Evidence items
38618
Deduplicated public sources
Theme Intelligence
The platform continuously maps a subject, separates claims from narratives, verifies them against multiple evidence spaces and publishes every verdict with its proof path.
Two clocks, one view
STS reads licensed scientific documents into immutable sentences, table cells and formulas. Models propose structure; deterministic validators and the Constitution alone commit proof.
STS 1.0.0 · Parser rules visible in every derivation
Some studies are metadata-only because no reusable full text is available. They remain visible as an evidence limitation.
Live knowledge layer
These figures are read from the current knowledge graph. They can change independently of a software deployment as new public evidence is processed.
Evidence items
38618
Deduplicated public sources
Current claims
2443
Canonical factual statements
Proof-path coverage
43%
1051 of 2443 claims linked
Mapped actors
90
Current theme-specific map
Intelligence flywheel
Collect public science, official material, editorial reporting and social signals.
Deduplicate, identify language and preserve origin and provenance.
Turn material into concepts, questions, actors, narratives and atomic claims.
Test claims against relevant evidence and explicit applicability boundaries.
Expose how sources, actors, narratives, questions and proof paths relate.
Coverage gaps become the next research frontier instead of hidden uncertainty.
Signal to proof
Models may extract and critique. Only the deterministic constitution commits the multi-axis verdict and freezes the proof bundle used by the platform and sister products.
v4
Constitution
v4
Proof schema
theme-knowledge-pack/v4
Export contract
Extraction and critique leave traces, confidence and candidate evidence; they do not create truth by themselves.
Explicit rules commit, hold or contradict the claim and enforce applicability and evidence thresholds.
The current claim, append-only claim events and versioned proof bundle remain inspectable and exportable.
Knowledge model
A question organizes investigation. It does not receive a truth verdict; the atomic claims beneath it do.
Support, strength, applicability, fidelity, confidence and lifecycle remain separate axes instead of one misleading score.
Every active link records whether evidence supports, contradicts or merely contextualizes a specific claim.
Credibility is always theme- and time-window-specific, includes sample size and never becomes a global person score.
Narratives reveal which claims travel together, who amplifies them and where attention diverges from evidence.
Briefs and knowledge packs are projections of versioned objects, not detached AI summaries.
Executable process map
This sequence is rendered from the same catalogue used by the autonomous pipeline. A code change cannot silently leave this process map behind.
Latest run
Running · 03.11.2026 · 13:04 UTC
Process hierarchy
Root process
A theme, steering priorities and budgets initiate one causal knowledge-building run.
Phase 1
Defines the thematic ontology, concepts and open research frontier.
1 production step
Governed handoff
Ontology and research frontier
Phase 2
Discovers and collects relevant public material across evidence spaces.
2 production steps
Governed handoff
Deduplicated public signals
Phase 3
Resolves studies, reads lawful full text and creates stable sentences, tables, formulas and assertions.
8 production steps
Governed handoff
Source-addressed scientific statements
Phase 4
Normalizes material and projects it into typed, provenance-preserving evidence.
1 production step
Governed handoff
Typed evidence objects
Phase 5
Extracts atomic statements, assigns their type and merges semantic duplicates.
4 production steps
Governed handoff
Canonical atomic claims
Phase 6
Fills evidence gaps and commits inspectable multi-axis verdicts.
2 production steps
Governed handoff
Verdicts and proof bundles
Phase 7
Relates claims, evidence, actors and narratives and produces traceable syntheses.
2 production steps
Governed handoff
Knowledge graph and syntheses
Phase 8
Projects governed knowledge into metrics, briefs and stable consumer contracts.
2 production steps
Governed handoff
Metrics, briefs and knowledge packs
Linear execution order
The hierarchy explains responsibility and outputs; the sequence below shows the exact causal order persisted for every run.
Research frontier
Ontology
Signal acquisition
Discover sources
Signal acquisition
Collect signals
Scientific reading
Targeted research
Scientific reading
Resolve scientific works
Scientific reading
Acquire lawful documents
Scientific reading
Parse documents
Scientific reading
Extract scientific assertions
Scientific reading
Normalize quantities
Scientific reading
Interpret formulas
Scientific reading
Appraise studies
Evidence preparation
Project evidence
Claim formation
Extract claims
Claim formation
Type claims
Claim formation
Canonicalize claims
Claim formation
Link claims to source nodes
Truthing and proofing
Validate source anchors
Truthing and proofing
Adjudicate claims
Knowledge graph
Build graph and narratives
Knowledge graph
Synthesize knowledge
Intelligence outputs
Calculate actor metrics
Intelligence outputs
Publish brief
Bar length is relative to the largest evidence space, not a fictitious completeness score.
A live distribution of current atomic claims. Insufficient is an explicit research state, not a system failure.
Intelligence platform
Sister product, for example Longevity
Inspect the system