ATLAS
Extracts atomic claims, detects contradictions, and distinguishes independent reporting from repetition.
Active research
Accipiter Group · Experimental Research
ALTUM
Defy all reason.
A controlled laboratory for analytical methods that are too new, uncertain, or unconventional for operational adoption.
09Active research lines
03Promotion gates
02v1-complete systems
∞Permanent experimental record
Research mandate
Novelty without diluted standards.
Altum explores analytical gaps that are too speculative, immature, or methodologically uncertain for operational adoption. Experiments can graduate, become standalone tools, or end in a documented postmortem.
ModelSmall, self-contained experiments with a written specification
and success criteria.
RecordRunning logs preserve results, limitations, and failed
approaches.
RelationshipAltum may read suite outputs; operational systems do not depend
on lab code.
OutcomePromote, spin out, or retire—with no experiment disappearing
without a record.
Current portfolio
Research into claims, causality, warning, and assessment fragility.
CASCADE
Tests temporal relationships in claim activity using explicit statistical methods and corrected significance thresholds.
Active researchAURORA
Combines multiple signal classes into ranked threat horizons while preserving the components behind each score.
Active researchRUBICON
A structured record of stated red lines, threshold crossings, responses, and observed enforcement history. v1: 41 lines, 33 crossings, 14 actors, 6 domains, with Wilson-interval credibility scoring.
v1 completeORRERY
Tests whether interactive assumption models help analysts identify fragile conclusions faster than static confidence tables.
PrototypeSILENCE
Investigates negative intelligence: when the absence of expected activity becomes analytically meaningful. 836k events processed against a live collector; 33 grade-A alerts to date.
Active — session 1 completeOVERLAY
A synthetic simulation lab: sealed baseline populations and disposable counterfactual branches for testing defensive interventions against belief propagation. Synthetic entities and declared equations only—it does not model or predict real people.
Active researchDELTAV
Maneuver and co-orbital detection from public element sets. Synthetic gates pass and live false positives are retracted; final real-maneuver persistence confirmation is pending.
Release candidateSHARD
Fragmentation forensics across 70,472 catalogue objects, dating 93 events. Validated against NASA data: median dating error 0.11 years, collision precision 1.00.
v1 completeOVERLAY in practice
A sealed baseline, a disposable branch, one paired divergence.
Every run compares an intervention branch against a read-only sealed baseline on the same synthetic population, and reports the divergence rather than a single outcome.
A methodological result from this instrument: its evidence × trust
product was found to be structurally degenerate—the observable
trajectories cannot attribute an effect to either factor on their
own. That finding is reported as fragile rather than promoted as a
discovery, per Altum's own gates.
Promotion standard
An experiment leaves Altum only after three gates.
Reproducibility
A fresh environment can run the work end to end from its written specification without undocumented operator knowledge.
Rigor
Outputs remain traceable to a defined method, source, equation, or testable analytical rule.
Orthogonality
The experiment performs a capability the suite does not already have—or demonstrates a measurable improvement to an existing one.
Altum is not a catalog of finished products. It is a controlled record of research in progress.
Public descriptions intentionally distinguish active systems,
prototypes, and planned research. Experimental outputs are not
presented as operational intelligence capabilities.