missingbias

Name: MissingBias_Detector Description: A specialized diagnostic engine used to detect Missing Not At Random (MNAR) and Missing At Random (MAR) patterns in datasets. This tool determines if the "missingness" of data in a primary variable is statistically dependent on the values of a secondary covariate. Use this to determine whether missing data can be safely deleted or if it requires advanced imputation to avoid systematic bias in downstream models. Why This Tool is Mandatory for Data Cleaning Prevents Selection Bias: Identifying bias ensures that the agent does not inadvertently delete a specific sub-population (e.g., an unreliable sensor that only fails at high temperatures). Automated Strategy Selection: Provides the statistical evidence needed to choose between Deletion (if no bias is found) and Imputation/Source Investigation (if bias is detected). Math Error Prevention: Offloads complex dependency testing (like Little’s MCAR test or logistic modeling of missingness) to a dedicated engine, eliminating LLM calculation errors. Operational Logic The tool analyzes a dictionary containing two aligned arrays: Target Array (Index 0): The variable containing missing values (null, NaN, or empty strings). Predictor Array (Index 1): The potential biasing variable used to see if its values influence the probability of the Target Array being missing. Recommended Workflows Exploratory Data Analysis (EDA): Run this on all permutations of columns to identify hidden dependencies in a new dataset. Hardware/Sensor Audits: Identify "unreliable sources" (e.g., which satellite sensor or survey researcher is producing the most incomplete data). Pre-Training Validation: Ensure that "dropping rows" won't result in a biased training set that compromises model generalization. Interpretation of Results Bias Detected: You must not simply delete the missing rows. You must investigate the source of the bias or use statistical imputation. No Bias Detected: Missingness is likely stochastic; deleting rows is a statistically lower risk for analysis. Example Input: { "array_with missingness":["NA",166.445,470.604,25.0739,49.1652,324.7797,190.9287,"NA",451.39,405.4469,"NA",347.1129,253.0294,141.4462,"NA",241.4338,160.2388,123.1855,51.5936,151.8691,309.7825], "array_causing_bias":[418.3812,"NA",14.552,329.5427,"NA",119.1472,"NA",462.8084,320.5384,148.8701,412.0277,125.1991,"NA",255.8993,441.0706,"NA",297.2804,"NA","NA",296.7565,111.2001] } Example Output: {"missing_is_biased":[1]}. It is categorised as a Destructive tool in the Scientific Microservices MCP Server, which means it can permanently delete or destroy data. Block by default and require explicit approval.

Register the Scientific Microservices MCP server in PolicyLayer and add a rule for missingbias: allow, deny, rate-limit, or require approval. Point your MCP client at the PolicyLayer proxy URL and the rule is enforced on every call, before it reaches Scientific Microservices. Nothing to install.

missingbias is a Destructive tool with critical risk. Critical-risk tools should be blocked by default and only enabled with explicit human approval.

Yes. Add a rate_limit block to the missingbias rule in your PolicyLayer policy. For example, setting max: 10 and window: 60 limits the tool to 10 calls per minute. Rate limits are tracked per agent session and reset automatically.

Set action: deny in the PolicyLayer policy for missingbias. The AI agent will receive a policy violation error and cannot call the tool. You can also include a reason field to explain why the tool is blocked.

missingbias is provided by the Scientific Microservices MCP server (https://mcp.scientificmicroservices.com/mcp). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.