A home to build.
A wilderness to understand.
Preparing your island…
Make a life from what you find.
Your island is paused.
September 9, 2026. Original browser survival prototype. This release audits and hardens the existing island instead of replacing its game systems.
Tidehaven combines first-person survival with camp life: a procedural island, finite harvesting, tool progression, production stations, land and marine predators, and a companion with social needs. It runs as one downloadable HTML file with bundled assets and no runtime network requests. The implementation uses JavaScript, Three.js r160, WebGL, procedural geometry, instanced vegetation, CSS and browser storage. Asset notices are included with the build.
The current island contains 591 resource nodes across 17 types, 54 inventory items, 33 crafting recipes, 18 building plans and 61 animals across 17 species. These are catalog and scene counts, not a claim of commercial-game scope or photorealism.
Twenty thousand independently seeded policy sessions were run against the original 2.0 simulation and rerun against the fixed 2.1 simulation. The same runner, fixtures, profiles and seeds were used in both runs. This is 20,000 paired synthetic cases, executed twice.
| Check | Before | After |
|---|---|---|
| Completed policy sessions | 20,000 | 20,000 |
| Active simulated hours | 8,334.81 | 8,334.81 |
| Action attempts, including blocked requests | 5,539,014 | 5,538,995 |
| Save/load round trips | 110,087 | 110,087 |
| Sessions with detected exceptions | 0 | 0 |
| Sessions failing checked state invariants | 0 | 0 |
| Sessions hitting the harness action limit | 0 | 0 |
| Independent randomized placement probes | 20,000 | 20,000 |
| Accepted placements intersecting a world collider | 237 | 0 |
| All accepted placement probes | 11,371 | 11,970 |
Zero detected exceptions is scoped to this execution model. It does not mean the game is bug-free. The original simulation already passed these integrity checks; the demonstrated fixes concern placement and controller/UI defects identified separately.
Each session also received a geometry probe with its own seeded random building type, position and rotation. Those probes tested actual model footprints against the island's tree/rock colliders. They were not 20,000 observed player building attempts. The new validator checks each plan's rotated footprint instead of one shared center-distance approximation. Small plans can fit in legitimate spaces while larger structures are rejected when they overlap obstacles. Existing saves retain their buildings; they are not automatically relocated.
| Finding | Correction | Evidence |
|---|---|---|
| A freshly selected building plan could reuse the previous preview's validity before a frame updated it. | Reset validity on selection/cancel and validate again at placement commit. | Reproduced on 2.0; targeted controller check passes on 2.1. |
| Large plans could intersect a tree or rock despite a green preview. | Validate rotated model footprints, slope across their corners and other structures; explain the specific rejection. | 237 accepted overlaps before, zero in the matched probes after. |
| Wildlife could hunt and kill while the player was still on the start screen. | Start-screen frames animate the atmosphere without advancing animal behavior. | Forced predator/prey controller fixture fails before and passes after. |
| A broken iron tool hid its healthy basic counterpart in the quickbar. | Prefer a usable upgrade, then a usable basic tool. | Broken-iron/working-basic fixture passes. |
| An unconscious survivor could use the free hand-foraging branch. | Reject harvesting before any resource mutation or reward. | Dead-player palm-foraging check passes. |
| Walking-motion preference was lost between visits. | Persist and validate the setting, including an explicit zero. | Setting and startup checks pass. |
| A queued unique tool still appeared available to craft. | Disable it and label it “In production.” | Live crafting-panel check passes. |
| An imported diagonal position could be outside the radial movement boundary. | Constrain the restored position to the reachable radius. | Imported-position fixture passes. |
| Canceling capture through a menu could leave a captured drag pointer. | Release pointer capture alongside the existing key and motion reset. | Included in the controller cleanup; actual browser capture semantics remain unverified. |
| Queued recipes showed only their own duration, ignoring jobs ahead. | Show cumulative time at each workstation while keeping different stations concurrent. | 28/56-second forge jobs and a concurrent 25-second kiln job checked. |
| Direct forage/fishing rewards bypassed the common inventory bounds. | Route them through the simulation's bounded reward API. | Simulation award validation and controller integration checks. |
Seven pre-existing controller defects and one new resource-guidance requirement were captured in an eight-check baseline regression report. All eight pass after the changes. Additional usability and geometry checks cover the new filtering, queue estimates, persisted setting startup, and batching behavior.
Static building parts are now combined by material. Moving parts, flames, growing crops and lights retain their separate objects. A preview shares one transparent material, and its color updates only when validity changes.
| Building | Meshes before | Meshes after | Triangles retained |
|---|---|---|---|
| Cabin | 130 | 3 | 2,472 |
| Fish trap | 118 | 3 | 4,468 |
| Forge | 80 | 4 | 1,422 |
| Kiln | 77 | 3 | 1,368 |
| Well | 68 | 5 | 904 |
The geometry check compared every transformed vertex, triangle count and precise bounds for all 18 building models, allowing a small Float32 rounding tolerance. Animated references remain attached and an opened gate still rotates. Mesh reduction should reduce draw submission overhead, but no FPS or GPU-time improvement was measured.
These are design hypotheses, not survey results. Ingredient hunting, unavailable recipes, unexplained placement rejection and misleading queue times can interrupt play. This release adds ingredient source hints, search by ingredient, a “Can make now” filter, clearer placement reasons and accurate queue waits. Wrong-tool feedback is rate-limited to avoid flooding messages.
The synthetic policies also exposed their own limitations. Frequent blocked signal-fire interactions came from scripts retrying without enough fuel; wrong-tool errors were intentionally injected. The hunter and explorer policies spent much of their time traveling. Their modeled collapse rates were 28.12% and 7.32%, respectively, in both versions. Those numbers depend on assumed encounters, hit rates and movement paths. They do not establish that the game is too difficult or that humans would abandon it. No difficulty or resource rebalance was justified from those rates alone.
The paired progression summaries were effectively unchanged. Interface improvements were checked directly; the runner does not model human interpretation of the new UI and cannot quantify their effect on enjoyment.
Eight profiles have 2,500 cases each: guided survivor, first-time player, camp builder, hunter, explorer, forager, distracted worker and experimental player. Profile names describe scripted policies. The runner records each profile's explicit probabilities for self-care, wrong-tool choice, hunting accuracy and threats, plus travel and menu-dwell assumptions. These probabilities have no empirical player-study calibration.
Each case receives 1,200–1,799 seconds of active time. Menu dwell adds wall-clock time without advancing survival. Four local worker threads execute independent sessions. They are not simultaneous online clients and do not exercise server capacity.
The actual simulation, harvesting, recipes, repairs, tool conditions, needs, damage, production queues, save validation and structure collision code execute in the runner. Resource positions and finite animal stocks come from the real scene. Bots know recipes and source locations. Travel uses straight segments with building collision and a detour factor. Hunting accuracy and threat encounters are sampled; the 61-animal live ecology runs in separate CPU tests. There is no full navigation model, rendered camera, browser interaction or long-campaign model in the 20,000-case runner. Progression times are not human onboarding estimates.
Checked invariants include finite bounded inventory and resource values, non-increasing resource stocks during harvesting, bounded needs and preservation of inventory, resource work, queue count and building count through serialization. The raw per-case logs and runner should be inspected for the exact coverage. This is not exhaustive property verification of every save field.
Reproduction uses tests/monte-carlo.cjs, tests/fixtures/island.json, and the recorded source files. Run node tests/monte-carlo.cjs --players 20000 --source /absolute/path/to/dist --out /absolute/output. Run npm test for the smaller regression suites. No dependency installation is required.
The original baseline modules are included in tests/fixtures/baseline-2.0/; their hashes match the before-summary evidence. The baseline and after summaries, compressed per-case JSONL records, controller results and building measurements are retained in docs/audit-2.1/. Summaries record SHA-256 hashes of the runner, fixture and four executed source modules. Calibration runs were excluded from the reported results. The baseline and after runner hash and fixture hash match exactly.
The CPU suites pass for simulation, controller flows, resource depletion, ecology, geometry and interface behavior. The controller harness uses real Three.js scene data with a mocked WebGL renderer. It cannot verify shader compilation, browser pointer-lock behavior, audio, GPU memory, frame rate or visual quality. The standalone static project has no compatible supervised Sites browser-preview workflow in this environment, so no browser playthrough was performed.
The finite island can eventually be exhausted. These sessions cover 20–30 active minutes each; multi-hour progression, long-term saves and actual player reactions remain unmeasured. Buildings are complete plans, hunting is hitscan, fishing is a shoreline action and the companion is rule-driven. Portfolio presentation should describe these choices and limits honestly. A measured desktop playthrough and real player feedback would be the next evidence to collect before a public quality claim.
Tidehaven is an original browser survival prototype combining a procedural island with finite harvesting, crafting chains, wildlife and companion needs. I developed it with AI assistance and used reproducible testing to harden the implementation: 20,000 paired synthetic sessions, 110,087 save/load checks per run, targeted controller regressions and geometry-preserving building optimization. The placement audit eliminated 237 accepted obstacle overlaps in the matched sample, and cabin geometry was reduced from 130 mesh objects to 3 while retaining its triangles. The project demonstrates game-system integration and evidence-based iteration; human usability and GPU performance are separate validation steps.
This update expands the original playable browser game. It remains an original, stylized procedural game. It does not reproduce commercial game art or claim photorealistic rendering.
| System | Original build | This update |
|---|---|---|
| Resource types | 6 | 17 |
| Resource nodes | 343 | 591 |
| Item types | 21 | 54 |
| Crafting recipes | 8 | 33 |
| Building plans | 9 | 18 |
| Animal species | 2 | 17 |
| Individual animals | 12 | 61 |
New resources include bamboo, reeds, herbs, mushrooms, flint, iron, copper, clay, salt, shells, and fallen hardwood. Resource placement retains the original node identities for save compatibility.
Trees require repeated axe strikes and fall when cut through. Pickaxes open mineral deposits. Loose stones can be collected by hand to bootstrap the first pickaxe. Cutting and digging accumulate work; rewards arrive when a harvesting stage completes. Better tools work faster. Tools wear out and can be repaired. Reachable fruit and fibers can be gathered once before felling a tree.
Build a kiln and forge to progress from stone to metal. Drying racks cure hides into leather; smokers preserve food. Wells and baited fish traps support a lasting camp. Cabins have interiors and doorways, and timber gates open and close. Production takes active simulation time and pauses with menus. Station requirements refer to owned camp structures, not proximity to the station while using the crafting panel.
The world uses layered leaves, branches, roots, vines, bamboo, ground cover, a mangrove margin, coral, kelp, a rock arch, and a shipwreck landmark. Surface materials include procedural grain and bump detail. Water has animated waves, shoreline foam, reflections, and shallow transparency.
Wildlife includes boar, deer, goats, rabbits, monitors, vipers, mountain lions, gulls, parrots, ospreys, reef fish, turtles, rays, reef sharks, a tiger shark, a crocodile, and crabs. Birds flap, fish school, land animals flee or defend, and predators can hunt prey. Ospreys hunt fish and do not attack the player. Marine predators threaten swimmers. Wildlife kills are finite; predators can consume carcasses. Fishing rods, gardens, and baited traps offer ongoing food sources.
Diving uses Ctrl and surfacing uses Space. Breath drains underwater and recovers above the surface. A depleted breath meter causes damage. Spears work underwater; drawing a bow requires surfacing. Viper bites cause temporary poison, treated with herbal medicine. Carried leather armor absorbs some animal damage until worn out.
Mouse aim uses direct accumulated input. Mouse capture requests raw movement where supported. When capture is unavailable, right-drag uses client-coordinate deltas and pointer capture. Walking accelerates and stops smoothly. Head motion, sensitivity, field of view, and graphics quality are adjustable.
Menus use an icon sidebar, categorized cards, search, ingredient counts, selected-item details, equipment condition, and production queues. The quickbar has seven tool slots. The journal includes a wildlife field guide. Search fields and settings retain their keyboard input instead of triggering game shortcuts.
| Control | Action |
|---|---|
| WASD / Shift | Move / sprint |
| Mouse or right-drag | Look |
| E / left click | Interact or use held tool |
| 1–7 | Hands, axe, spear, bow, pickaxe, knife, torch |
| R | Repair held tool; rotate a building preview |
| Space / Ctrl | Jump or surface / dive |
| F / Q | Eat / drink |
| Tab / C / B | Backpack / crafting / building |
| M | Map |
| Esc / P | Pause |
Download Tidehaven.html and open it in a desktop browser with WebGL enabled. Engine code, art geometry, icons, styles, and the research report are embedded. No installation or network connection is required for gameplay. Source and the standalone download are both retained.
Existing version-1 JSON saves load into version 2. Resource depletion, per-swing harvesting work, wildlife state, inventories, tool condition, buildings, and production queues persist. Browser file origins can differ between downloads; export a save from the old game and import it in the new game's Settings if Continue does not appear. Menus, pausing, focus loss, and hidden tabs stop the simulation.
Verification uses the bundled Three.js implementation, real geometry, actual controller and generated menu handlers, and a mocked WebGL renderer. It covers simulation economics, tool wear, queues, finite resource and carcass rewards, save migration, input accumulation, paused resize redraw, keyboard focus, structure geometry, cabin/gate collision, shoreline trap placement, and three simulated minutes of ecology.
No real-browser rendering, shader compilation, screenshots, visual assessment, or frame-rate measurements were performed in this update. The drawing-buffer retention and resize redraw changes address a known blank-paused-canvas code path, but the reported screenshot issue is not confirmed resolved on the user's device. Graphics and mouse feel still require checking on the target browser. Buildings are complete plans; freeform wall-by-wall construction, complex human animation, and commercial-scale content are outside this build.
Three.js r160 is MIT licensed. Lucide icons are ISC licensed. Microsoft Barramundi Fish geometry is CC0, obtained from the official Khronos glTF Sample Assets repository; Tidehaven adds its own procedural coloring and tail deformation. Original photograph textures are not included. Other world and wildlife assets are generated from original game code. The full license notices and fish credit are included in the standalone file.
Tidehaven should make a small island feel inhabited, consequential, and worth returning to. Its first complete experience is a 15–30 minute browser session in which the player turns a bare shoreline into a functioning camp with a companion. Gathering, crafting, construction, food preparation, personal needs, and companion behavior must interact. A beautiful island with disconnected buttons would fail this brief.
The recommended identity is grounded survival with a warm domestic layer: explore in first person, establish reliable food and water, choose where to build, and watch another survivor develop a routine around the resulting camp. The player should experience both the satisfaction of securing a meal and the pleasure of improving a home. A single dense island supports this better than a large, mostly empty archipelago.
This is an original browser game, not a reproduction of commercial assets or an equivalent to years of commercial development. All Tidehaven timings, quantities, thresholds, behavior rules, and production priorities below are design recommendations, not claims about the reference games. The evidence describes the originals; the proposed implementation deliberately has a much smaller scope.
Endnight describes a first-person survival world where vegetation can be cut, materials become camps and fortifications, fire provides warmth, food can be scavenged or grown, and players alternate daytime exploration with nighttime defense. It also explicitly presents stealth and direct confrontation as alternatives. These are interacting activities: the environment supplies construction, construction changes exposure to danger, and danger influences the next gathering trip.[1]
The useful interpretation for Tidehaven is that gathering must visibly alter the place. A cut tree should disappear or leave a stump, a harvested bush should look depleted, and a built shelter should occupy real ground. The player's camp should be a physical record of decisions, with a fire, bed, garden, storage, and collector in locations the player selected.
The Forest's release notes provide a second lesson: quality lives in details. Its 1.0 notes discuss visual decay of meat, cooking appearance, clearer menu activation, fire lighting, map orientation, keyboard issues, and save-related defects. Endnight also describes four years of player-supported development before the April 2018 full release.[2] A browser prototype can adopt the standard of clear consequences; it cannot honestly promise that accumulated level of finish immediately.
The official Stranded Deep site emphasizes land and underwater exploration, material progression in housing, hunting, fishing, trapping, and survival pressure from hunger, thirst, exposure, and dangerous wildlife.[3] Its publisher-maintained Steam description adds a survival watch, injuries, crafting from inventory or nearby objects, skill advancement, customizable rafts, procedural islands, and farms with watering and plant growth.[4]
Tidehaven should adopt a compact supply chain: shore resources make tools; tools improve gathering and hunting; food needs a fire; reliable water and crops create surplus; surplus allows longer exploration and better housing. A water collector should therefore produce an observable resource rather than merely add a statistic. A garden should have visible growth states. Cooking should transform a raw item into a more useful one.
Stranded Deep's breadth should also constrain expectations. Its listed PC early-access release was January 2015 and its full release was August 2022.[4] Recreating the raft system, underwater exploration, bosses, extensive item catalog, and cooperative networking would dilute the first playable build. A shore and a few shallow-water interaction spots are enough for the first complete survival loop.
EA's main Sims 4 page describes distinctive personalities and aspirations alongside home layout, furnishings, and terrain editing.[5] The important bridge to survival is that a built object should enable an action and improve daily life. Shelter, seating, a cooking station, and crops should change what the inhabitants can do, not merely decorate the scene.
EA's Natural Living guide explicitly connects outdoor activities to hygiene, hunger, energy, fun, and social needs. This is an expansion feature, not a description of every base-game Sim.[6] Its relevance is the coupling: a single action can satisfy a practical requirement and a personal desire. Tidehaven can use a meal shared beside a fire to restore hunger and morale, while comfortable shelter improves recovery.
The Gardening guide gives a clear progression from planting, watering, and harvesting to more capable interactions; later activities can also improve fun and social needs.[7] The Fishing guide similarly connects skill growth to new opportunities, bait, and the ability to fish through difficult conditions.[8] Tidehaven needs only a small version: improved harvesting speed or meal quality is enough to make repeated work feel like mastery.
EA's March 2026 autonomy update describes adjusting preference, familiarity, distance, trait restrictions, and cooldowns to address repetitive object use and unwanted social behavior.[9] This is strong evidence that believable behavior requires selection and restraint. It is not evidence of a particular internal AI architecture, and it does not justify claiming a clone of that architecture.
The recommended Tidehaven companion uses simple local rules. It should choose a useful nearby activity, complete it, then reconsider. It should remember whether it recently ate, rested, socialized, or gathered. It should react to shortages and danger without thrashing between tasks. A visible status such as “Gathering wood” or “Resting at camp” makes the system understandable.
EA's historical introduction of Neighborhood Stories explains that traits influence autonomous decisions and that the active player can influence some changes, while also discussing controls on unwanted repetition.[10] Tidehaven can express the same broad principle with one companion and a handful of orders. It does not need simulated generations, careers, romance, or a town economy.
The following timing is a target for playtesting, not a promise about an untested build. The player should enter a scenic shoreline immediately, see an obvious first resource nearby, and be able to perform the first useful action within 30 seconds. Do not begin with a long cinematic, character editor, or explanatory wall of text.
| Session time | Player experience | System proof |
|---|---|---|
| 0–2 minutes | Look around, move, gather driftwood, stone, fiber, and a coconut | Controls, targeting, inventory, and needs have clear feedback |
| 2–5 minutes | Make a basic axe and spear; collect construction materials | Recipes consume real ingredients; tools improve available actions |
| 5–9 minutes | Place a shelter and campfire on chosen ground | Placement preview, cost, collision, and usable structures work |
| 9–14 minutes | Hunt or fish, bring food back, and cook | Animal response, raw/cooked states, fire interaction, and replenishment connect |
| 14–20 minutes | Assign a companion task; create a water collector or garden | Another survivor visibly contributes; camp creates ongoing output |
| 20–30 minutes | Rest through night, improve comfort, explore a landmark, save | Progress persists; a meaningful next-session goal remains |
Give the opening enough water and food to learn without a death spiral. Need decay should create a reason to act within the session, but a player reading controls should not be punished. The first night is an atmosphere and preparation milestone. If hostile wildlife is included, first-night pressure should be telegraphed and survivable with the taught equipment.
A short journal can present three current goals, automatically completing them from actual game events. The historical Sims goals guide describes objectives as a way to lead players toward available content.[11] The original guide concerns 2014 systems, so it is useful as historical design rationale rather than a current Sims feature specification. Tidehaven's goals should orient the player while leaving room to build freely.
Use five clearly named values: health, hunger, thirst, energy, and morale. Keep short-term exertion separate only if sprinting and tool use need a stamina bar. Otherwise, energy can handle exertion and rest without a sixth permanent meter. The numbers should have observable effects and accessible recovery routes.
| Variable | Main drains | Recovery | Consequence |
|---|---|---|---|
| Hunger | Time and exertion | Coconut, cooked food, garden produce | Low levels reduce recovery; critical levels gradually damage health |
| Thirst | Time and exertion | Coconut, collector water | Low levels limit exertion; critical levels gradually damage health |
| Energy | Running, harvesting, late hours | Rest, shelter sleep | Work slows and sprinting becomes limited |
| Health | Wildlife and sustained critical needs | Safe rest with adequate food and water | Incapacitation or recoverable failure |
| Morale | Isolation, discomfort, repeated hardship | Conversation, shared meals, comfortable camp | Modest work/recovery benefits and companion willingness |
Suggested initial tuning gives the player several minutes before the first need becomes urgent, with warnings before penalties. Do not synchronize all meters so the player must immediately solve everything. Show the cause of a penalty in plain language. Avoid claiming biological accuracy from normalized game meters or accelerated time.
Morale is the key connection to domestic play. It should matter enough that a shared meal or better bed is useful, but never become a frequent punishment for exploring alone. Hygiene and bladder can be later additions only if their actions are enjoyable and integrated. Adding six unrelated chores would make the game busier without making it deeper.
Start with wood, stone, fiber, raw food, cooked food, water, and seeds. Tools and built structures remain distinct inventory/world entities. Limit recipes to items that unlock an action, solve a recurring problem, or visibly improve camp. A three-stage progression is sufficient: emergency survival, reliable camp, comfortable home.
Recommended initial recipes are an axe, spear, campfire, shelter, water collector, storage crate, garden plot, bed, and a simple decorative seat or lantern. Display ingredient availability, the exact missing amount, and the resulting function. Crafting should be one deliberate action, with an animation or short completion beat. Do not hide recipe discovery behind trial-and-error combinations in the first release.
All material changes need one authoritative update path. Successful crafting subtracts ingredients once and creates one result. Canceling placement must not consume materials. A failed build must leave the inventory untouched. Companion gathering must draw from the same world resource model as player gathering, or be explicitly represented as an off-screen expedition with bounded returns.
The player needs a translucent placement preview, rotation, a clear valid/invalid state, and a way to cancel. Placement should reject water, excessive slope, overlap, and positions beyond reasonable reach. The constructed object must match the preview and survive a reload in exactly that location and orientation.
A compact catalog of complete structures is the simplest initial solution. A genuinely modular foundation, wall, and roof system is a larger second step because it introduces snapping, support, interiors, doorway clearance, and camera occlusion. If modular pieces are present, all must have actual collision and a usable purpose. An attractive hut that the player cannot enter is a decoration, not a shelter simulator.
Give structures interaction points: the fire cooks and warms, shelter supports sleep, collector stores water, garden accepts seeds, crate stores goods, and seat supports rest or conversation. Favor visible effects such as smoke, embers, filled containers, sprouts, and companions using furniture. These make the camp feel alive without a costly content catalog.
Use one huntable land animal and fish or crabs near the shore. The land animal should wander, notice a nearby player, flee or defend itself, respond to a successful strike, and produce a finite harvest. A stationary target with a resource label does not satisfy hunting.
An attack needs a readable range, cooldown, visible or audible contact, and a clear distinction between a miss and a hit. Wildlife should avoid steep terrain, water inappropriate to its species, and buildings. A few animals behaving coherently are preferable to dozens clipping through trees. Respawning should happen away from the player and after a delay so hunting cannot become an instant infinite food button.
Start with one survivor whose name, current activity, energy, and relationship are visible. Orders should be limited to Follow, Gather, Rest, and Help at Camp. If there is only time to implement three orders, Follow, Gather, and Rest provide a complete useful set. Avoid a text-chat panel unless the dialogue produces concrete in-game choices.
The behavior priority should be danger first, critical needs second, direct orders third, and autonomous camp routines fourth. A companion must not continue harvesting while visibly starving. A direct order can temporarily outweigh ordinary preferences but should never suppress emergency self-preservation. Brief explanations such as “Too tired to gather” make refusals intelligible.
For task selection, score a small set of available actions from need, distance, assigned role, preference, and recent repetition. Add a commitment period so a character walks to the chosen task and completes a meaningful chunk. Reevaluate immediately only when the target disappears or danger occurs. If navigation stalls, select another reachable target instead of walking endlessly into a wall.
Relationships can begin with a single trust value. Conversation, shared food, and comfortable living raise it slowly; unmet basic needs reduce it slowly. Trust should provide a modest benefit, such as slightly more efficient cooperation, rather than locking the core game behind repetitive conversation. Keep dialogue tied to circumstances: rain, hunger, a new shelter, a successful hunt, or a completed task.
The camp should generate small emergent stories: the player returns with fish while the companion has gathered enough wood to finish a bed; a storm fills the collector; poor sleep delays a morning expedition; a shared cooked meal restores both survivors. These outcomes are the desired life-simulation layer. They require consistent state changes, not an external language-model service.
Visual realism in this scope comes primarily from coherent scale, terrain, lighting, water, shadows, foliage density, and motion. Establish a convincing beach-to-interior transition, a strong horizon, shallow-water color variation, and enough landmarks to navigate without a constantly open map. Use restrained interface panels that allow the environment to dominate the screen.
The island should have a safe arrival beach, resource-rich grove, open building ground, a rocky lookout, and a shoreline food area. Each region needs a reason to visit and an identifiable silhouette. Scatter resources intentionally around this layout rather than uniformly. A small island that teaches navigation is more useful than procedural acreage with no distinctive locations.
Sound should begin only after a deliberate user interaction and have a readily available mute control. Ambient surf, foliage, footsteps, chopping, cooking, and wildlife calls can improve presence substantially. All essential information must remain visible when muted. Do not require headphones or voice input to play at work.
Pause is a core feature. Escape should release the pointer and pause the simulation; losing tab visibility should also pause. Returning after a meeting must not reveal a dead survivor or depleted camp. Save after meaningful changes and at regular safe intervals, with a visible manual save command. Clearly label browser-local persistence and provide export/import if implemented; do not imply cross-device sync without a real backend.
| Priority | Required outcome | Acceptance evidence |
|---|---|---|
| P0 | Load, move, look, pause, and resume reliably | A clean browser session enters play and handles pointer-lock loss |
| P0 | Complete the resource-to-meal-and-shelter loop | Gather, craft, place, hunt, cook, eat, drink, and sleep all change state correctly |
| P0 | Save and reload the changed world | Inventory, needs, depleted nodes, and placed buildings persist |
| P1 | Companion performs actual useful work | Orders change behavior; completed tasks produce bounded, valid results |
| P1 | Camp improves survival and comfort | Collector, crops, rest, and morale have observable consequences |
| P1 | Maintain readable performance | Test actual browser behavior, controls, and frame pacing on available hardware |
| P2 | Expand replay value | More recipes, landmarks, weather, wildlife, modular homes, and exploration after stability |
The auditor should test the whole loop without using developer shortcuts. Important edge cases are insufficient materials, canceled placement, repeated interaction, inventory limits if present, pausing during an action, a hidden tab, a reload after construction, dead or removed targets, and a companion whose path is blocked. A build is not verified merely because it compiles.
Balance needs a separate short playthrough. Record time to first useful action, first tool, first cooked meal, first shelter, and first sustainable water source. Watch for excess walking, unclear targeting, unreadable night lighting, and a recipe that blocks every later activity. If the player cannot complete the opening loop in roughly one break, reduce friction before adding content.
The honest target is a polished, playable browser prototype with coherent survival and domestic systems. It is not “the most realistic simulator,” a complete commercial-scale replacement, or validated on every work computer. Target hardware performance must be measured, and the final delivery should state which features actually work and which remain planned.
The official materials examined do not provide an open-source license for the shipped code and assets of these three reference games. Their public pages are evidence about design, not permission to copy their characters, maps, recordings, art, interface, or code. Tidehaven should use original implementation and assets with independently verified reuse licenses. Separate repository and asset review should record origin, license, attribution requirements, and any excluded material.
Keep runtime spending at zero: no cloud GPU requirement, paid asset dependency, metered dialogue API, or subscription needed to play. A downloaded engine or large repository is not inherently an advantage. Adopt technology only when it removes a concrete delivery obstacle. The first success criterion is a game the player can open and enjoy immediately, with a persistent camp and a reason to return.
All sources were accessed September 9, 2026. Dates below are publisher publication/update dates when available. Undated product pages describe the current accessible marketing text, not a fully audited list of every shipped mechanic. Historical notes are explicitly used as historical design evidence.
1. Endnight Games. The Forest. Undated official product page. Used for world interaction, construction, food, farming, daytime exploration, nighttime defense, and stealth.
2. Endnight Games. V1.0. April 30, 2018. Used for development duration, release context, visual feedback, interaction polish, and bug categories.
3. Beam Team / Stranded Deep. Stranded Deep official site. Undated; footer copyright 2020. Used for island survival premise, building tiers, hunting, fishing, trapping, hunger, thirst, and exposure.
4. Beam Team Games / Beam Team Publishing. Stranded Deep on Steam. Live publisher-maintained product listing; PC release August 10, 2022, early access January 23, 2015. Used for survival watch, injuries, crafting, skills, rafts, procedural islands, farming, and release dates. User review claims were not used.
5. Electronic Arts. The Sims 4. Undated official product page. Used for personalities, aspirations, home building, furnishings, and terrain modification.
6. Electronic Arts Help. The Sims 4 Skills: Natural Living. Updated April 2026 according to the indexed help record. Used for the connection between outdoor actions and personal needs; expansion content.
7. Electronic Arts Help. The Sims 4 Skills: Gardening. Updated April 2026 according to the indexed help record. Used for planting, watering, harvesting, skill progression, and links with fun/social needs.
8. Electronic Arts Help. The Sims 4 Fishing skill guide: levels, tips, and cheats. Updated April 2026 according to the indexed help record. Used for progression, bait, and fishing under difficult conditions. Individual expansion requirements should be checked before treating any listed activity as base-game content.
9. Electronic Arts / The Sims 4 Team. Bug Bash and Maker Packs. March 17, 2026. Used for autonomy changes involving preferences, familiarity, distance, traits, and cooldowns.
10. Electronic Arts. Introducing Neighborhood Stories. November 23, 2021. Historical source used for trait-influenced autonomous changes, player influence, pacing, and control principles; not a current complete feature inventory.
11. Electronic Arts / SimGuru Grant. Whims, Aspirations, and Goals: A Quick Guide. October 24, 2014. Historical rationale for goals that introduce available activities; not a current specification of Sims 4 whims or wants.
Research checked 9 September 2026. This is an implementation recommendation, not a claim that any proposed game has passed runtime testing. Only official repositories, repository notices, and project documentation support the factual comparisons below.
Use Three.js for a new, compact, single-player browser game with original geometry, original gameplay code, and a small deterministic simulation. Deliver a complete playable loop before increasing graphical scope. Browser delivery, zero spending, and playability on an ordinary work computer make this a stronger starting point than porting a native survival game. That recommendation is an engineering judgment. An engine license can remove software purchase cost; it cannot supply a commercial game's animation, level design, assets, tuning, or QA.
| Project | Verified license and target | Fit for this request |
|---|---|---|
| Three.js | MIT; JavaScript rendering library with WebGL and WebGPU renderers | Recommended for a purpose-built browser island. Permissive, no engine charge. Gameplay, collision, AI, persistence, and UI remain our responsibility. Preserve the copyright and permission notice. |
| Babylon.js | Apache 2.0; JavaScript game/rendering engine | Credible alternative when engine-provided game features justify integration. Maintain required notices and account for separately licensed assets. Switching engines after a working Three.js prototype would not create more realism by itself. |
| Luanti | Engine code LGPL 2.1 or later; voxel game platform | Mature world/mod architecture, but a different visual and technical direction. Engine licensing does not determine the license of its games or media. |
| VoxeLibre | GPL 3 or later code; mixed free media licenses | Substantive survival sandbox with gathering, hunting, mining, construction, and exploration. Appropriate to adapt only if a voxel game is acceptable and all applicable copyleft/media obligations are handled. |
| Cataclysm: Dark Days Ahead | Repository states CC BY-SA 3.0 for code/content, with separately listed font licenses | A major turn-based survival reference, not a browser 3D foundation. Useful for studying how many dependent survival systems create meaningful choices. Direct transplantation would create substantial porting and license work. |
| One Hour One Life | Public-domain copyright dedication with stated retained non-copyright rights | Substantive multiplayer survival and civilization-building reference. Strong conceptual reference for interdependence and family continuity; not a ready-made 3D island game. Do not represent a derivative as the author's official product. |
VoxeLibre's legal file is more specific than a repository badge: its code is GPL 3 or later, individual mods may offer alternatives, textures/sounds carry several Creative Commons licenses, and some assets have their own attribution. Its media should not be treated as a single unrestricted asset pack. No VoxeLibre code or assets are required for the recommended implementation. VoxeLibre legal notice.
Luanti browser work exists. The upstream paradust7/luanti-wasm repository describes itself as experimental and says its build was tested on Ubuntu 20.04. This prevents the false claim that Luanti cannot run in a browser. It also means a browser port is not automatically a maintained, drop-in, low-risk foundation. Its wrapper's MIT badge does not relicense the Luanti engine or games bundled into a build.
Terasology is an additional open voxel-world reference: Apache 2.0 code and generally CC BY 4.0 artwork, with exceptions noted by the project. Its Java, modular, multiple-repository workspace is a poor match for the immediate browser implementation. Unknown Horizons is valuable for settlement needs and production-chain design, but its GPL code, separately licensed media, and different engine architecture make it a research reference rather than a shortcut.
Realism should start with consequences. Trees must yield finite material. Consumed food must leave inventory. A shelter must affect rest or exposure. Rain must affect water collection and wetness. Hunting needs an observable approach, attack, reaction, and reward. Relationships need a state change and a consequence. A cosmetic bar with no gameplay effect does not fulfill a simulation promise.
Recommended scope is one coherent island, one controllable survivor, a few wildlife types, gathering, crafting, cooking, free placement of a useful set of structures, needs, weather/time, one companion, and local saving. Prefer consistent controls and a compact recipe set over dozens of shallow panels. Use different resources and travel risks to give the island reasons to explore. The Sims influence can be needs, mood, relationships, and furnishing utility; full household simulation, careers, generations, complex animation, and autonomous village life remain unfulfilled unless implemented and verified.
Art direction should be described honestly as stylized naturalism if it uses procedural low-poly assets. Water, lighting, shadows, fog, vegetation placement, scale, camera motion, and sound can make that scene convincing. They do not justify claims of photorealism or parity with The Forest. Blender is an asset authoring tool, not a substitute for gameplay engineering. A cloud-rendered Unreal game would introduce a different deployment and operating-cost problem.
Keep a dependency lockfile and an asset/provenance list. Public visibility is not permission to reuse. The Forest, Stranded Deep, and The Sims should supply research inspiration only unless explicit asset or code rights are established. The safest immediate route is original implementation with the selected engine's notice preserved.