Latentmachine

Verify Data Transformation

Check whether a batch of AI-transformed data rows all follow one deterministic rule. Takes the original records and transformed output, infers the majority rule, and returns a capped diagnostic summary. Sparse optional fields are scoped to their source domain and may be unverifiable without flagging out-of-domain rows. Coherent alternative rules are reported as additive clusters with privacy-safe path signatures, support, and share; equal splits are not reported as row-level defects. Candidate i

Infer Transformation Rule

Infer a deterministic data transformation rule from input/output example pairs. Returns the symbolic rule, confidence assessment, and any ambiguities. The rule can then be applied to new data with apply_transformation_rule.

Apply Transformation Rule

Apply a previously inferred transformation rule to new input data. Requires an executable rule object from infer_transformation_rule. Executes deterministically: same input plus same rule gives the same output.

Detect Data Format

Detect the structured data format of a text string. Returns json, csv, yaml, toml, xml, env, sql, or unknown.

Fingerprint Data

Compute a deterministic structural fingerprint of a dataset, or a structural diff between two datasets. Same data always produces the same fingerprint: object key order is ignored and array order is significant. Useful for asserting transformation stability, detecting config drift, or verifying agent output stability. Raw JSON integer literals beyond JavaScript's safe range are disclosed because parsed values may lose precision. Non-cryptographic. Deterministic engine, not an LLM.

Learn Transformation Contract

Learn a versioned Transformation Contract from input/output examples. Returns an unapproved or review-required contract and never claims human approval. Remote HTTP use sends the supplied examples to Latentmachine for stateless processing; use the local stdio server for local-only data.

Get Contract Challenges

Return unresolved review questions for a Transformation Contract. Does not answer or approve them. Remote HTTP use sends the contract to Latentmachine for stateless processing.

Test Transformation Contract

Mutation-test every learned operation target against its evidence, disclose target coverage and detection gaps, and downgrade the reported inference status when mutation evidence is incomplete. Remote HTTP use sends the contract and its embedded evidence to Latentmachine for stateless processing.

Run Transformation Contract

Run an already-approved Transformation Contract deterministically. Unapproved contracts are rejected and MCP cannot create local-human-review approval. Remote HTTP use sends the contract and input to Latentmachine for stateless processing; use local stdio for local-only data.

Check Transformation Contract

Check external output against an already-approved Transformation Contract. Unapproved contracts are rejected and MCP cannot create local-human-review approval. Remote HTTP use sends the contract, input, and output to Latentmachine for stateless processing; use local stdio for local-only data.

Compare Transformation Contracts

Compare two Transformation Contracts and classify behavioral, review, evidence, and metadata changes. Remote HTTP use sends both contracts to Latentmachine for stateless processing.