Capabilities / C · Data and program

C.1 Data processing Active

Given raw data and processing rules, the agent cleans, maps, aligns and recomputes it into an output file. Graded by field-by-field agreement with a reference recomputation.

Evidence: frozen. A frozen benchmark version tests this capability. No site-verified result exists yet.

Domain
C · Data and program
Evidence form
program
Veto constraints
—
Depends on
A.1 Medical semantic understanding
Required by
C.2C.3C.4D.1D.11D.13D.14D.3D.5E.5
Internal task families
omics_qc, pathogen_qc
External anchors
BixBench
Version
0.1.0 · 0.1.0

Definition in detail

Most medical data work is cleaning, mapping and recomputing before any analysis. Small processing errors such as dropped rows or double conversions survive into every downstream result.

Inputs a task gives

  • Raw tables or files with inconsistent formats
  • Processing rules: cleaning steps, mapping tables, unit conversions and formulas for derived fields
  • A data dictionary with field definitions and tolerances
  • An output schema

Artifact required

An output file in the specified schema and a processing log listing each rule applied and the rows it affected.

How it is graded

A program compares the output field by field with a reference recomputation, using the numeric tolerances in the dictionary. Row counts and keys must match. Values that the mapping rules cannot resolve must be handled exactly as the rules direct.

In scope

  • Cleaning, mapping, joining and aligning records
  • Unit conversion and derived-field computation
  • Handling unmappable values as the rules direct

Out of scope

  • Generating synthetic records - graded under C.2
  • Checking a gold standard against its raw data - graded under C.4
  • Domain-specific recomputation of scales, scores and QC metrics - graded under D.3, D.5 and D.13

Typical failures

Expected failure patterns; not measured rates.

Standards: std-omop-cdm-5.4

Benchmarks

Grading references

No grading reference is mapped to this capability.

Levels by model and harness

L1 not computable (no calibration subset)The frozen benchmark versions that test this capability have no designated calibration item yet, so L1 and above cannot be derived.