Tank game reference and reuse inventory

Read-only audit, October 5, 2026. Source projects were not modified, built, downloaded, copied, or launched. This inventory is the only file written by this research task.

Style authority recovered

1. The current user request selects Diablo I as the dominant image, motion, interface and sound family for this game. Apply industrial tank subject matter through that family. Original current instructions outrank older blanket palette prescriptions.

2. Current complete local guide: Local archive: 00 START HERE - CURRENT Golden Keep Art and Atmosphere Bible - Timed Walkthrough.pdf. Verified PDF metadata: 282 pages, 88,455,408 bytes, title Golden Keep Art and Atmosphere Bible - Timed Walkthrough Edition. The 49-page current guide precedes an unchanged 233-page archive. Also at Local archive: Golden_Keep_Art_Atmosphere_Bible_Timed_Walkthrough_Edition.pdf. Existing extracted first 49 pages, read for this audit: Local archive: current-style-bible-first49.txt.

3. Original character bible read live through the connected Google Drive plugin: Style Bible Golden Keep Grim Pixilation - 1990s Blizzard. Modified August 22, 2026. It establishes miniature-like pre-rendered form, deliberate pixel clusters and serious adult anatomy. Its isolated character presentation and 24–32-color example are a specialized branch, not a newer prohibition on effects/UI/scenery or one mandatory framebuffer palette.

Precise direction to carry forward

Mass before polish: convincing cavities, overlap, thickness and contact. Severe adult faces and substantial hands; physically legible armor and robes. Avoid interchangeable rounded hero shapes, cute scale, vector/cartoon contours, and waxy uniform shading.
Materials before filtering: metal flashes selectively; cloth turns in folds; machinery is constructed from functional assemblies. Wear at heat, friction, impact and joints. Do not turn every surface uniformly matte brown. The current guide explicitly allows early-CGI shine when it describes metal; shiny plastic-like uniform materials are the problem.
Construction before noise: three scales—structural masses, assembly detail, surface incident. Dirt/noise over primitives cannot supply the desired density. Preserve quiet regions between meaningful detail.
Staging before spectacle: connected dark masses and bounded pockets of legibility; dominant motivated light; visible route, hands, instrument, threat. Diablo/Clans darkness is appropriate for this specific project. Protect gameplay reading from excessive black crush.
Diablo I material UI: one assembled artifact with recessed spaces, credible thickness and a clear hierarchy of metal, stone, parchment, wax and glass. Bright constrained warning lights/globes can belong. Readable type/cursor is independently crisp.
Sampling: establish on-screen footprint first, render convincing clean forms, then deliberately reduce; inspect role/facing/tool/interaction at actual gameplay size and in movement. Hard nearest-neighbor sampling and discrete color ramps help but do not create the style. Avoid soft paintings under a global pixel filter. Do not force an arbitrary 24/32/64 color limit if it destroys role/material identity.
Palette for this tank: soot/iron/steel/old bronze/blood red/dirty bone/olive, with practical amber and controlled warning red. Reserve accent ramps for state and crew identity. No broad teal-orange grading, candy saturation, global glowing edges, or blanket bloom.
Motion: anticipation, contact, applied force, recovery. Loader really grips and seats shell; gunner turns the mechanism; commander braces and drives. No decorative endless orbit or slow motion as a substitute for staged action.
Rejected reference: Torchlight is explicitly a negative comparator, not a target. Diablo I and Clans dominate this project; StarCraft cockpit imagery can support industrial materials and selective illumination without displacing the user's Diablo I direction.
Acceptance: reference study is not a gameplay screenshot, authored screenshot is not proof of motion, objective sound checks are not listening approval. User explicitly postpones cinematic/narrated gameplay until demo approval.

Exact existing visual references

Local archive: Diablo_1_saved_town_inventory.jpeg
Local archive: diablo HUD.jpeg
Local archive: Clans_saved_interior_HUD.jpeg
Local archive: Golden_Keep_Art_and_Atmosphere_Bible.pdf — older 174-page master; retain as archive, current 282-page guide takes precedence.
Local archive: Grim_PC_Fantasy_Illustrated_Bible_v1.1.pdf
Local archive: GoldenKeep_Visual_Rendering_Bible.pdf
Local archive: goldenkeep-old-style.png
Same art/references/: goldenkeep-races.png, orc-architecture.png. Useful material/silhouette evidence, not appropriate tank runtime assets.

Golden Ocean runtime: best technical inheritance

Root: Local archive: goldenkeep-hulks.

Verified clean main, HEAD 0aad2ef698cbe73f6cd3a6e015e30a1bc24b6c9c; remote https://github.com/redgargoyle/goldenkeep-hulks.git. Package 0.7.5-alpha.1, plain ES modules, no runtime package dependencies. Existing version is not automatically the desired visual end state.

Source	Useful pattern	Adaptation boundary
`vendor/three.module.js` (589 KiB), `three.core.js` (1.4 MiB), `THREE-LICENSE.txt`	Fully local Three.js rendering, ~2 MiB pair	Copy both required modules and MIT license if chosen; optional GLTFLoader 113 KiB, SkeletonUtils 11 KiB, BufferGeometryUtils 35 KiB only when needed
`src/world-renderer.js`	Shared scene, orthographic overhead and perspective 72-degree first-person cameras; same local walker transformed through moving vessel	Its native-resolution rendering is not Diablo rasterization; set deliberate internal low resolution for new game; do not import enormous nautical renderer wholesale
`src/captain-walk.js`, `src/deck-layout.js`	DOM-independent kinematic walker, finite checks, swept small steps, body radius, solids/floors/ramps	Tank own dimensions and colliders; same physical player position in both camera modes
`src/hulk-model.js`, `src/environment-details.js`	Textures with nearest magnification, material separation, shared geometry	Techniques reusable; wood/orc subject matter mostly inappropriate
`src/soundscape.js`, `docs/AUDIO_V2.md`	Shared bus gains, compressor/master analyser, gesture start, mute covers all buses, simulation-driven looping, cue cooldowns, diagnostics, destroy cleanup	Start with limited tank-specific cues; don't carry hundreds of nautical voices
`tools/qa-browser.mjs`	3-minute watchdog, renderer limits, bounded heap, deduplicated arguments, idempotent browser close, SIGINT/SIGTERM cleanup	Adapt/import as bounded helper; all owned browser tests use it and close in finally
`tools/qa-first-person.mjs`, `tests/captain-walk.test.mjs`, `tests/qa-browser.test.mjs`	Physical route tests and first-person state checks	Write actual tank route and actions, not source-matching tests
`docs/PRESENTATION_AND_LEARNING.md`	Stable controls and response in place; protected training; one current instruction and one observable completion condition	Essential tank teaching contract; don't explain all failures before first agency
`src/main.js` lifecycle	Hidden tab saves/pauses and suspends audio/3D work	Prevent background simulation losses and resource burn

Recommended new architecture

A small independent browser project with a deterministic fixed-step simulation, separate content definitions, 3D scene, cameras, audio and UI. Share one tank-local coordinate system between overhead navigation, first-person movement, interactable stations and crew. Build a cutaway tank with floor/collision volumes first; hide roof/near walls only for overhead. Keep real first-person walls and physical narrow slit openings. Put a bounded streaming outside scene around the tank, visible to overhead and through the same openings. Seed event pieces with distance-along-route and lateral lane; pool/recycle them, preserve shared timing so two portholes show one coherent external event. Add per-opening grime/shutter/debris overlays as occlusion, not random unrelated movies. External state must drive traction, hull shocks, visibility and crew demand so the glimpses mean something.

Separate physical symptom from cause (heat, misfire, drag, vibration, chatter) to support diagnosis; make machine-spirit state, damage, crew condition and corruption interact through observable evidence. Integrate rites into hands-on sequences with time/reagent cost. Keep correct execution and clever prioritization assessable in deterministic scenarios.

Kadabra cassette/page integration

Root Local archive: kadabra.games. Verified clean main, tracking sites/main, ahead 6; read-only audit only. Do not reset/rebase based on that difference.

Path	Responsibility
`public/index.html`	Physical cassette shelf buttons, labels and accessibility text
`public/assets/js/vhs-console.js`	`GAMES`, `GAME_EDITORIAL`, cassette titles, transport, showcase renderer
`public/assets/css/vhs-cassette-v27.css`	Six-face cassette model, held/loaded/ejected states
`public/assets/css/cassette-series.css`	Existing Golden Keep series hierarchy; new game is not automatically a Golden Keep subtitle
`public/assets/css/showcase-v28.css`	Responsive individual showcase pages
`public/assets/js/demo-builds.js`	Browser demo entry, build links; web-only game belongs here
`scripts/build-collection.mjs`	Games collection list/order and generation
`public/assets/js/tape-films.js`, `film-catalog.js`	Actual reviewed cinematic/gameplay tape routes; do not fabricate these before approval
`public/golden-ocean.html`	Small standalone page/social-metadata entry routes to `/#golden-ocean`
`public/play/poot-poot/`, `docs/poot-poot-web.md`	Recent small standalone browser game integration pattern
`review-tests/cassette-series.test.cjs`, `vhs-flow.test.cjs`, `poot-poot-browser.cjs`	Shelf-to-player and real player controls/persistence verification patterns

Existing build checks preserve native downloads, film routes, edge functions and public file boundary. Publication must update the complete existing deployment package, not replace it with a game-only directory. Existing source is plain public HTML/CSS/JS with Vite build, Node >=22.12. Run appropriate integration checks after editing. A new game can initially get honest concept/foundation status, cover art and playable prototype link; it should not claim a finished cinematic before the user's demo approval.

Reuse, provenance and licensing

Three.js license is explicit MIT in vendor directory. Preserve it.
Golden Ocean original code/geometry/textures and generated AudioX/Fish outputs have project provenance in art/PROVENANCE.md, docs/SOURCES.md, docs/AUDIO_V2.md, docs/SPOKEN_WORLD.md; not a blanket third-party asset license.
Images in original style bibles are reference evidence, not game texture/sprite source. Do not ship Blizzard art, logos, recordings or source footage as new assets.
Existing Orc/Jabberwocky models are user project assets and visually wrong for human Imperial crew; use only when a genuinely appropriate part and its provenance are established.
Generated imagery must have exact prompt/source provenance and be labeled concept versus runtime use. Generate new tank-specific art at established view footprint; do not mistake a beautiful still for a working scene.
Fish voices retain separate permission/licensing and human listening gates before narrated public release. Runtime sound architecture is reusable independently.

Audit references for memory attribution

Memory located projects and QA rules; current filesystem and Drive contents above were verified directly.

MEMORY.md:574-581 (Golden Ocean path/server pattern; rollout 01a0c1bc-82af-7d72-bd49-afedb2cde2fe).

MEMORY.md:759-773 (bounded QA and player journey; rollout 01a0a718-2d43-7870-a588-aeaa45fcd66e).

MEMORY.md:665-677 (Kadabra ownership/preservation and source paths; rollout 01a0b1e5-fb40-70e3-88e1-891e8d3fb3a6).

No builds, recordings, large writes or downloads were performed. This research therefore did not consume storage reserve. All large future work still needs live filesystem headroom checks.