You already have the maker. We are the other half of the loop.
Claude or GPT can easily write paytables, weight tables, and features. What it cannot do is calculate exact RTP, test designs on players, or say no to itself. That is the half we sell, delivered as tools your LLM can call.
Your problem
RTP is made up
The model confidently returns 96.1%. No one has enumerated the reels. The first real number will not appear until the game is live.
No players to test against
No round data, no cohort, and no way to know whether the design can keep anyone beyond the first session.
No mechanism that says no
The LLM grades its own work. There is no checker with cost bounds, perception budgets, or consistency checks.
What you get
An exact engine as a tool
PAR sheet and weight table engines that calculate exact RTP, hit frequency, and volatility for anything an LLM drafts, with Monte Carlo confidence intervals.
Synthetic players from day one
Behavior populations with ten switchable mechanisms, including loss aversion, chasing, stopping rules, and house money, pretrained on aggregate corpora and responsive to math outcomes.
A checker that says no with reasons
Gates from the Loop Engineering paper: cost upper bounds at explicit confidence levels, perception budgets, and consistency checks. Pass or fail, with reasons.
Private access available today
The engines are live in the browser. Hosted REST and MCP endpoints are in private access; apply and we will onboard you.
Evidence
Run the loop manually first
The Weight Table simulator runs the same maker, checker, and memory against synthetic player populations in your browser.
Open simulatorThe population is the parameter
802 simulated runs show behavior populations and how each mechanism affects retention and the ledger.
Read the studyLoop Engineering for Self-Improving Slot Agents
Six primitives, five phases, and the maker–checker separation your LLM is missing.
Read the paperFirst step
- 1
Drop a draft into the simulator
Paste the paytable or weight table your LLM generated and read back the exact numbers.
- 2
Request tool access
Tell us which model you are building with and what you are delivering. We will enable private endpoints for you.
- 3
Close the loop
Your LLM proposes, reSlot calculates, simulates, and scores, then the experience feeds back into the next proposal.
Common studio questions
Start with your next decision
Bring a table.Tell us your goal.
Want to reduce pre-submission iteration, evaluate a compensation budget, or validate an AI-generated design? Tell us which metric you want to optimize, and bring a PAR sheet, weight table, or draft. Together we’ll define the evaluation scope, deliverables, and next step.