Capabilities / C · Data and program

C.5 Pipeline orchestration Active

Given a multi-step pipeline specification, the agent organises and runs it, including resuming from a checkpoint. Graded by the pipeline end state and by the presence of the required intermediate artifacts.

Evidence: none. No benchmark card lists this capability yet. No level is issued: methodology 0.1.0 sets no level thresholds.

Domain
C · Data and program
Evidence form
program
Veto constraints
—
Depends on
B.3 Tool invocation
Required by
D.10D.5E.5
Internal task families
traj_lit_ops
External anchors
MedAgentGym
Version
0.1.0 · 0.1.0

Definition in detail

Medical data work runs as multi-step pipelines that are interrupted and resumed. Restarting from scratch wastes time; resuming incorrectly mixes stale and fresh outputs.

Inputs a task gives

  • A pipeline specification with steps, dependencies and required intermediate artifacts
  • Partial state left by an interrupted run
  • The tools each step uses
  • Resume rules

Artifact required

The final output, the required intermediate artifacts at their specified paths, and a run log (step, status, input hashes, outputs).

How it is graded

A program checks the pipeline end state and the presence and validity of each required intermediate artifact. On resume items, steps that completed before the interruption must not be redone where the specification forbids it, and their outputs must be unchanged.

In scope

  • Running steps in dependency order
  • Resuming from a checkpoint without redoing completed steps
  • Producing the required intermediate artifacts

Out of scope

  • Correctness of individual tool calls - graded under B.3
  • Correctness of the data transformation inside each step - graded under C.1

Typical failures

Expected failure patterns; not measured rates.

Standards: std-atif

Benchmarks

No benchmark card lists this capability.

Grading references

No grading reference is mapped to this capability.

Levels by model and harness

No verified results yetLevels would be derived from site-verified result records; methodology 0.1.0 sets no level thresholds, so the level shows not issued.