Skip to content

Analyses and research objects

Formats that record a scientific analysis or package research outputs. In each case ASB Schema supplies the paper-grounded content: tasks, evidence, decisions and files.

ASTRA

ASTRA (Agentic Schema for Transparent Research Analysis), published by Lightcone Research, records one scientific analysis while it is designed and run: inputs, outputs, decisions with named options, universes and insights backed by evidence1. It is written in LinkML and ships a JSON Schema and a JSON-LD context2.

ASB Schema produces ASTRA documents and is upstream of them for published research. The exporter of the reference pipeline writes two documents per run:

  • astra/astra.yaml, the science view, from the paper, its cards and its exploration trace;
  • astra/run_trace.yaml, the build view: one decision per resolved build flag, one per agent model, one sub-analysis per pipeline unit.
ASB source ASTRA element
Paper processed in the run Root analysis
Task card Sub-analysis
Trace decision node Decision with options; untested alternatives stay ordinary options
Trace pivot node Decision whose new approach is the default and former approach an excluded option, with trigger as the reason
Trace dead_end under a decision Excluded option, with failure_mode as the exclusion reason
Trace dead_end with no enclosing decision Finding of the root analysis
Claim-ledger claims Findings with quoted evidence
Build flags and agent models Decisions of run_trace.yaml

Trace question and experiment nodes are reported as unmapped.

ASB Schema to ASTRA

existing

The exporter targets ASTRA 0.0.12; the current release is 0.0.143. Moving to it means writing a root id in astra.yaml and setting Output.format on outputs that do not alias another output.

Further bridge What it needs Status
EvidenceSnippet to ASTRA Evidence snippet_id to id, text to quote.exact, the card's crossref_doi to doi; reason to the notes of the enclosing insight feasible
ASTRA analysis to task cards A reader and a rule for task boundaries, since one analysis can yield several cards feasible
astra: mappings in the LinkML files close_mappings, for example from ASBTaskCard to astra:Analysis feasible
Evaluation records in ASTRA An evaluation extension, proposed in issue #744 needs new code

RO-Crate

RO-Crate packages research data with their metadata in one JSON-LD file, ro-crate-metadata.json, based on schema.org5. The Workflow Run RO-Crate profiles describe executions: Process Run Crate for tool executions recorded as CreateAction entities, Workflow Run Crate for one workflow execution and Provenance Run Crate for per-step detail6.

Every capsule carries an RO-Crate 1.1 metadata file whose CreateAction links the tool (instrument), the inputs (object) and the outputs (result); see Capsule.

Bridge What it needs Status
RO-Crate 1.1 Written for every capsule; a root datePublished and a license entity complete it existing
Process Run Crate The profile IRI in the root conformsTo (https://w3id.org/ro/wfrun/process/0.5 for RO-Crate 1.1), and agent and endTime on the action7 feasible
RO-Crate 1.2 and 1.3 New context and descriptor IRIs and Profile entities; ro_crate_version is a free string8 feasible
Workflow RO-Crate A class for workflow.yaml and one crate per workflow with a ComputationalWorkflow main entity9 needs new code

rocrate-validator checks a crate against RO-Crate and its profiles10.

ARA

ARA (Agent-Native Research Artifact) defines one artifact per paper or project, a directory with PAPER.md, logic/, src/, trace/ and evidence/ layers11. Its trace/ layer holds an exploration tree with the same five node types and the same two support levels as the exploration trace.

ASB Schema to ARA

existing

asb ara-export writes an ARA package for each finished run under ara_export/, builds each layer from the run's records and copies the exploration trace. ARA names the top-level list tree where the trace uses nodes, and asks for a description on question nodes12; the converter renames the key and fills the description.

Workflow languages

asb export-workflow converts the run's workflow.yaml into CWL, Nextflow and Snakemake files under exports/1314. Command bodies come from each card's runtime executable command; a card without one yields a structural stub. The optional card slots executable (ExecutableSpec), execution_environment (ExecutionEnvironment) and data_in, data_out (TypedArtifact) carry that command, its software stack and its typed files on the card itself.

existing

See also


  1. ASTRA, "Specification". https://astra-spec.org/latest/specification/ ↩

  2. Lightcone Research, "astra.schema.json". https://astra-spec.org/latest/schema/astra.schema.json ↩

  3. GitHub, "Tags · LightconeResearch/astra-spec". https://github.com/LightconeResearch/astra-spec/tags ↩

  4. GitHub, "Proposal: evaluation extension for ASTRA records · Issue #74". https://github.com/LightconeResearch/astra-spec/issues/74 ↩

  5. Research Object, "Root Data Entity" (RO-Crate 1.1). https://www.researchobject.org/ro-crate/specification/1.1/root-data-entity.html ↩

  6. Research Object, "Workflow Run RO-Crate". https://www.researchobject.org/workflow-run-crate/ ↩

  7. Workflow Run RO-Crate, "Process Run Crate". https://www.researchobject.org/workflow-run-crate/profiles/process_run_crate/ ↩

  8. Research Object Crate (RO-Crate), "RO-Crate 1.3". https://www.researchobject.org/ro-crate/specification/1.3/ ↩

  9. WorkflowHub, "Workflow RO-Crate". https://about.workflowhub.eu/Workflow-RO-Crate/ ↩

  10. GitHub, "crs4/rocrate-validator". https://github.com/crs4/rocrate-validator ↩

  11. ARA-Labs, "README.md", Agent-Native-Research-Artifact repository. https://raw.githubusercontent.com/ARA-Labs/Agent-Native-Research-Artifact/main/README.md ↩

  12. GitHub, "skills/compiler/references/exploration-tree-spec.md". https://raw.githubusercontent.com/ARA-Labs/Agent-Native-Research-Artifact/main/skills/compiler/references/exploration-tree-spec.md ↩

  13. Common Workflow Language, "Home". https://www.commonwl.org/ ↩

  14. Nextflow, "A DSL for parallel and scalable computational pipelines". https://www.nextflow.io/ ↩