new file mode 100644
index 0000000..67896be
Binary files /dev/null and b/docs/character/concepts/concept-silhouette-strip.png differ
diff --git a/docs/character/concepts/f01-ash-veil-halberd-matron-sheet.png b/docs/character/concepts/f01-ash-veil-halberd-matron-sheet.png
new file mode 100644
index 0000000..f21903b
Binary files /dev/null and b/docs/character/concepts/f01-ash-veil-halberd-matron-sheet.png differ
diff --git a/docs/character/concepts/f02-glass-moth-duelist-sheet.png b/docs/character/concepts/f02-glass-moth-duelist-sheet.png
new file mode 100644
index 0000000..5f4e329
Binary files /dev/null and b/docs/character/concepts/f02-glass-moth-duelist-sheet.png differ
diff --git a/docs/character/concepts/f03-bell-root-hex-huntress-sheet.png b/docs/character/concepts/f03-bell-root-hex-huntress-sheet.png
new file mode 100644
index 0000000..121d6c2
Binary files /dev/null and b/docs/character/concepts/f03-bell-root-hex-huntress-sheet.png differ
diff --git a/docs/character/concepts/generation-log.md b/docs/character/concepts/generation-log.md
new file mode 100644
index 0000000..61c8099
--- /dev/null
+++ b/docs/character/concepts/generation-log.md
@@ -0,0 +1,53 @@
+# Character Concept Generation Log
+
+Status: record of generated 2D concept sheets before any Meshy/Maya spend.
+
+Tool path:
+
+- Built-in `image_gen` tool.
+- Outputs copied from `$CODEX_HOME/generated_images/...` into
+  `docs/character/concepts/`.
+
+## IP Safety Notes
+
+The user provided a hooded reference image for mood. It was not used as a direct
+edit target. The derived H09 direction keeps only broad, generic ideas:
+
+- hooded anonymity
+- ritual weapon weight
+- occult pilgrim mood
+- heavy beads/charms as silhouette support
+
+Explicitly rejected from H09:
+
+- key-shaped weapon
+- star charm
+- black zipper coat
+- school-uniform silhouette
+- childlike proportions
+- recognizable franchise character design
+
+## Generated Sheets
+
+| File | Prompt Direction | Current Verdict |
+| --- | --- | --- |
+| `h07-white-charcoal-judgement-axe-sheet.png` | Oversized-arm execution axe hero. | Reuse as enemy/boss mass or impact reference; too rugged for first hero. |
+| `h02-black-iron-pincer-knight-sheet.png` | Asymmetric pincer knight with shears. | Keep: useful for elite/alternate class. |
+| `h08-navy-lantern-threadbinder-sheet.png` | Lantern and thick thread-loop trap hero. | Keep: novel but production-risky. |
+| `h09-veiled-oath-relic-duelist-sheet.png` | Hooded relic duelist with broad reliquary greatblade. | Part library: hooded field-worker mass and relic weight. |
+| `f01-ash-veil-halberd-matron-sheet.png` | Female halberd warrior with ash veil crest. | Part library: pale veil/crest and polearm readability. |
+| `f02-glass-moth-duelist-sheet.png` | Female glass-moth rogue with crescent daggers. | Keep with caution: strong art, risky ninja/fairy read. |
+| `f03-bell-root-hex-huntress-sheet.png` | Female bell-root occult hunter/summoner. | Strong alternate; could become heroine or elite enemy. |
+| `concept-comparison-contact-sheet.png` | Combined overview of all generated sheets. | Used for visual triage. |
+| `concept-silhouette-strip.png` | Cropped top silhouette strips in grayscale. | Used for tiny-readability triage. |
+
+## Next Gate
+
+Before Meshy preview:
+
+1. Accept or replace `docs/product/game-definition-threadlight.md`.
+2. Create a world contact sheet and gameplay mock.
+3. Choose no more than two assets that prove the world rule.
+4. Run only Meshy preview/draft first.
+5. Do not refine, HD texture, rig, or animate until the preview survives the
+   game-camera test.
diff --git a/docs/character/concepts/h02-black-iron-pincer-knight-sheet.png b/docs/character/concepts/h02-black-iron-pincer-knight-sheet.png
new file mode 100644
index 0000000..a7a4b2d
Binary files /dev/null and b/docs/character/concepts/h02-black-iron-pincer-knight-sheet.png differ
diff --git a/docs/character/concepts/h07-white-charcoal-judgement-axe-sheet.png b/docs/character/concepts/h07-white-charcoal-judgement-axe-sheet.png
new file mode 100644
index 0000000..2cc8f13
Binary files /dev/null and b/docs/character/concepts/h07-white-charcoal-judgement-axe-sheet.png differ
diff --git a/docs/character/concepts/h08-navy-lantern-threadbinder-sheet.png b/docs/character/concepts/h08-navy-lantern-threadbinder-sheet.png
new file mode 100644
index 0000000..8e6a64f
Binary files /dev/null and b/docs/character/concepts/h08-navy-lantern-threadbinder-sheet.png differ
diff --git a/docs/character/concepts/h09-veiled-oath-relic-duelist-sheet.png b/docs/character/concepts/h09-veiled-oath-relic-duelist-sheet.png
new file mode 100644
index 0000000..5a038d8
Binary files /dev/null and b/docs/character/concepts/h09-veiled-oath-relic-duelist-sheet.png differ
diff --git a/docs/product/game-definition-threadlight.md b/docs/product/game-definition-threadlight.md
new file mode 100644
index 0000000..ef31ecb
--- /dev/null
+++ b/docs/product/game-definition-threadlight.md
@@ -0,0 +1,284 @@
+# Game Definition: Threadlight Pilgrimage
+
+Status: proposed product direction after market-expression research.
+Date: 2026-06-30.
+
+This document replaces weapon-first character exploration as the current
+creative gate. Character, Meshy, Maya, Unreal, and Unity work should serve this
+definition until it is intentionally changed.
+
+## One-Line Pitch
+
+Enter frayed provinces as a lantern-thread surveyor, read the traces of a
+broken land, bind its monsters and routes with threadlight, and bring lost
+memories back to a growing pilgrimage caravan.
+
+## What The Player Buys
+
+The player is not buying a sword, axe, or armor set.
+
+The player is buying the fantasy of becoming a careful, brave field worker in a
+beautifully damaged world:
+
+- reading danger before it kills them
+- restoring paths others can no longer cross
+- rescuing memories, names, and people from places that have come apart
+- returning to a caravan that becomes warmer and stranger after each expedition
+- mastering action combat because they understand the land, not because they
+  found a bigger weapon
+
+## Genre
+
+Top-down action roguelite / compact ARPG.
+
+The first commercial prototype should prove:
+
+- readable movement, dodge, and melee timing
+- one exploration tool that changes combat and navigation
+- short expeditions with retreat/deeper risk
+- one hub that visibly grows
+- one boss whose attacks are explained by the region's world rule
+
+## Player Role
+
+The player is a `Threadlight Surveyor`.
+
+Surveyors belong to a travelling repair pilgrimage. They enter provinces where
+roads, names, memories, and buildings have frayed into physical thread. Their
+job is not conquest. Their job is to make a route safe enough that others can
+return.
+
+The hero should read as profession first, warrior second:
+
+- lantern or thread tool
+- field coat, mantle, sash, or work apron
+- measuring pins, spools, charms, chalk, map cloth, and route tags
+- one practical weapon that also supports the job
+- clear top-down value mark: pale lantern, white thread, red route cloth, or
+  blue-black coat
+
+Avoid hero-first signals:
+
+- chosen-one crown
+- generic holy knight
+- huge spiked armor
+- skull overload
+- weapon collector silhouette
+- black-on-black body with no job identity
+
+## World Premise
+
+The world entered a `Long Night`. Some provinces did not simply fall into ruin;
+they came unstitched.
+
+Memory became thread. Grief became knots. Roads now loop back on themselves.
+Old flags, wedding veils, work clothes, banners, curtains, and burial cloth
+wake as hostile relics. Lamps and bells are the safest way to mark a true path.
+
+People survive by travelling in pilgrimage caravans built around looms,
+lantern racks, map cloth, kitchens, sleeping shelves, and memory archives.
+
+## Hub
+
+The hub is a mobile `Loom Caravan`.
+
+It is not a menu room. It is the emotional anchor.
+
+Hub upgrades should be visible:
+
+- lantern shelf fills with recovered lights
+- NPC bunks appear and become lived-in
+- map cloth expands across the wall
+- repaired tools hang from hooks
+- cooking area becomes warmer
+- memory fragments become small exhibits
+- failed expeditions leave stains, torn cloth, or missing markers
+
+This is how runs gain emotional weight beyond loot.
+
+## Core Loop
+
+1. Prepare at the Loom Caravan.
+2. Choose a frayed province and one route objective.
+3. Enter a compact top-down expedition.
+4. Read traces: loose threads, lamp color, footprints, corpse direction, bell
+   rhythm, cloth movement, floor seams.
+5. Use the threadlight tool to reveal, bind, anchor, or cut a route.
+6. Fight enemies whose attacks come from the same local rule.
+7. Decide to retreat with rescued memory or push deeper for a route seal.
+8. Return to the caravan and see the world, NPCs, or tools change.
+
+## First Prototype Slice
+
+The first 30-second slice should not try to show the whole game.
+
+It should show:
+
+- a surveyor entering a rain-dark cloth street
+- the threadlight lantern revealing a hidden route seam
+- one enemy that eats light
+- one enemy that tangles the player's route line
+- one elite with a clear bell or banner telegraph
+- a short fight with hit stop, knockback, and readable danger colors
+- one rescued memory returned to a visible caravan object
+- one boss tease: a giant loom gate or bell tower begins moving
+
+## Visual Pillars
+
+Use these before character detail:
+
+1. `Thread and cloth`: route lines, torn banners, veils, knots, sashes, ropes,
+   stitching, repair tags.
+2. `Lantern light`: warm safety, pale memory light, reflected wet stone,
+   readable player mark.
+3. `Soft dark fantasy`: dangerous and melancholy, not rugged armor brutality.
+4. `Profession silhouettes`: surveyor, weaver, bell keeper, route guard,
+   lantern carrier, map stitcher.
+5. `Repair over conquest`: visual payoff is opening, mending, lighting, and
+   returning, not only killing.
+
+Primary palette:
+
+- ink blue
+- black cloth
+- wet stone gray
+- warm lantern gold
+- pale memory white
+- oxblood route cloth
+- moss green
+
+Avoid:
+
+- one-note slate/dark-blue UI
+- generic brown dungeon
+- all-black hero on all-black floor
+- purple-blue magic as the main palette
+- sand/beige fantasy ruins
+
+## Character Direction
+
+Existing generated sheets are not final protagonist direction. Preserve them as
+parts:
+
+- H09: hooded field-worker mood, relic weight, top-down cloak mass.
+- F01: pale veil/crest and long polearm readability, but convert armor into
+  ritual field uniform.
+- F03: bell-root motif and occult ecology, useful for NPC, elite, or region
+  boss language.
+- H08: lantern/thread action language, but make it a physical survey tool
+  instead of thin VFX loops.
+- H02/H07: reserve for enemy, guard, or boss silhouettes, not first hero.
+- F02: hold for redesign; current ninja/fairy signal is off-direction.
+
+New character prompts should lead with job and world:
+
+- rain surveyor with lantern-thread spool
+- route mender with map cloth and short blade
+- bell keeper with cracked bronze mantle
+- caravan stitcher with field apron and measuring pins
+- ash veil route guard with practical polearm
+
+Do not lead with:
+
+- dark fantasy warrior
+- magic swordsman
+- execution axe hero
+- armored valkyrie
+- rogue assassin
+
+## Enemy Ecology
+
+Enemies are not demons for their own sake. They are frayed memories and hostile
+relic behavior.
+
+First enemy set:
+
+- `Light-Moth`: small fast creature that dims safe lantern zones.
+- `Knot Hound`: low runner that pulls thread routes into hazards.
+- `Banner Husk`: cloth soldier that hides attack direction until lit.
+- `Bell Root`: stationary pulse enemy that changes floor rhythm.
+- `Loom Gate Keeper`: first boss; a doorway/loom/bell machine that controls the
+  route network.
+
+Each enemy must teach the world rule:
+
+- light reveals
+- thread binds
+- knots trap
+- bells warn
+- repaired routes create safety
+
+## Store Page Promise
+
+The first store-facing image should show:
+
+- a small readable surveyor with a warm lantern
+- a visible thread route crossing a dark floor
+- a hostile relic or cloth creature pulling the route apart
+- a large landmark: loom gate, bell tower, sunken wedding hall, or stitched
+  street
+- a hint of the caravan or rescued memory, so the game is not only combat
+
+First trailer structure:
+
+1. `0-3s`: lantern lights a thread route in a dark street.
+2. `3-7s`: the route reveals danger before an enemy strikes.
+3. `7-12s`: combat with a white hit core, warm thread trail, and knockback.
+4. `12-16s`: retreat/deeper choice at a fraying doorway.
+5. `16-21s`: caravan return, recovered memory changes a visible object.
+6. `21-26s`: boss landmark wakes.
+7. `26-30s`: high-readability combat beat, title card.
+
+## Audio Direction
+
+The sound should be dark but not generic.
+
+Signature materials:
+
+- taut thread plucks
+- loom wood knocks
+- lantern glass clicks
+- soft bells
+- rain on cloth
+- low strings
+- dry hand percussion
+- breath-like choir fragments
+
+Combat mix:
+
+- player attacks: short cloth/metal motion plus clear contact
+- enemy tells: bells, thread tension, breath, low scrape
+- route tool: satisfying pluck, glow, and latch
+- reward: small memory chime, not long fanfare
+
+## Production Gate
+
+Before any new Meshy/API spend:
+
+1. Create a world contact sheet for Threadlight Pilgrimage.
+2. Create a 1280x800 gameplay mock with surveyor, thread route, enemy, hub
+   payoff, and boss tease.
+3. Create one environment style tile.
+4. Create a 30-second trailer beatboard.
+5. Only then choose up to two assets for 3D preview.
+
+Preferred first 3D targets:
+
+1. threadlight route tool
+2. light-moth enemy
+3. surveyor body blockout
+
+Do not start with a fully polished hero model.
+
+## Rejection Rules
+
+Reject any direction that:
+
+- can only be explained as "dark fantasy ARPG"
+- looks like weapon concept art without a world rule
+- makes every character rugged, armored, and grim
+- has no hub or non-combat desire
+- cannot show the game promise in one screenshot
+- requires paid generation to look convincing
+- would still work if the thread, lantern, caravan, and repair ideas were
+  removed
diff --git a/docs/production/maya-meshy-gate.md b/docs/production/maya-meshy-gate.md
index c4f53dd..6f08358 100644
--- a/docs/production/maya-meshy-gate.md
+++ b/docs/production/maya-meshy-gate.md
@@ -2,6 +2,14 @@
 
 Status: active production gate.
 
+## World Gate Precondition
+
+New character generation is paused until the world direction in
+`docs/product/game-definition-threadlight.md` is accepted or replaced.
+
+Maya and Meshy should not be used to rescue a vague fantasy character. They are
+only useful after the asset proves a specific game promise on screen.
+
 ## Allowed Use
 
 Maya:
@@ -33,10 +41,19 @@ Do not spend API credits on:
 
 ## First Candidate If Needed
 
-The only next permitted new generation target is one playable hero candidate:
+The next permitted generated asset is not a fully polished hero. Use one
+low-cost preview only after the world contact sheet and gameplay mock exist.
+
+Preferred order:
 
-- top-down action RPG playable hero
-- original dark fantasy magic swordsman
+1. threadlight route tool
+2. light-moth enemy
+3. surveyor body blockout
+
+The first playable body candidate must be:
+
+- top-down action RPG surveyor
+- original threadlight repair-pilgrimage field worker
 - readable head, shoulders, cape, weapon
 - asymmetrical facing cues
 - broad color blocking
@@ -47,8 +64,8 @@ Failure means:
 
 - generic armor NPC
 - unreadable at camera scale
-- weapon too thin
+- tool or weapon too thin
 - facing unclear
+- job identity unclear
 - famous-character resemblance
 - requires changing the game direction to justify it
-
diff --git a/docs/production/task-split.md b/docs/production/task-split.md
index 4916278..a3b07c2 100644
--- a/docs/production/task-split.md
+++ b/docs/production/task-split.md
@@ -11,10 +11,14 @@ Status: first PR task map for Veripsa/Core review.
 
 ## Milestone 1: Visual Target Without Engine Lock
 
+- [ ] M1-T00: Accept or replace `Threadlight Pilgrimage` as the concrete world
+  and store-facing promise.
 - [ ] M1-T01: Create one-page final-frame brief.
 - [ ] M1-T02: Build free/CC0 asset shortlist for hero, enemy, floor, and props.
 - [ ] M1-T03: Create visual mood board using only licensed links and notes.
 - [ ] M1-T04: Define 30-second storyboard acceptance checklist.
+- [ ] M1-T05: Create gameplay mock showing world rule, hub payoff, and one
+  combat beat before any Meshy/Maya spend.
 
 ## Milestone 2: Engine Bakeoff
 
@@ -47,4 +51,3 @@ Status: first PR task map for Veripsa/Core review.
 - Do not mix market research, engine setup, gameplay, and art import in one PR.
 - Every asset PR needs source/license notes.
 - Every visual PR needs screenshot evidence.
-
diff --git a/docs/research/commercial-expression-benchmark.md b/docs/research/commercial-expression-benchmark.md
new file mode 100644
index 0000000..fbffd73
--- /dev/null
+++ b/docs/research/commercial-expression-benchmark.md
@@ -0,0 +1,96 @@
+# Commercial Expression Benchmark
+
+Status: market-facing expression research for the next art direction gate.
+Date: 2026-06-30.
+
+Use these references as market evidence only. Do not copy their characters,
+costumes, UI, worlds, palettes, trailers, logos, music, or named systems.
+
+## Core Finding
+
+Successful top-down action RPG and roguelite games do not sell a weapon first.
+They sell a role, a world rule, and an emotional reason to continue.
+
+The buyer-facing promise is usually simple:
+
+- become a specific kind of person
+- enter a world with a memorable rule
+- repeat a dangerous loop that changes the world, the cast, or the player
+- see growth in the screen, home base, relationships, or build identity
+
+Weapon art is useful only when it makes that promise easier to understand.
+
+## Benchmarks
+
+| Game | Market Signal | What The Store/Trailer Sells | Expression Principle To Absorb |
+| --- | --- | --- | --- |
+| Hades | Reported over one million copies before/around launch; Steam page remains a benchmark for action roguelite presentation. | A rebellious escape from the underworld, family conflict, gods as vivid personalities, fast readable combat. | Sell relationships and role before builds. Combat rewards work because characters and return visits matter. |
+| Hades II | Strong Steam reception and high launch attention; official positioning centers on witchcraft, the underworld, and the Titan of Time. | A witch against Time, ritual identity, moon/night colors, new protagonist role. | A sequel or adjacent design must shift job, color, ritual, and emotional fantasy, not just add weapons. |
+| Cult of the Lamb | Reported one million copies in its first week. | Cute lamb, evil cult, rituals, followers, base life, dark humor. | A strong mascot plus an uncomfortable verb can sell instantly. The base and rituals are as important as combat. |
+| V Rising | Reported five million copies sold by 2025. | Weak vampire becoming a ruler, castle building, blood hunting, daylight as a hard rule. | A visible world constraint, such as sunlight, makes screenshots explain gameplay without text. |
+| No Rest for the Wicked | Reported near two million copies during Early Access. | Painterly isometric world, heavy manual melee, disease and island dread. | Premium appeal comes from art quality plus input weight. Screens must show craft, not just darkness. |
+| Diablo IV | Large mainstream dark fantasy benchmark. | Lilith, hell, gothic religion, class fantasy, season/live-service scale. | The villain and world architecture can be the product face. Darkness needs strong iconography. |
+| Path of Exile 2 | High-interest ARPG benchmark with deep systems. | Brutal bosses, build depth, grim world, complex mastery. | Depth is also a visual promise. Boss tells, skill trees, and build screens can be sellable images. |
+| Death's Door | High-signal indie action adventure benchmark. | A small crow reaper, dead-end job premise, melancholy humor, clean diorama worlds. | A small protagonist plus a large job concept can be more memorable than detailed armor. |
+| Children of Morta | Strong family/action roguelite reference. | A family of heroes, home warmth, corruption outside the home. | A roster is stronger when it is a relationship network, not a class list. |
+| Curse of the Dead Gods | Smaller but useful readability benchmark. | Torches, greed, traps, curses, temple risk. | A player vice can be a mechanic. Make greed, fear, or retreat visible on screen. |
+
+## Trailer And Screenshot Patterns
+
+- First 3-5 seconds show the promise, not the tutorial.
+- Strong trailers alternate world, combat, boss, reward, and identity.
+- HUD is usually hidden or reduced in the first store-facing shots.
+- A hero needs a dedicated value/color accent that survives dark floors.
+- Hit impact usually uses a white or yellow core, colored outer shape, hit stop,
+  knockback, and debris or body reaction.
+- Bosses sell by screen control: large silhouette, strong telegraph, and a tiny
+  readable player.
+- A screenshot should expose the world rule. Good examples: sunlight, cult
+  ritual, family home, underworld exit, torch/curse pressure, or a giant gate.
+
+## Audio Patterns
+
+Strong games have a sound identity that can be described in one line:
+
+- Hades: mythic percussion and rock energy.
+- Cult of the Lamb: cute ritual unease.
+- V Rising: gothic predator ownership.
+- Diablo IV and Path of Exile 2: heavy corruption, metal, flesh, and dread.
+- Death's Door and Children of Morta: melancholy warmth around dark subject
+  matter.
+
+Our direction should avoid generic cinematic fantasy. Use a specific sonic
+signature tied to the world rule: thread tension, lantern glass, looms, rain,
+soft bells, breath, low strings, and dry percussion.
+
+## Implications For This Project
+
+The current generated characters are too often rugged armor plus oversized
+weapon. That can support a combat prototype, but it is not enough to sell the
+game.
+
+The product needs a world-language first:
+
+- a clear player job
+- a hub or community worth returning to
+- a visible rule that changes exploration and combat
+- enemy ecology, not generic demons
+- a non-combat desire such as repair, rescue, memory, cult, castle, family, or
+  home growth
+
+## Sources
+
+- Hades Steam: https://store.steampowered.com/app/1145360/Hades/
+- Hades sales report: https://www.gamedeveloper.com/business/-i-hades-i-has-sold-over-1-million-copies-in-under-two-years
+- Hades II Steam: https://store.steampowered.com/app/1145350/Hades_II/
+- Cult of the Lamb Steam: https://store.steampowered.com/app/1313140/Cult_of_the_Lamb/
+- Cult of the Lamb one million sales report: https://www.gematsu.com/2022/08/cult-of-the-lamb-sales-top-one-million
+- V Rising Steam: https://store.steampowered.com/app/1604030/V_Rising/
+- V Rising five million sales report: https://www.gematsu.com/2025/01/v-rising-sales-top-five-million
+- No Rest for the Wicked Steam: https://store.steampowered.com/app/1371980/No_Rest_for_the_Wicked/
+- No Rest for the Wicked sales report: https://www.gamesradar.com/games/action-rpg/after-2-years-in-steam-early-access-and-almost-2-million-copies-sold-arpg-no-rest-for-the-wicked-is-not-slowing-down-on-its-way-to-1-0/
+- Diablo IV Steam: https://store.steampowered.com/app/2344520/Diablo_IV/
+- Path of Exile 2 Steam: https://store.steampowered.com/app/2694490/Path_of_Exile_2/
+- Death's Door Steam: https://store.steampowered.com/app/894020/Deaths_Door/
+- Children of Morta Steam: https://store.steampowered.com/app/330020/Children_of_Morta/
+- Curse of the Dead Gods Steam: https://store.steampowered.com/app/1123770/Curse_of_the_Dead_Gods/
diff --git a/docs/research/game-character-production-pipeline.md b/docs/research/game-character-production-pipeline.md
new file mode 100644
index 0000000..35a597d
--- /dev/null
+++ b/docs/research/game-character-production-pipeline.md
@@ -0,0 +1,194 @@
+# Game Character Production Pipeline
+
+Status: research note for deciding how to create the first commercial-looking
+ARPG character without wasting Maya/Meshy/API spend.
+
+## Blunt Finding
+
+The current Phase 1 character fails as a commercial character because it is a
+procedural gameplay marker, not a production character. A production character
+needs readable identity, authored shape language, deformation-safe topology,
+materials that survive the game camera, animation poses that sell intent, and
+engine evidence. Meshy can help create starting geometry. Maya is where the
+result must be judged, cleaned, rigged, skinned, and made usable.
+
+Do not spend paid generation credits until the character has a one-page
+direction sheet and a pass/fail test at gameplay camera distance.
+
+## Pipeline
+
+| Stage | Output | Pass Gate |
+| --- | --- | --- |
+| 1. Role and camera brief | One sentence role, weapon, scale, camera-distance crop, color/value target. | Readable as hero/enemy at 256 px tall screenshot. |
+| 2. Silhouette sheet | 6-12 black silhouettes, front/side/back roughs, weapon profile. | Head, shoulders, weapon, facing direction readable in black. |
+| 3. Color/material key | 2-3 material/value studies, no final render dependency. | Hero separates from blue-gray dungeon floor and warm torches. |
+| 4. Generation or blockout | Meshy preview, Maya blockout, or kitbash blockout. | Shape works untextured from top-down. |
+| 5. Cleanup/retopo | Animation-ready low/mid poly mesh, named parts, pivot/origin set. | No melted limbs, no ambiguous hands/weapon, no tangled intersections. |
+| 6. UV/materials | Base color, normal, roughness/metal where useful. | Materials read under engine lighting, not only in DCC viewport. |
+| 7. Rig/skin | Skeleton, skin weights, weapon sockets, optional cloth/cape bones. | Idle/run/attack/death poses deform without collapsing volume. |
+| 8. Animation set | Idle, run, dodge, attack, hit, death, and one skill pose. | Each pose reads in a still frame before motion blur/VFX. |
+| 9. Engine import | Unity or Unreal prefab/blueprint with scale, materials, animator. | Screenshot and contact sheet prove commercial direction. |
+| 10. Ledger | Source, license, prompts, task IDs, edits, export formats. | Rebuildable, reviewable, legally traceable asset. |
+
+## Maya's Correct Role
+
+Maya is not where we should improvise a finished character from nothing. It is
+the production control point:
+
+- clean Meshy/kitbash geometry
+- fix scale, forward axis, pivot, naming, hierarchy
+- retopologize or simplify problem areas
+- create/adjust skeleton joints
+- bind mesh to skeleton and paint skin weights
+- pose-test shoulders, cape, weapon, hands, knees, and death collapse
+- export FBX/GLB for engine import
+
+Unreal's FBX skeletal mesh pipeline describes the same basic production facts:
+characters are exported from DCC software as FBX, contain mesh and skeleton, can
+include animations/morph targets, and support materials, multiple UVs, vertex
+colors, smoothing groups, and LODs. It also calls out Maya Joint Tool, Smooth
+Bind, Paint Skin Weights, manual triangulation control, normal-map baking, and
+Skeletal Mesh LOD setup as production concerns.
+
+## Meshy/API Role
+
+Meshy is useful only if treated as a draft asset generator plus cleanup input.
+The official Text to 3D API uses a two-step preview/refine flow: preview creates
+untextured geometry, refine textures it. It can request `glb`, `obj`, `fbx`,
+`stl`, `usdz`, and `3mf` outputs, supports A/T pose requests, target polycount,
+and PBR/HD texture options. The docs also show that tasks consume credits, so
+preview spam is waste.
+
+Meshy's Rigging API can add skeletons to humanoid models, but the official docs
+state that programmatic rigging currently works well only for standard bipedal
+humanoids with clear limbs/body. It is not suitable for untextured meshes,
+non-humanoids, unclear humanoids, or models over 300,000 faces when using a
+task ID. Model URL rigging requires textured GLB and forward-facing +Z.
+
+Meshy's Animation API applies library animations to a previously rigged
+character. That can accelerate tests, but it does not solve game-specific attack
+timing, hit stop, dodge frames, weapon sockets, or top-down pose readability.
+
+## Unity/Unreal Import Facts
+
+Unity:
+
+- Humanoid import needs a valid Avatar mapped to a human-like skeleton.
+- Unity distinguishes Humanoid and Generic rigs; non-human enemies may need
+  Generic.
+- Unity recommends checking Avatar mapping even when automatic mapping succeeds.
+- The Rig import tab defaults to 4 bone skin weights for performance, and higher
+  bone influence has cost.
+- Animation clips can be masked, clipped, loop-optimized, and event-marked at
+  import.
+
+Unreal:
+
+- The foundational character asset is Skeletal Mesh: geometry plus skeleton.
+- FBX import can bring mesh, skeleton, animations, morph targets, materials,
+  multiple UVs, smoothing groups, vertex colors, and LODs.
+- Character behavior is assembled in a Character Blueprint and Animation
+  Blueprint.
+- Skeleton compatibility and retargeting matter if we want to reuse animations.
+- LODs should align to the same skeleton/pivot and can reduce material/detail
+  complexity at distance.
+
+## Top-Down ARPG Character Readability Rules
+
+For this project, character quality is judged from gameplay camera first, not
+from a beauty render.
+
+Required:
+
+- oversized weapon or signature prop visible from top-down
+- head/shoulder silhouette separated from cape/body mass
+- one bright hero value band and one dark anchor band
+- warm/cool material contrast that survives dungeon lighting
+- asymmetry: shoulder plate, cape tear, lantern, relic, horn, or weapon shape
+- attack pose with clear anticipation, contact, follow-through
+- enemy silhouette that differs in size, color, and facing language
+
+Rejected:
+
+- white blob hero
+- generic armored human with no weapon identity
+- details only visible close-up
+- hair/cape/cloth fused into a single soft mass
+- Meshy output used directly in-engine without cleanup
+- animation that reads only when moving, not in still frames
+
+## First Character Target
+
+Do not start with a complex full hero. Start with a "hero-readable prototype
+character" that can pass gameplay distance:
+
+- role: cursed foldblade knight / relic duelist
+- camera: top-down/isometric gameplay distance, character occupies 12-16% of
+  screen height in hero evidence shot
+- silhouette: broad left shoulder, long diagonal folding blade, short torn cape,
+  bright mask/crest, dark lower body
+- materials: pale mask/crest, blackened armor, oxblood cloth, cyan relic glow,
+  warm edge highlights from torches
+- animations: idle, run, dodge, light attack, heavy attack, hit, death
+
+This is intentionally narrow. If this fails, adding more enemies or skills will
+not fix the game.
+
+## Spend Gate For Meshy
+
+Before any paid/API generation:
+
+- [ ] final role sentence approved
+- [ ] 6+ silhouettes reviewed at gameplay crop
+- [ ] front/side/back rough or reference sheet exists
+- [ ] prompt includes pose, camera-readability, weapon, material bands, and
+      forbidden details
+- [ ] one preview limit per candidate; no refine until preview passes silhouette
+- [ ] task IDs and prompts will be logged in asset ledger
+- [ ] exported model will be cleaned in Maya before engine import
+
+First allowed API experiment:
+
+```text
+Create one Meshy preview only, A-pose, GLB output, low-to-mid poly target,
+for a dark fantasy top-down ARPG hero with exaggerated readable silhouette:
+bright mask/crest, asymmetrical shoulder armor, torn short cape, long diagonal
+foldblade weapon, blackened armor, oxblood cloth, cyan relic accent. Avoid tiny
+ornaments, fused hands, hidden weapon, symmetrical generic knight, realistic
+face detail, and long trailing cloth that would break top-down readability.
+```
+
+Do not refine/texture until the preview silhouette passes against the current
+game camera.
+
+## Immediate Next PR
+
+The next implementation PR should not be another procedural body polish pass.
+It should create a character-evidence lane:
+
+1. Add `docs/asset-ledger/character-ledger.md`.
+2. Add a `docs/character/hero-readable-prototype-brief.md` with silhouette,
+   material, animation, and Meshy spend gate.
+3. Add a Unity/Unreal neutral screenshot checklist for imported character
+   evidence.
+
+Only after that should a Meshy/Maya/engine import PR be opened.
+
+## Sources
+
+- Meshy API introduction and endpoint list: https://docs.meshy.ai/en
+- Meshy Text to 3D API: https://docs.meshy.ai/en/api/text-to-3d
+- Meshy Rigging API: https://docs.meshy.ai/en/api/rigging
+- Meshy Animation API: https://docs.meshy.ai/en/api/animation
+- Meshy Remesh API: https://docs.meshy.ai/en/api/remesh
+- Meshy for Maya plugin: https://docs.meshy.ai/en/maya-plugin/introduction
+- Unity Humanoid import and Avatar configuration:
+  https://docs.unity3d.com/Manual/ConfiguringtheAvatar.html
+- Unity Rig import settings:
+  https://docs.unity3d.com/Manual/FBXImporter-Rig.html
+- Unreal Skeletal Mesh assets:
+  https://dev.epicgames.com/documentation/en-us/unreal-engine/skeletal-mesh-assets-in-unreal-engine
+- Unreal FBX Skeletal Mesh Pipeline:
+  https://dev.epicgames.com/documentation/en-us/unreal-engine/fbx-skeletal-mesh-pipeline-in-unreal-engine
+- Unreal Animation Retargeting:
+  https://dev.epicgames.com/documentation/en-us/unreal-engine/animation-retargeting-in-unreal-engine
diff --git a/docs/research/market-fit.md b/docs/research/market-fit.md
index a40b9b5..f76c5c4 100644
--- a/docs/research/market-fit.md
+++ b/docs/research/market-fit.md
@@ -24,27 +24,30 @@ characters, UI, logos, costumes, palettes, silhouettes, lore, or named systems.
 
 Small commercial target:
 
-- 3D top-down dark fantasy survivor ARPG
+- top-down action roguelite / compact ARPG
 - manual movement, dodge, and melee timing
-- semi-automatic magic, cursed relics, traps, or summons layered on top
-- 15-25 minute runs
+- one exploration tool that changes navigation and combat
+- short expeditions with retreat/deeper pressure
+- a visible hub that grows after successful runs
 - one clear boss or elite moment in the first short capture
 - low-to-mid price premium indie positioning
 
-The store hook should be:
+Current product definition:
 
-> dodge through a cursed cathedral, stack forbidden relics, and turn one
-> readable melee hero into a screen-clearing build within minutes.
+> Enter frayed provinces as a lantern-thread surveyor, read the traces of a
+> broken land, bind its monsters and routes with threadlight, and bring lost
+> memories back to a growing pilgrimage caravan.
+
+See `docs/product/game-definition-threadlight.md`.
 
 ## First Store-Facing Promise
 
 The first screenshot/video must show:
 
-- one readable hero
+- one readable surveyor
+- a thread route, lantern mark, or repaired path
 - diagonal motion
-- enemy density rising from edges
-- one strong skill shape
-- one medium enemy or boss threat
-- one reward/build choice
+- one enemy interacting with light/thread
+- one medium enemy or boss threat tied to the world rule
+- one visible hub or rescued-memory payoff
 - dark setting with readable foreground action
-
diff --git a/docs/research/video-analysis-tooling.md b/docs/research/video-analysis-tooling.md
new file mode 100644
index 0000000..bda04fb
--- /dev/null
+++ b/docs/research/video-analysis-tooling.md
@@ -0,0 +1,171 @@
+# Video Analysis Tooling
+
+Status: local free-tool setup note.
+
+## Installed Tool
+
+Installed with Homebrew:
+
+```bash
+brew install ffmpeg
+```
+
+Verified tools:
+
+- `ffmpeg 8.1.2`
+- `ffprobe 8.1.2`
+
+`ffmpeg` is the baseline free tool for trailer/gameplay evidence work:
+
+- read video metadata
+- extract exact frames
+- generate contact sheets
+- extract audio
+- measure volume
+- generate spectrogram/waveform evidence
+- transcode Unity/Unreal captures to reviewable MP4
+
+Do not commit downloaded commercial reference videos to the repository. Store
+temporary analysis inputs under `/tmp` or another ignored local evidence
+folder, and cite the source URL in research docs.
+
+## Smoke Test Performed
+
+Created a one-second synthetic video and audio test:
+
+```bash
+ffmpeg -y \
+  -f lavfi -i testsrc2=duration=1:size=960x540:rate=24 \
+  -f lavfi -i sine=frequency=440:duration=1 \
+  -pix_fmt yuv420p \
+  -shortest \
+  /tmp/arpg-ffmpeg-smoke.mp4
+```
+
+Read metadata:
+
+```bash
+ffprobe -v error \
+  -show_entries format=duration:stream=index,codec_type,width,height,r_frame_rate,sample_rate \
+  -of default=noprint_wrappers=1 \
+  /tmp/arpg-ffmpeg-smoke.mp4
+```
+
+Extracted one frame:
+
+```bash
+ffmpeg -y \
+  -ss 0.5 \
+  -i /tmp/arpg-ffmpeg-smoke.mp4 \
+  -frames:v 1 \
+  -update 1 \
+  /tmp/arpg-ffmpeg-smoke-frame.png
+```
+
+Measured volume:
+
+```bash
+ffmpeg \
+  -i /tmp/arpg-ffmpeg-smoke.mp4 \
+  -af volumedetect \
+  -f null -
+```
+
+Observed output included:
+
+- video stream: 960x540, 24 fps
+- audio stream: 44.1 kHz
+- duration: 1.000000 seconds
+- mean volume: about `-21.2 dB`
+- max volume: about `-17.7 dB`
+
+## Reference Trailer Analysis Commands
+
+For local, permitted trailer/gameplay files:
+
+Metadata:
+
+```bash
+ffprobe -v error \
+  -show_entries format=duration,bit_rate:stream=index,codec_type,width,height,r_frame_rate,sample_rate \
+  -of json \
+  /path/to/reference.mp4
+```
+
+One frame every two seconds:
+
+```bash
+mkdir -p /tmp/arpg-reference-frames
+ffmpeg -y \
+  -i /path/to/reference.mp4 \
+  -vf fps=1/2,scale=1280:-2 \
+  /tmp/arpg-reference-frames/frame_%03d.png
+```
+
+Contact sheet from extracted frames:
+
+```bash
+ffmpeg -y \
+  -pattern_type glob \
+  -i "/tmp/arpg-reference-frames/*.png" \
+  -vf "scale=320:-2,tile=4x3" \
+  /tmp/arpg-reference-contact-sheet.png
+```
+
+Audio waveform image:
+
+```bash
+ffmpeg -y \
+  -i /path/to/reference.mp4 \
+  -filter_complex "aformat=channel_layouts=mono,showwavespic=s=1600x300:colors=cyan" \
+  -frames:v 1 \
+  /tmp/arpg-reference-waveform.png
+```
+
+Audio spectrogram image:
+
+```bash
+ffmpeg -y \
+  -i /path/to/reference.mp4 \
+  -lavfi showspectrumpic=s=1600x900:legend=disabled \
+  /tmp/arpg-reference-spectrogram.png
+```
+
+Volume:
+
+```bash
+ffmpeg \
+  -i /path/to/reference.mp4 \
+  -af volumedetect \
+  -f null -
+```
+
+## What To Measure For This Game
+
+For each reference trailer/gameplay clip:
+
+- first readable hero frame time
+- first enemy-threat frame time
+- first hit/impact frame time
+- first reward/build-choice frame time
+- camera distance and character screen-height percentage
+- screen center occupancy by character/enemy/VFX vs empty floor
+- audio presence in first 3 seconds
+- mean/max loudness from `volumedetect`
+- waveform density during attacks vs downtime
+
+For our own capture:
+
+- reject silent clips for store-facing evidence
+- reject clips where the hero is not readable in extracted frames
+- reject clips where the first 5 seconds show only floor/idle motion
+- always produce a contact sheet and a waveform alongside any video PR
+
+## Tooling Gaps
+
+`ffmpeg` is enough for immediate free analysis. Optional later tools:
+
+- `yt-dlp`: free downloader/metadata extractor for public video URLs, but use
+  only where source terms and rights permit local analysis.
+- Python OpenCV: useful for automated frame statistics, but not required before
+  we have local video captures and reference files.
diff --git a/docs/research/visual-audio-benchmark.md b/docs/research/visual-audio-benchmark.md
index 8d8adb4..6735d62 100644
--- a/docs/research/visual-audio-benchmark.md
+++ b/docs/research/visual-audio-benchmark.md
@@ -28,14 +28,14 @@ Status: research input for PR review.
 
 ## First 30 Seconds Storyboard
 
-1. `0-2s`: hero alone on dark readable floor, weapon/identity visible.
-2. `2-6s`: diagonal movement, first hit, dodge, enemy death, drop feedback.
-3. `6-10s`: first signature skill with a clear shape.
-4. `10-15s`: enemies enter from edges; keep a readable player pocket.
-5. `15-20s`: medium enemy or mini-boss with size, color, and telegraph.
-6. `20-24s`: reward/build choice, three options maximum.
-7. `24-28s`: chosen upgrade immediately changes the combat shape.
-8. `28-30s`: highest-density frame, then title/CTA.
+1. `0-2s`: surveyor lights a thread route on a dark readable floor.
+2. `2-6s`: diagonal movement, hidden danger revealed, first dodge and hit.
+3. `6-10s`: threadlight tool binds or anchors one route/enemy.
+4. `10-15s`: light-moth or knot enemy attacks the route; keep player readable.
+5. `15-20s`: medium enemy or mini-boss with bell/banner telegraph.
+6. `20-24s`: rescued memory or route seal returns to the caravan.
+7. `24-28s`: boss landmark wakes and changes the room rule.
+8. `28-30s`: highest-readability combat frame, then title/CTA.
 
 ## Audio Direction
 
@@ -43,7 +43,7 @@ Music:
 
 - low drone
 - metal or skin percussion
-- short string/guitar/synth motif
+- short thread pluck, string, or hammered-loom motif
 - optional low choir for boss pressure
 - SFX-forward mix with space in the 2-5 kHz range
 
@@ -53,7 +53,8 @@ Combat SFX:
 - dodge = short cloth/leather/air burst, 0.2-0.4 seconds
 - enemy death = short collapse plus soul/ember tail
 - player death = low drop, breath/heartbeat stop, momentary BGM duck
-- reward = rarity-coded stings, not long fanfares
+- threadlight route tool = taut pluck + lantern glass click + latch
+- reward = memory chime, not long fanfares
 
 Minimum audio set:
 
@@ -71,4 +72,3 @@ Minimum audio set:
 - inconsistent UI fonts, frames, or icon resolution
 - no knockback, hit stop, death reaction, or reward feedback
 - 30 seconds of the same enemy, same attack, same floor
-