Current line / 01
Hardware purchase screen
Turn the catalog into an actual buy decision.
Public development line / updated 2026-08-06
Roadmap
This is the working state of Scaling Laws, not a promise list. No 82% progress bars. A feature moves from foundation to playable, polish and finally shipped only when the label actually means something.
Tracked features
180
end-to-end game line
Shipped
76
checked off under the strict rule
Current line
48
playable, polish or active foundation
Categories
11
from simulation core to release
On the bench now
The next expensive decisions are becoming visible.
Hardware economics already exist underneath the UI. The current line is about letting the player actually rent, buy, configure and time those decisions, then tying that ownership back into the physical company.
Current line / 02
Server configuration
Accelerator, CPU, memory and fabric must fit together.
Current line / 03
Hardware launch timeline
Make waiting visible before it becomes expensive.
Current line / 04
Staff & larger offices
The starter house needs a real second rung.
180 features shown
Core / 11 items
Foundation
The rules that keep every campaign reproducible, saveable and testable.
10 shipped
1 now
Deterministic campaign clock
Campaign time advances through one GameDate timeline beginning on 1 January 2022.
#One ordered daily simulation tick
Deliveries, compute, market demand, bills, gates and solvency resolve in a fixed order.
#Saved deterministic random state
The same seed and the same decisions replay to the same numbers.
#Simulation separated from Unity
Economics and game rules run without UnityEngine so balance tests do not need a scene.
#Versioned save migrations
Old save shapes move forward one version at a time instead of being silently rewritten.
#Save sanitization
Loaded fields are clamped and invalid enum values fall back to legal states.
#Projection honesty flags
Estimated hardware and rival data stay distinguishable from known reference data.
#Training projection separated from outcome
A projected capability never becomes a deployed score until the run actually finishes.
#Playability campaign test
A scripted multi-year company must survive and stay competitive without trivially dominating.
#Cross-catalog consistency tests
Catalog sweeps catch impossible ranges, dangling references and dead research paths.
#Scene wiring guards
PlayMode checks protect critical visual references that EditMode simulation tests cannot see.
#Founder / 14 items
Founder & Company
The opening choices that shape cost, research, training and where the company operates.
8 shipped
5 now
Founder creation screen
Create the person behind the lab before the campaign begins.
#Seven founder skills
Development, Management, Teamwork, Concept, Software, Data Engineering and Safety.
#Neutral skill baseline at 20
Ignoring a skill is a real weakness instead of merely skipping a bonus.
#Skill XP from completed work
Experience is awarded for finished runs, research, upgrades and releases, never for idle time.
#Founder traits with downsides
Every opening trait trades one advantage against a real cost.
#Company identity choices
Opening company profiles change starting cash, reputation, data and default pricing.
#Custom company identity
Name and define the player's own lab rather than taking a preset.
#World region selection
Choose the region used to navigate the registration map.
#Sixteen registration countries
Countries differ on hardware access, tax, innovation and local competition.
#Corporate tax on operating profit
A loss-making company is not punished with tax on turnover.
#Regional hardware price modifier
Registration changes what owned silicon costs.
#Regional innovation modifier
Country choice changes research and upgrade timing.
#Local competition modifier
Registration changes the brand pressure in the market split.
#Founder appearance in the 3D world
Connect the creator identity to the physical starter home and later offices.
#Models / 19 items
Models & Data
From a blueprint to a trained model, a shelf decision and a live product.
8 shipped
8 now
Staged model creator
Build a model through Foundation, Scale, Data, Compute and Review instead of one giant form.
#Live capability projection
The creator shows how the last change moved projected capability.
#Training-time projection
The bill is not the only cost. Calendar time is visible before committing the run.
#Training-cost projection
The creator prices the run against cash on hand before the player commits.
#Scaling-law quality model
Parameter count and token count feed one Chinchilla-style quality function.
#Compute-optimal training pressure
Undertraining or overtraining can waste the same compute budget.
#Dataset catalog
Eight corpora feed the data side of the training plan.
#Dataset blending rule
Multiple corpora combine through one normalized mechanism.
#Five audience segments
Consumer, Developer, Enterprise, Creative and Autonomous demand change over time.
#Five model types
Specialized model types trade broad reach for stronger audience fit and price tolerance.
#Shelf before release
A finished run becomes a trained model on the shelf, not an automatic launch.
#Manual release timing
The player chooses when a trained model reaches the market.
#Per-model pricing
Multiple live models can carry different prices.
#Eleven model traits
Reasoning, knowledge, coding and other traits are measured against the market par of the day.
#Trait upgrades consume time and compute
Model maintenance competes with training for the same calendar and infrastructure.
#Production serving descendant factor
Inference is priced as served descendants rather than the full training artifact.
#Safety stage in model creator
Make safety an explicit creation-stage decision before training.
#Keep local vs release flow
A finished run should branch into internal use or public release instead of one fixed path.
#Serving-capacity decision after training
Choose rented serving capacity or owned infrastructure after deciding to release.
#Research / 14 items
Research & Architecture
Technology unlocks, in-house architecture work and the calendar cost of getting there.
7 shipped
5 now
Technology tree
Seventeen nodes gate architectures, data, model types, upgrades and compute tiers.
#Four research eras
Foundations, scaling, autonomy and the late frontier structure the tree.
#Research prerequisite graph
Every unlock must be reachable and cannot open before its own prerequisite.
#Calendar gates
Research cannot be brute-forced purely by having more cash.
#Cash cost on research
Time is hard to buy, but research still competes for company capital.
#Artificial Superintelligence node
The end-state node is visible early but intentionally unreachable in the opening years.
#In-house architecture designer
Create a custom family instead of only adopting a published architecture.
#Five architecture research directions
Sparsity, throughput, quality-per-parameter, serving cost and reasoning divide programme focus.
#Budget × calendar power model
More money cannot fully substitute for more time, and more time cannot replace all budget.
#Focus dilution
A programme chasing every direction makes less depth in each one.
#Architecture outcome variance
Cheap or rushed research is deliberately less predictable.
#Six custom family slots
House architectures resolve through the same downstream interfaces as published families.
#Iterate an owned family
Improve a house family more cheaply and quickly, with diminishing returns.
#Research-era visual polish
Finish the visual language and card art around the four research eras.
#Hardware / 22 items
Hardware & Compute
The current active development line: rent, own, configure, wait and sometimes regret the timing.
10 shipped
5 now
Hardware generation catalog
Accelerators, host CPUs, node memory and fabric carry launch dates, price, power and capacity data.
#Known vs projected hardware flag
Future entries can be labelled as projections rather than blended into history.
#Time-based hardware depreciation
Owned accelerators lose resale value as calendar time passes.
#Successor depreciation
Meaningful newer parts apply another phased hit after the player's purchase.
#Performance-per-dollar aging index
The fleet can be compared with what the same money buys today.
#Rented compute in petaflops
Cloud rental is capacity, not a fragile count of whatever hardware generation happens to be current.
#Cloud hardware availability lag
Rental availability follows the hardware frontier with a delay.
#Three compute tiers
Rented cloud, colocated servers and own datacenter form the infrastructure ladder.
#Owned-cluster balance factor
Accelerators need enough host CPU, memory and fabric to approach their nominal rating.
#Training scaling-efficiency loss
Large accelerator counts lose efficiency as the fleet scales.
#Inference utilization model
Serving and training do not pretend to use the same hardware in the same way.
#Hardware purchase screen
Turn the existing catalog and economics into a player-facing buy flow.
#Hardware sell flow
Make resale value visible before the player exits an owned batch.
#Server configuration
Pair accelerators with host CPUs, node memory and fabric instead of buying one magic server score.
#Hardware launch timeline
Surface generation timing so waiting becomes an explicit decision rather than hidden knowledge.
#Colocation acquisition flow
Buy owned hardware with lead time before the company is ready for a full datacenter.
#Own-datacenter acquisition flow
Turn the late compute tier into a real capital and infrastructure commitment.
#Supply ramp after launch
A new accelerator generation should not become infinitely available on day one.
#Supply squeezes and shortages
Make availability another reason a technically good purchase can fail operationally.
#Physical server racks
Owned compute should appear in company spaces instead of living only in a table.
#Datacenter power planning
Owned infrastructure eventually needs limits beyond purchase price.
#Datacenter cooling constraints
Cooling becomes a late physical limit if the infrastructure simulation grows deep enough.
#Rivals / 16 items
Rivals & Intelligence
Competitors that react to timing, plus information that can cost money and still be wrong.
12 shipped
2 now
Eight rival labs
The active field is agent-driven rather than one static score table.
#Historical release seed timeline
Known releases seed the 2022–2026 opening before agents diverge.
#Patient rival waiting
Some rivals delay a run when better silicon is close to the planned launch window.
#Rival rush response
A frontier-focused lab can ship early with a quality penalty when the player opens a large gap.
#Rival capability drift
Competitors improve between releases instead of freezing until the next launch.
#Procedural post-history releases
After the known timeline ends, rival strategies generate their own cadence.
#Parody rival naming
Labs remain recognizable as jokes without using real company names.
#Parody rival logo set
Finish a consistent small-mark system that remains readable at UI scale.
#Three paid intelligence tiers
Information quality improves with spend without turning into perfect foresight.
#Real accuracy vs stated confidence
A report can sound more certain than the underlying signal deserves.
#Hardware launch signals
Intel can surface an upcoming generation before the player has certainty.
#Price-collapse signals
Intel can warn about a market event that changes hardware economics.
#Supply-squeeze signals
Intel can warn that availability is about to tighten.
#Rival-holding signals
Higher-tier intelligence can flag that a competitor may be waiting on purpose.
#Expanded intelligence report UI
Give evidence, history and uncertainty more room than a single note card.
#Competitor release history view
Show how each lab actually moved through the campaign, not just where it stands today.
#Business / 20 items
Business & Capital
Turn model quality into demand, demand into cash and cash into the next dangerous decision.
12 shipped
6 now
Per-model token pricing
Each live model carries its own price instead of inheriting one company-wide number.
#Free-tier control
Free access can be used as a deliberate distribution decision.
#Shared market-share model
The economy and ranking read from the same demand model.
#Capability term in demand
Better models earn an advantage but do not automatically own the market.
#Brand term in demand
Reputation changes how equal products split demand.
#Price term in demand
Price matters independently of capability.
#Model-age term in demand
A model that nobody replaces gradually becomes stale.
#Audience-segment curves
Consumer, Developer, Enterprise, Creative and Autonomous markets change size over time.
#Specialist price tolerance
A narrower model can reach fewer people while selling to an audience that tolerates higher pricing.
#Company marketing programmes
Marketing spend acts on demand instead of being a decorative button.
#Model launch push
A release can receive focused marketing rather than only permanent company spend.
#Funding valuation
Valuation combines frontier proximity, run rate and the sentiment of the year.
#Investor sentiment cycle
The same company can be priced very differently depending on when it raises.
#Cap table and founder dilution
Equity rounds permanently change ownership.
#Down-round penalty
Raising below the previous valuation costs extra equity.
#Debt facilities
Bridge, venture debt, bond and sovereign compute debt trade dilution for repayment pressure.
#Arrears and default
Missed payments accumulate before a public default damages standing.
#Ranking board
Capability, share and brand feed one sorted company board.
#Ranking history
Track movement through time instead of showing only today's order.
#Contextual market events
Add discrete events only where they reinforce an existing market mechanism.
#World / 23 items
Office, Staff & World
Make company growth visible in rooms, people, cars, buildings and infrastructure.
1 shipped
5 now
3D starter home
The opening company site exists as a real orthographic 3D environment.
#Starter living space
Living room, workspace and bedroom sell the scale of the first company before expansion.
#Garage
The starter site includes a garage connected to the planned travel flow.
#Starter car
The first vehicle gives the world-map transition a physical starting point.
#Fixed orthographic camera
The scene uses a locked 2.5D presentation rather than free camera navigation.
#Room visual-polish pass
Keep refining props, lighting and composition without changing the core camera.
#Wall hide / transparency
Remove room obstruction when the founder enters the garage or another hidden space.
#Founder 3D character
Place the created founder into the physical company view.
#Founder idle animation
Give the opening scene life without turning movement into simulation authority.
#Founder waypoint walking
Move between fixed activity points rather than overengineering pathfinding for the camera distance.
#Walk to workstation
Training and work actions gain a short physical animation at the desk.
#Walk to garage and car
Travel begins with the founder physically leaving the starter site.
#World / location map
Turn location selection into a navigable world layer beyond company registration.
#Rent larger offices
Company growth should require leaving the starter house.
#Office tiers
Progress through larger spaces instead of one infinitely expanding room.
#Hiring employees
Add the team layer that turns empty office capacity into useful company capacity.
#Employee skills
Staff quality should influence work rather than every hire being interchangeable.
#Employee salaries
Headcount becomes a continuing operating commitment.
#Physical employee desks
A hire should appear in the space the company is paying for.
#Walking staff
Employees move between work points to keep larger spaces alive.
#Owned racks populate the office
The physical view reads the compute pool and shows what the company actually owns.
#Dedicated company HQ
A late office tier separates the management fantasy from the starter site.
#Dedicated datacenter site
Owned infrastructure eventually becomes a place rather than only a finance line.
#Safety / 09 items
Safety & Incidents
Safety should be a decision with tail risk, not a virtue stat that only moves brand.
2 shipped
1 now
Founder Safety skill
Founder creation already changes incident-risk pressure through the skill system.
#Model Safety trait
Safety is one of the traits measured against the market par of the day.
#Safety brand effect
Falling behind on safety already affects how a product is received.
#Incident-risk multiplier foundation
The simulation has the inputs needed to turn weak safety choices into tail risk.
#Safety stage in model creator
Make safety visible before training instead of burying it in later upgrades.
#Model safety incidents
A model far below the safety expectation can produce a costly event.
#Reputation damage from incidents
Incidents should hit standing that took time and money to build.
#Regulatory attention
Serious safety failures can create pressure outside the normal demand model.
#Incident response decisions
Give the player a recovery choice rather than only applying an automatic penalty.
#UI / 14 items
Interface & Presentation
The game needs to stay legible while the simulation gets denser.
0 shipped
9 now
Bottom control HUD
The main navigation lives in a bottom control bar instead of a wide left rail.
#Day clock dial
A half-disc shows date and day progress without controlling the simulation rules.
#Pause / X1 / X2 / X3
Time controls map cleanly onto the whole-day simulation.
#Skip Day
Run exactly one day and stop, even when the game is paused.
#Unified accent gradient
Coral, wine and pale violet are treated as one line moving through the interface.
#Bottom category icon set
Finish the white single-weight icon language across the major screens.
#Skill icon set
Seven founder-skill icons are already present and being normalized as one family.
#Section banner art
Research, Business, Funding and other screens share low-contrast panoramic art.
#Responsive UI Toolkit layout
Creators and panels defend against shrink, wrap and small-window layout faults.
#Tooltip pass
Move long explanations out of cramped controls without hiding decisions.
#Resolution and scaling pass
Test the game across the target window and display sizes.
#Accessibility pass
Review contrast, focus, motion and readability before a public release.
#Audio feedback
Add restrained confirmation and warning sounds once the interaction language is stable.
#UI motion pass
Animate state changes only where motion clarifies what changed.
#Release / 18 items
Public Build & Release
The public footprint around the game: source, website, roadmap, playtests and store pages.
6 shipped
1 now
Public source repository
The code and engineering notes are visible while the game changes.
#HCK Labs hub
Scaling Laws has a permanent home beside the other HCK Labs work.
#Scaling Laws landing page
The public game page separates current systems from planned work.
#Development mechanism page
The simulation rules and real-world references have their own crawlable technical page.
#Public roadmap
This page is the living status map and the JSON behind it is the source of truth.
#Roadmap JSON data source
One structured file drives the roadmap state and can feed other presentation formats.
#Devlog index
Create a home for build notes that can be indexed outside social platforms.
#Media gallery
Collect screenshots, room-evolution clips and later trailers in one crawlable place.
#itch.io project page
Prepare the first game-specific public page and devlog channel.
#IndieDB project page
Add a second game-specific public footprint with indexed development updates.
#First public playable build
Move from screenshots and source to something outside testers can actually run.
#Closed playtest loop
Collect focused feedback before a wider demo.
#Feedback capture
Give early players one clear place to report friction and balance problems.
#Steam Coming Soon page
Start collecting wishlists once the visual direction and core loop can sell the game honestly.
#Playable demo
A public slice of the loop that proves the game beyond screenshots.
#Gameplay trailer
Cut the strongest in-game stories once the footage can represent the real loop.
#Release candidate
Freeze the feature line and spend time on bugs, balance and compatibility.
#Commercial release
Ship when the game is stable enough that the roadmap no longer has to explain away missing fundamentals.
#
Nothing matches this view. Clear the search or open another lane.