Build a system as a runnable diagram
Represent sources, stores, sinks, conversions, probabilities, gates, and feedback so dependencies no longer live only in prose or a spreadsheet.
O01O04Machinations
Turn a game-system question into a diagram you can run before you build it.
Evidence statusPublic-evidence reviewMedium confidenceLast checked Sep 1, 2026By MakeGameWithAI
PRODUCT OVERVIEW
What Machinations is—and how it works.
Machinations is a browser-based modeling and simulation tool for game systems and economies. Ask AI Builder to draft a diagram from text, or connect sources, resource pools, drains, converters, probabilities, parameters, and feedback loops yourself. Step and Interactive Play expose how resources move; repeated Predictions and charts show distributions and extreme paths under the assumptions you entered.
AI Balancer can search parameter combinations toward one numerical target. Google Sheets supports parameter handoff, and an open-beta Unity route can connect selected model values with a project. These are tools for making rules and consequences discussable—not an automatic certificate that a game is balanced, fun, fair, or ready to ship.
HOW YOU USE IT
Use it in a desktop browser. Community starts without a credit card but makes the account and diagrams public under CC0; private work begins with a paid account. Every simulation, formula, resource movement, and AI context can consume Events.
Step through or sample many outcomes
Use Step and Interactive Play to inspect paths, then Predict and charts to compare distributions. More playthroughs reduce sampling noise, not modeling error.
O05O06O07Draft and search with AI
AI Builder lowers the blank-canvas barrier, while AI Balancer searches up to ten influencers toward one target. Humans still define the target and judge alternatives.
O01O08O09O15Hand values into the next workflow
Sheets, CSV, and Unity paths can move selected data onward, but synchronization scope, ownership, and validation must be planned explicitly.
O10O11O12O13O14INTERFACE & EXAMPLES
See how it works
Machinations' official Balancer illustration combines a target metric, value distribution, and balancing controls.
Official source retrieved September 2026; interface version not specified.Source: Machinations
REVIEW SCOPE
This page evaluates whether Machinations is worth trying for one bounded game-system question. It does not measure simulation correctness, current AI quality, typical Event usage, Unity performance, player prediction accuracy, or business outcomes.
- Beginner friendlinessSome basics needed
- AI and templates offer a starting point, but meaningful system models require understanding nodes, resource flows and step rules.
- Agent integrationLimited integration
- The official Socket.IO API reads and updates existing diagrams; it does not establish an end-to-end agent workflow for AI diagram creation and simulation.
Scope, integration requirements and sourcesChecked
Beginner scope: Build and interpret a small resource loop or economy model.
Agent scope: A design-to-simulation workflow; the confirmed interface covers data interaction with existing diagrams.
Confidence: learning Medium · Agent Medium
- Machinations API ServicesOfficial
- Access
- Requires a User API Key and manually generated Diagram Token; confirm current account entitlement.
- Billing
- Diagram execution remains subject to plan and Event conditions; this document does not establish complete API pricing.
- Getting results
- Connect over WSS, retrieve diagram elements, send updates and listen for changes.
- Limitations
- In-product Agentic Simulations is not external MCP. API details were retrieved from the official search index; the direct page returned only its shell.
These are independent editorial judgments, not an overall score or a claim of hands-on agent integration testing.
- Machinations API Services (official indexed documentation) · Official · Learning evidence · Integration evidence
KEY FINDINGS
What the evidence supports—and what it does not.
User reportsMedium confidence
The diagram is useful because it exposes assumptions
Across a small-game review, a Defold example, an economy analysis, an academic project, and a management-game report, users repeatedly found value in turning loops and dependencies into something runnable. The benefit was often discovering tradeoffs and better questions—not receiving a final prediction.
U01U03U05U06U08Editorial inferenceMedium confidence
No-code entry does not remove systems thinking
Text-to-diagram and visual nodes lower the starting barrier. Independent reports still surface a learning curve around nodes, values, formulas, abstraction, and deciding what a Step means. AI can draft structure; it cannot decide which rules faithfully represent the game.
O01O04U01U03U04Editorial inferenceMedium confidence
Start with one costly question, not the whole game
A very small team may get the same answer from Lua, a spreadsheet, or a rules engine. A commissioned project also found integration unjustified while mechanics were still moving. Scope the first model around one decision whose implementation or reversal would be expensive.
U04U07U08Official factHigh confidence
More simulations do not make a wrong model right
Monte Carlo playthroughs reduce random sampling noise inside the model. They cannot repair missing behaviors, incorrect parameters, or a false assumption about players. Important conclusions still need a hand check, prototype, playtest, or real data.
O06O07U05U08Official factMedium confidence
AI Balancer optimizes a target, not fun
The documented Balancer searches up to ten influencer values around one target. That can narrow parameter candidates, but the target, constraints, simulation horizon, and comparison criteria are human choices. Current independent evidence is too thin to claim a reliable complex-game success rate.
O08O09U03Editorial inferenceMedium confidence
The diagram and the game can become two sources of truth
Sheets and Unity create handoff routes, but current Unity coverage is Open Beta and real-time sync is limited. Parameters, formulas, experiments, and analytics can drift between the diagram and implementation unless ownership and validation are explicit.
O10O11O12O13O14U07U09EDITORIAL VERDICT
Our take
Medium confidenceMachinations is worth a small, low-risk trial when a game has interdependent resources, random rewards, progression curves, production chains, or feedback loops that would be expensive to implement incorrectly. A non-programmer can begin with text and visual nodes, but trustworthy work still requires clear rules, Step definitions, probabilities, and independent validation. Start with one costly question—not a second representation of the entire game.
Conclusion scopeMedium-confidence fit judgment from public evidence. The supported claim is that a bounded model can clarify assumptions and compare scenarios; it is not a ranking, an accuracy certification, or a claim that AI can balance a complete game.
BETTER FIT
Worth auditioning when
- Resource loops, progression, random rewards, crafting or production chains, and multi-currency economies with meaningful dependencies.
- A designer or beginner who can explain what happens each step, where resources come from, where they go, and what changes probability.
- A project with one costly-to-implement question that can be checked before committing engineering time.
- A team willing to maintain assumptions, compare the model with a prototype, and validate important conclusions with playtests or live data.
POORER FIT
Do not depend on it yet when
- A simple system already answered adequately by a spreadsheet, hand calculation, or small script.
- Rules changing so quickly that keeping a separate diagram current would cost more than it saves.
- A confidential unreleased economy placed in a free public Community account.
- Anyone expecting one prompt, more Monte Carlo runs, or one optimized number to prove that players will find the game fun or fair.
WORKFLOW FIT
Where it fits in a production workflow.
Before implementation
Model one resource loop, progression curve, drop system, or production chain and compare a few explicit scenarios before paying the engineering cost.
GuardrailDefine the question and what one Step means before adding detail.
During prototyping
Use the diagram to explain surprising behavior and select parameters worth trying, then implement them in a playable build.
GuardrailTreat charts as model output; validate every important claim in the prototype.
For handoff and live tuning
Move selected values through Sheets, CSV, or the conditional Unity path when the maintenance benefit is clear.
GuardrailAssign one owner, document sync scope, keep backups, and compare against playtests or telemetry.
RECOMMENDED WORKFLOW
Use this tool inside a complete production workflow.
Start from a testable brief, then move through prototyping, assets, audio, testing, and release with an explicit handoff and human check at every step.
SHORTEST RESPONSIBLE PATH
Measure one question, one model, and one repeatable run
This is a low-risk starting workflow synthesized from the evidence.
- 01
Write one decision question, such as whether difficulty spikes within 50 levels, and define exactly what one Step represents.
- 02
Use synthetic, non-confidential values to build the smallest loop in Community. Do not upload an unreleased economy, client material, or a core secret design.
- 03
Run step by step until every resource movement and trigger behaves as intended; only then run a small Prediction and inspect unusual paths.
- 04
Repeat the same task once, read its Event tracker, and use that measured E to calculate Starter or Pro capacity. Do not borrow another diagram's Event count.
- 05
Sanity-check one result with a spreadsheet, hand calculation, or small script, then validate the important conclusion in a playable prototype or real data.
- 06
Pay for privacy or expand the model only when repeat simulation or collaboration clearly outweighs subscription and maintenance cost; retain parameter and result backups.
PRICING & RIGHTS
What Event quotas buy—and why one useful answer has no fixed price
Pricing and terms last checked: Sep 1, 2026
- Community is $0 with 50,000 Events per user per month. The account and diagrams are public, can be viewed, run, and forked, and Community content is dedicated under CC0.
- Starter is $20 monthly for 200,000 Events, or $180 paid annually with a displayed $15 monthly equivalent. It provides private-by-default work for solo creators and flexible collaboration.
- Pro is $200 monthly for 4,000,000 Events, or $1,800 paid annually with a displayed $150 monthly equivalent. It adds team and performance allowances; more quota is not a guarantee of more accurate conclusions.
- Events can be created by node triggers, formulas, and resource movement. One transfer already includes send, move, and receive events; diagram complexity, steps, playthroughs, and AI context can multiply use.
- Monthly Events reset and do not roll over, including on annual subscriptions. Top-up Events are consumed after monthly quota and do not expire; final package, discounts, tax, and checkout amount were not verified.
- AI Builder consumes Events from prompts, the current diagram, uploaded files, simulation and chart data, and response length. No public raw-token-to-Event formula establishes a fixed number of AI requests.
- The annual help text says 17% off, while the current displayed $180 and $1,800 amounts equal three months free versus monthly payment. This page uses explicit current amounts and preserves the conflict.
COST TO OUTPUT
First measure E Events for one repeatable task
Monthly nominal allocation assumes the whole included quota is used: Starter $20 ÷ 200,000 = $0.10 per 1,000 Events; Pro $200 ÷ 4,000,000 = $0.05 per 1,000 Events. Actual payment is still $20 or $200 even if you run once, and unused quota raises the effective unit cost.
O02O03At full use, 1,000 included Events carry a $0.10 nominal allocation. This is an allocation inside a subscription, not a pay-per-run checkout.
200,000 ÷ 10,000 fits 20 theoretical runs, and $20 × 10,000 ÷ 200,000 allocates $1 to each. It does not say a run answers the question or that 10,000 is typical.
4,000,000 ÷ 10,000 fits 400 theoretical runs, and the monthly allocation is $0.50 each. The real subscription threshold is still $200 per month.
A high-use task can consume Starter quickly. Four runs are not four complete economies, and the scenario is illustrative rather than measured product usage.
Community, Starter, and Pro display $0.00020, $0.00016, and $0.00010 per Event before applicable volume and tier discounts. Checkout totals and tax were not tested.
Use total subscription and top-up spend plus modeling, debugging, maintenance, and validation time, divided by conclusions later accepted and confirmed in a prototype or real data. No acceptance rate makes that denominator available today.
The 1,000, 10,000, and 50,000 Event scenarios are sensitivity arithmetic—not typical diagram sizes, measured bills, or promises of a useful answer. Failed-request charging and automatic Event refunds were not established. Measure your own repeatable task before choosing a plan.
Commercial-use condition
Paid-account content is private by default and user content remains owned by the user under the service license. Community accounts and diagrams are public, and their content is automatically dedicated under CC0; later privatizing or deleting it does not revoke rights already received by people who accessed it. Terms section 4 limits Community to personal, nonprofit, or academic use, while section 7 also names professionals below $100,000 in revenue or funding. Because that eligibility conflicts internally, use allowed_with_conditions: confirm the applicable contract for commercial work and use a paid private account for confidential designs. This is not legal advice.
Subscriptions renew automatically, monthly quota does not roll over, and the service is not a backup. Current privacy text predates the AI-first entry point and does not clearly identify AI model providers, training use, a dedicated retention period, or an opt-out. Use synthetic non-confidential data for a free trial, keep your own copies, and verify checkout and account-specific terms before paying.
PRODUCTION RISKS
Resolve these before production use.
Community means public and CC0
A free diagram is not merely visible—it is dedicated for broad reuse under CC0. Do not place an unreleased economy, customer data, or a proprietary core loop there. Moving private later cannot recover rights already granted.
O16A precise chart can still answer the wrong question
Simulation output inherits the model's structure, Step meaning, parameters, and player assumptions. Use independent checks and treat unusual certainty as a reason to inspect the model rather than trust it more.
O05O06O07U05U08Maintenance can erase the modeling benefit
When rules change, the diagram, spreadsheet, engine values, tests, and analytics can drift. Unity is still Open Beta and sync covers selected structures. Assign ownership and define which representation wins.
O10O11O12O13O14U07U09AI data terms are not current enough to assume confidentiality
AI Builder can read prompts, diagrams, uploaded files, and simulation data, while the published privacy policy predates this workflow and the subprocessor list does not name an AI provider. Get current contractual answers before sensitive use.
O03O16O17NOT VERIFIED
Claims this page does not make
- Current AI Builder success rate on complex game rules and AI Balancer's rate of finding parameter sets a project actually adopts.
- Typical Event consumption for a diagram, Prediction, Balancer run, or AI request, including failed-request charging and any automatic refunds.
- Average cost and time per design conclusion later confirmed by a prototype, playtest, telemetry, or live operation.
- Current Unity Open Beta compatibility, synchronization coverage, stability, and runtime performance for a specific engine version and project.
- Error between a model's player assumptions and real behavior, plus enterprise-contract and AI-processing details for a specific account.
RESEARCH METHOD
Research-reviewed from public evidence
Research-reviewed from public evidence. We checked 18 official sources and coded 9 independent records across 8 effective platforms and 6 source environments: review platforms, an engine community, professional analysis, academic projects, a broad design community, and a competitor analysis. Fourteen platforms or site groups were searched. Defold contributes 2 of 9 records (22.2%); Reddit contributes 1 of 9 (11.1%). Invited or incentivized review, vendor support, historical pricing, anonymous identity, and competitor bias are disclosed and down-weighted. Conflicting reports about value, complexity, precision, project stage, and model drift are retained. Confidence is medium for bounded workflow fit and low for current AI quality, Event distribution, Unity production behavior, and accepted-conclusion cost.
Research windowOfficial facts, pricing, terms, and independent evidence were last checked September 1, 2026. Nine independent records span 2023–2026; current AI Builder, AI Balancer, post-Events usage, and Unity Open Beta evidence remains thin.
OFFICIAL SOURCES18
- O01Machinations product homepageChecked Sep 1, 2026
Current text-to-playable-diagram entry point and product positioning. Efficiency and accuracy claims are not treated as measured outcomes.
- O02Machinations pricingChecked Sep 1, 2026
Community, Starter, and Pro prices, monthly Event quotas, top-up rates, visibility, performance, and current annual amounts.
- O03Machinations Events and Billing PolicyChecked Sep 1, 2026
What creates Events, AI Builder metering, top-ups, shared-diagram charging, and quota expiry. Its 17% annual-discount wording conflicts with current prices.
- O04What is Machinations?Checked Sep 1, 2026
Browser-based visual modeling, simulations, charts, AI, Sheets, and CSV capabilities. Vendor outcome claims are excluded.
- O05Machinations Play ModesChecked Sep 1, 2026
Step, Interactive Play, and Predict behavior. A Step is defined by the model author and is not automatically a real-world time unit.
- O06Monte Carlo SimulationsChecked Sep 1, 2026
Repeated random sampling reduces sampling noise but cannot repair incorrect structure, parameters, or player assumptions.
- O07Machinations ChartsChecked Sep 1, 2026
Mean, median, range, PNG, and plan-dependent CSV outputs. Charts describe the model, not observed player behavior.
- O08AI Balancer OverviewChecked Sep 1, 2026
AI Balancer searches up to ten influencer values toward one selected numerical target; that target is not a definition of fun or fairness.
- O09Interpreting AI Balancer resultsChecked Sep 1, 2026
Candidate parameters and averages depend on the target, steps, and simulation count; alternatives still need human comparison and validation.
- O10Google Sheets integrationChecked Sep 1, 2026
Documented import and export route. Changes need another import or save, so the sheet and diagram do not remain synchronized automatically.
- O11Game Engine Plugin overviewChecked Sep 1, 2026
Current ready-made plugin coverage is Unity Open Beta; Unreal remains on the roadmap. A roadmap is not a delivery date.
- O12Unity Integration GuideChecked Sep 1, 2026
Unity Open Beta setup and ScriptableObject handoff. This review did not install or benchmark the integration.
- O13Machinations Real-Time SyncChecked Sep 1, 2026
Current sync scope includes Pools and Resource Connections; it does not make every node, formula, or runtime state production game logic.
- O14Ruby's Adventure Unity exampleChecked Sep 1, 2026
Official example repository showing one integration route. Its README labels some formula handling very early and it is not independent production evidence.
- O15Machinations changelogChecked Sep 1, 2026
Tracks the August 2025 Events transition and 2026 AI Builder, Agents, Dashboard, and AI changes used to date older reports.
- O16Machinations Terms and ConditionsChecked Sep 1, 2026
Paid and Community content rights, account eligibility, renewal, cancellation, backup, and privacy text. Community eligibility clauses conflict internally.
- O17Machinations subprocessorsChecked Sep 1, 2026
Lists hosting, support, analytics, payment, and email providers. It does not identify the AI model provider or settle AI training and retention questions.
- O18Machinations FAQChecked Sep 1, 2026
Useful conservative limits and support context, but several agent, prediction, and SSO entries lag the current product and changelog.
INDEPENDENT EVIDENCE RECORDS9
- U01G2: small-game RNG and loop reviewChecked Sep 1, 2026
A verified small-game developer reports checking RNG and gameplay loops before prototyping, while still needing code or logic understanding for nodes and values. The review was invited and incentivized.
- U02Trustpilot: historical studio-adoption limitChecked Sep 1, 2026
A creative director liked the tool but said the then-current 100-component limit blocked studio adoption. The August 2025 Events model removed that specific limit, so this is historical change evidence.
- U03Defold Forum: first puzzle-progression modelChecked Sep 1, 2026
A first-time user turned a 50-level puzzle difficulty and churn question into a diagram and result, finding the workflow more suitable than Excel for that example. No later accuracy validation was provided.
- U04Defold Forum: overkill counterpointChecked Sep 1, 2026
Another developer argues that this class of tool can be excessive for very small teams when a simple Lua simulation or rules engine already answers the question. This is a fit counterpoint, not a product failure.
- U05Naavik: ILV and SPS economy simulationsChecked Sep 1, 2026
A multi-diagram economy analysis changes demand, pack, and sell assumptions while explicitly warning against over-reading simplified models. The author received free access and vendor technical help.
- U06UPC: creation of a mobile game economyChecked Sep 1, 2026
An undergraduate project reports that decomposing a mobile-game economy improved the workflow, exposed tradeoffs, and prompted mechanic changes. It remains one author's project report.
- U07Theseus: progression-model adoption counterexampleChecked Sep 1, 2026
A commissioned mobile-card-game project found formal integration unjustified while mechanics were changing and resources were limited, relying mainly on playtests, diaries, and analytics.
- U08Reddit: management-game customer-flow modelChecked Sep 1, 2026
An anonymous designer says Machinations helped think through customer flow, scaling, and formulas, but was not precise enough to serve as the sole final-optimization method.
- U09Leap: model-to-live-game drift analysisChecked Sep 1, 2026
A competitor article credits Machinations for systems thinking while criticizing duplicate wiring and drift between diagrams, implementation, A/B tests, and analytics. Its commercial bias sharply limits weight.
DISCLOSURE
How this research was supported
Owner Review passed on September 1, 2026, accepting this bounded public-evidence review. MakeGameWithAI has no affiliate link, sponsorship, vendor-provided account, free credits, interview, or technical support for this work, and did not run Machinations. The Naavik author received free access and vendor help; the G2 review was invited and incentivized; Leap sells a competing product. Those relationships are disclosed and limit their evidentiary weight.
This public research review will be revisited when pricing, terms, product versions, or material new evidence changes.