Data Analysis School
An open-source AI skills project for data analysis, integrated with Codex. It constrains AI data analysis behavior through mechanisms such as an evidence grading system, intra-table reconciliation, and a GrillMe clarification gate, preventing models from fabricating data-based conclusions.
Timeline (last 90 days)
Data Analysis School is a data analysis skills project, authored by the data analysis UP host "Xiao Ma Xue Shu Fen".
The Data Analysis School project has more than 30 stars on GitHub.
The project includes 74 regression tests, CI covering three Python versions, and byte-by-byte rendering comparison for case reports.
The project's capability matrix covers 16 data analysis methods, including anomaly location, funnel analysis, metric systems, A/B experiments, causal inference, user segmentation, contribution decomposition, channel attribution, cohort retention, survival analysis, path analysis, price elasticity, breakpoint detection, and multiple comparison correction.
The project has 11 built-in industry calibration templates covering retail e-commerce to finance and gaming, each specifying applicable signals, KPI definitions, common data pitfalls, and fallback paths.
The project adopts an in-table reconciliation mechanism, using a reconcile script to cross-validate each line item with the table's built-in checksum, blocking anything that does not match, with the exit code serving as a pipeline gate.
The project classifies conclusions by evidence levels from L0 to L4, stipulating that causal language at L2 or above fails outright if the corresponding evidence is missing.
The GrillMe gate must first ask the user a clarifying question when a query is ambiguous; clarifying rounds do not generate a report, and only when the user explicitly says to proceed directly is the system allowed to continue with default assumptions.
Explanatory questions involving "why" must run four independent roles: Locator, Mechanism, Falsifier, and Reviewer — the first three are dispatched in parallel, Reviewer consolidates at the end, and publication is not allowed if any one is missing.
Reports have the model write only the specification JSON, with HTML and charts generated by a separate renderer to achieve separation of content and style, and numerical geometric animations are prohibited to avoid being frozen into incorrect readings in screenshots, print, or background tabs.
3 more timeline events
All Facts (13)
Data Analysis School is a data analysis skills project, authored by the data analysis UP host "Xiao Ma Xue Shu Fen".
50%UnverifiedThe Data Analysis School project has more than 30 stars on GitHub.
50%UnverifiedThe project includes 74 regression tests, CI covering three Python versions, and byte-by-byte rendering comparison for case reports.
50%UnverifiedThe project's capability matrix covers 16 data analysis methods, including anomaly location, funnel analysis, metric systems, A/B experiments, causal inference, user segmentation, contribution decomposition, channel attribution, cohort retention, survival analysis, path analysis, price elasticity, breakpoint detection, and multiple comparison correction.
50%UnverifiedThe project has 11 built-in industry calibration templates covering retail e-commerce to finance and gaming, each specifying applicable signals, KPI definitions, common data pitfalls, and fallback paths.
50%UnverifiedThe project adopts an in-table reconciliation mechanism, using a reconcile script to cross-validate each line item with the table's built-in checksum, blocking anything that does not match, with the exit code serving as a pipeline gate.
50%UnverifiedThe project classifies conclusions by evidence levels from L0 to L4, stipulating that causal language at L2 or above fails outright if the corresponding evidence is missing.
50%UnverifiedThe GrillMe gate must first ask the user a clarifying question when a query is ambiguous; clarifying rounds do not generate a report, and only when the user explicitly says to proceed directly is the system allowed to continue with default assumptions.
50%UnverifiedExplanatory questions involving "why" must run four independent roles: Locator, Mechanism, Falsifier, and Reviewer — the first three are dispatched in parallel, Reviewer consolidates at the end, and publication is not allowed if any one is missing.
50%UnverifiedReports have the model write only the specification JSON, with HTML and charts generated by a separate renderer to achieve separation of content and style, and numerical geometric animations are prohibited to avoid being frozen into incorrect readings in screenshots, print, or background tabs.
50%UnverifiedThe project's main skill file at rest is just over 4,000 tokens, accompanied by a loading table specifying which file to read at which step, with method packages and industry templates prohibited from being loaded in their entirety.
50%UnverifiedThe project's cases use the UCI UK retailer public dataset, with everything from data construction to run artifacts to conclusions available in the repository.
50%UnverifiedThe project uses the non-commercial license Polyform Noncommercial, and companies must obtain authorization before use.
50%