Scope question: auditable multi-agent research for Mega-Sena indicators and covering designs

[FABIO8495]

Hello MiroFish maintainers and community,

We are developing “Mega-Sena · Histórico e quadrantes”, a Brazilian lottery analysis dashboard. We would appreciate guidance on whether any part of this project fits MiroFish, rather than assuming social-agent simulation can predict physical lottery draws.

Project and primary goal

Mega-Sena draws six distinct numbers from 1–60. Our dashboard groups them into 12 fixed groups of five numbers and currently selects nine groups (45 numbers) using an experimental standardized-delay indicator (“Z-core”). Our main research objective is to determine whether a frozen indicator can improve future coverage over a matched random baseline. We have no established predictive advantage. Under independent uniform draws, historical frequency, delay and narrative agent consensus do not provide such an advantage.

We also work on a separate, verifiable problem: arranging six-number tickets to cover four-number subsets of a chosen pool. This improves conditional combinatorial coverage, not the probability of drawing the chosen numbers.

Concrete progress

Using Covering-PDO (https://github.com/germuth/Covering-PDO), with independent quartet enumeration:

• 9-number pool: 12 tickets cover all 126 four-number subsets.
• 15-number pool: 129 tickets cover all 1,365 subsets, versus our earlier 145-ticket reference. This is a feasible covering, not a proven minimum.
• 30-number pool: a five-minute search with 2,604 tickets covered 27,345 of 27,405 subsets, leaving 60 uncovered. It therefore does NOT provide an integral conditional guarantee.

Our statistical work has explored frequencies, delays, momentum and conditional repetition, but we cannot claim a reproducible out-of-sample predictive edge. We want executable evidence rather than plausible explanations or invented execution results.

Questions about fit and implementation

We understand from your FAQ that MiroFish is a social simulation and reporting system, not a calibrated numerical forecasting engine or a service for generating betting predictions.

1. Is there a supported way to connect agents/reporting to external deterministic Python or C validators, and require every numerical claim to reference actual execution artifacts?
2. Could MiroFish help critically review pre-registered hypotheses and experiment designs without treating simulated agent agreement as statistical evidence? Or would that be outside its intended architecture?
3. If this research workflow is outside scope, could MiroFish instead support an explicitly exploratory study of how users misunderstand conditional guarantees, coverage percentages and experimental indicators in a lottery dashboard? Such simulation would still require validation with real users.
4. Which extension points or examples would you recommend, and what limitations should we preserve in our implementation and user-facing claims?

For context, we have also submitted a workflow-architecture question to ChatDev: OpenBMB/ChatDev#714. We are not asking you to run predictions or supply winning numbers; we are asking for a candid fit assessment and, if appropriate, a reproducible integration path.

Prepared with AI assistance and submitted on behalf of the project owner.