Schema table

Field names are LinkML (Linked Data Modeling Language) slot identifiers from skills/csag-extraction/assets/csag.yaml. Required fields are marked with *. Provenance and curation slots reach the extracted record classes through the Record and ExtractedItem mixins.

The schema has five modules. core is the minimum manuscript-local claim and evidence ledger. bio, reasoning, research_state, and benchmark are optional enrichment layers that reference core identifiers. Validation profiles maps the modules to --profile values and explains why the reasoning module is separate from core.

Module	Object groups
core	PaperExtraction, TextSpan, Assertion, Context, EvidenceItem, EvidenceLink, StudyCritique, KnowledgeGap, Artifact, Dataset, provenance and curation fields
bio	Entity, EntityMention, OntologyAnnotation, Condition, Qualifier, Study, Experiment, Variable, normalized assertion fields
reasoning	InferenceStep, AssertionRelation
research_state	ResearchStateRecord, NextAction, Execution
benchmark	QAItem, Answer, benchmark scoring metadata

Object class	Key fields	What it records
PaperExtraction	id*, title*, schema_version*, validator_version*, doi, pmid, journal, publication_date, authors, keywords, mesh_terms, full_text_url, license, assertions*, evidence_items*, evidence_links*	Bibliographic anchor and root container for one manuscript.
TextSpan	document_id*, section_type*, start_char*, end_char*, exact_text, page_number, artifact_ref	Character-level source grounding for extracted objects.
Entity with EntityMention, OntologyAnnotation	entity_category, ontology_annotations, xrefs, mentions, term_id, term_label	Normalized scientific term with cross-references and ontology mappings.
Context	context_facet, organism, cell_type, tissue, disease_state, strain, developmental_stage, sex, age, environment_description, additional_context_qualifiers	Biological or experimental scope. Every Assertion has at least one.
Condition	parameter_type, value_text, value_number, value_unit, entity_involved, logical_expression	Dose, time, genotype, treatment, or other claim parameter.
Study / Experiment / Variable	study_type, study_contexts, experiments, sample_size, assay_type, variables, variable_role, variable_entity, measurement_type	Study design, experiments, and measured or manipulated variables.
Assertion	assertion_text*, claim_role*, assertion_type, normalization_status*, subject, predicate, object, contexts* (at least one), conditions, qualifiers, criticality, falsification_criteria	Scoped claim or hypothesis, optionally normalized as a subject-predicate-object triple.
EvidenceItem	evidence_type*, evidence_text, contexts, conditions, associated_experiment, associated_artifacts, results, referenced_works	Structured evidence unit. Polarity is stored on EvidenceLink.
Result	result_text, outcome, comparator, effect_size, effect_size_type, p_value, ci_low, ci_high, statistic, n	Numerical result attached to evidence, such as an effect size, statistic, interval, or sample size.
EvidenceLink	evidence_item*, assertion*, polarity*, strength, rationale	Support, refute, mixed, or inconclusive edge from evidence to assertion.
InferenceStep	input_assertions, input_evidence_links, output_assertion*, inference_method*, assumptions, inference_rationale	Reasoning step that combines premises and evidence into a derived assertion.
AssertionRelation	from_assertion*, to_assertion*, relation_type*, relation_rationale, related_work	Typed claim-to-claim relation within or across manuscripts, including competing or alternative hypotheses.
StudyCritique	critique_type, risk_domain, severity, impacted_assertions, impacted_evidence_items, mitigation_suggestions	Threat to validity or risk-of-bias note tied to assertions or evidence.
KnowledgeGap	gap_type, severity, related_assertions, suggested_actions	Open question, missing test, replication need, contradiction, or generalizability gap.
ResearchStateRecord	target_assertions, state*, current_read, rationale, recommended_next_actions	Interpretation of a claim or branch after weighing the evidence.
NextAction	action_type*, description*, target_assertions, target_knowledge_gaps, priority, due_date	Next experiment, analysis, review, decision, branch, or merge step.
Execution	execution_type*, execution_status*, command, started_on, completed_on, output_artifacts, generated_evidence_items, tested_assertions	Run that generated evidence, tested claims, or produced durable artifacts.
Artifact / Dataset	artifact_type, artifact_label, caption, accession, repository, dataset_url, dataset_license	Figure, table, code, supplement, equation, or data deposit referenced by evidence.
QAItem / Answer	question_text*, expected_answer_type*, answers, answer_type*, answer_text, answer_number, answer_boolean, answer_entity, supporting_assertions, supporting_evidence_links, answer_confidence	Evidence-backed benchmark or downstream question-answer surface.
Provenance and curation mixins	id*, label, description, aliases, notes, origin, curation_status, confidence_score, text_spans, created_on, created_by, generated_by, derived_from, ExtractionActivity (activity_type, tool_name, tool_version, model_name, model_version, run_id, run_datetime, parameters)	Shared provenance and curation fields. Every mixed-in record has a stable ID; the other fields are optional.