Project Stars is a browser-based research atlas for representing thoughts, concepts, values, institutions, and social pressures as structured objects within an interpretable conceptual system.
The project combines:
- an evidence-aware graph of reviewed and reviewable relations,
- a semantic and geometric model of conceptual distance,
- a dynamical layout that distinguishes stable structure from hypothesis,
- a continuous activation field for local propagation and disturbance,
- an operative pathway layer that changes under experimental conditions,
- and an experimental interface for studying how receiver-state, environment, and historical pressure alter conceptual outcomes.
- Abstract
- Research Framing
- Formal Model
- Thought Objects
- Layered World Model
- Balanced and Real-World Graphs
- Relation and Pathway Layers
- Candidate-Link Inference
- Modeling Principles
- Layout Dynamics
- Environment, Mediation, and Outcome Formation
- Activation Field and Disturbance
- Receiver Quality and Heterogeneous Interpretation
- Scenarios and Experimental Conditions
- Inspector and Research Interface
- Navigation and View Modes
- Export Semantics
- Research Foundations
- Scientific Contribution
- Limitations
- Future Directions
- Implementation Mapping
- Copyright
Project Stars formalizes thoughts and social concepts as attributed objects embedded in a layered conceptual environment.
The atlas distinguishes between:
- a confirmed relation layer of reviewed structure,
- a candidate relation layer of interpretable but unconfirmed hypotheses,
- an effective pathway layer that expresses which routes are currently operative under a given condition,
- a balanced moral geometry for fair structural testing,
- and a real-world environment layer that introduces institutions, scarcity, enforcement, leadership, memory, and conflicting pressures.
The resulting system supports a research program in which:
- concepts are modeled as evidence-bearing entities,
- relations preserve provenance and review state,
- candidate links are surfaced through explicit signals,
- social and moral outcomes are mediated rather than directly assumed,
- operative pathways vary with receiver-state and environmental pressure,
- and conceptual dynamics can be examined under both symmetrical test conditions and realistic asymmetric conditions.
Project Stars studies a central question:
How can thoughts and social concepts be represented as mathematically structured objects inside a research atlas that preserves provenance, uncertainty, mediation, and interpretable dynamics?
The atlas treats a thought not as a simple label, but as a structured object with:
- semantic content,
- ontology placement,
- evidential and provenance metadata,
- graph relations,
- local dynamical behavior,
- and context-sensitive interaction with social and institutional environment.
A central research distinction in the current project is the separation between:
- fairness of structure, tested through balanced graph conditions,
- stability of evidence, represented through confirmed and candidate relations,
- and variability of operative pathways, tested through receiver-state, scenario, and environmental pressure.
Project Stars is best understood as a formal computational hypothesis.
Semantic precision and social outcome formation are modeled as emergent properties of a layered conceptual system shaped jointly by structure, activation, receiver-state tuning, environmental mediation, historical pressure, and conditionally active pathways.
The full system can be written schematically as:
M = (Theta, A, d, Rc, Rp, Reff, S, p(t), phi, Q, E, H, Xi)
Where:
Theta= set of thoughts or concepts,A= attribute structure on each thought,d= multi-part similarity or distance structure,Rc= confirmed relations,Rp= candidate relations,Reff= effective pathways under current conditions,S= deterministic candidate-link score,p(t)= time-dependent visual embedding,phi= continuous activation field over the display plane,Q= receiver-state tuning,E= environmental and institutional mediators,H= historical and memory terms,Xi= observability and experimental metrics.
This is intentionally hybrid: Project Stars is not only a graph, not only a semantic space, and not only a simulation. It combines:
- an attributed object model,
- a layered graph,
- an epistemic review model,
- a dynamic layout,
- a field-like activation layer,
- an environmental mediation layer,
- an operative pathway layer,
- and an experimental receiver-state interface.
Let Theta be the set of all thoughts or concepts.
Each thought theta is modeled as:
theta = (L, D, e, C, M)
Where:
L(theta)= label,D(theta)= description,e(theta)= semantic embedding or feature representation,C(theta)= category or ontology class,M(theta)= metadata bundle.
The metadata bundle may include:
- references,
- provenance,
- evidence class,
- consensus,
- review state,
- source type,
- layer membership,
- and layout state.
The atlas separates multiple levels of organization so that internal states, social relations, institutions, and collective effects are not conflated.
Examples include:
- Intention
- Fear
- Empathy / Compassion
- Shame
- Impulse
- Restraint
- Self-control
- Trauma
Examples include:
- Trust
- Cooperation
- Conflict
- Reputation
- Influence
- Reciprocity
- Distrust
- Repair
Examples include:
- Law
- Enforcement
- Media
- Education
- Governance
- Incentives
- Punishment
- Legitimacy
Examples include:
- Collective Intention
- Collective Identity
- Norms
- Polarization
- Fragmentation
- Nation / State
- Historical Memory
This layered design allows edges to run both within and across levels, producing a more realistic account of how internal states become social outcomes.
A key methodological feature of Project Stars is the distinction between two modes of interpretation.
The balanced graph is used to test whether the model itself is structurally biased.
It introduces mirrored moral branches with comparable depth and connectivity, including paired or opposing concepts such as:
- Kindness ↔ Cruelty
- Trust ↔ Distrust
- Cooperation ↔ Exploitation
- Empathy ↔ Dehumanization
- Repair ↔ Retaliation
- Collective Identity ↔ Fragmentation
In this mode, the goal is not realism but fairness of structural testing.
The real-world graph introduces asymmetry deliberately. It includes:
- institutions,
- scarcity,
- law,
- enforcement,
- media amplification,
- leadership,
- incentives,
- punishment,
- historical memory,
- and uneven social pressure.
In this mode, the goal is to study how morally or conceptually similar inputs may lead to different outcomes under different environmental conditions.
Project Stars maintains distinct structural and operative layers.
The confirmed relation layer is:
Rc subset of Theta x Theta x RelationTypes
A confirmed relation represents reviewed or accepted structure.
The candidate relation layer is:
Rp subset of Theta x Theta x RelationTypes x [0,1] x Evidence
Candidate relations are:
- deterministic,
- explainable,
- reviewable,
- visually subordinate,
- and distinct from accepted structure.
Strong candidate relations may exert weak geometric influence during layout while remaining epistemically separate from confirmed structure.
The operative pathway layer is condition-dependent:
Reff = f(Rc, Rp, Q, E, H, sigma)
Where sigma denotes the current scenario or experimental condition.
Effective pathways represent the routes that are currently most active or salient under a given receiver-state, environmental pressure profile, and scenario. They do not replace the evidence graph. Instead, they modulate visible and dynamical emphasis across the existing structure.
This allows the atlas to keep:
- stable evidence relations as reviewed structure,
- reviewable hypotheses as candidate relations,
- and dynamic operative routes as condition-sensitive pathways.
Candidate relations are computed by a deterministic scoring function.
For each thought theta, define signal sets such as:
Tok(theta)= token set from label and description,Ref(theta)= reference or citation set,Top(theta)= topic set,Year(theta)= representative temporal signal.
A candidate-link score can be expressed as:
S(theta_1, theta_2) =
0.28 * s_sem+ 0.24 * s_cit+ 0.18 * s_top+ 0.16 * s_nei+ 0.09 * s_ont+ 0.05 * s_rec
Where the components encode semantic similarity, citation overlap, topic overlap, confirmed-neighbor overlap, ontology match, and recency.
A candidate relation is admitted only when:
S(theta_1, theta_2) >= tau
with threshold tau defined by the current atlas configuration.
A candidate relation exists only if:
- the score exceeds threshold,
- and the basis of inference is non-empty.
This ensures that surfaced hypotheses remain interpretable rather than opaque.
The scientific stance of Project Stars is defined by the following principles.
A thought is an attributed object with semantic, relational, evidential, and review-bearing structure.
Candidate links are hypotheses. Confirmed links are reviewed commitments.
Every surfaced relation must preserve the basis on which it was proposed.
Accepted structure and hypothetical relation should not be conflated.
Intention does not map directly to social outcome. Outcomes are filtered through environment, institutions, norms, opportunity, and constraint.
A pathway may be structurally present without being equally operative under every condition.
Symmetry is used for fairness of theory testing. Asymmetry is introduced for realism.
The visible 2D map is a projection of the underlying system, not the full conceptual structure.
The visible graph is a time-dependent embedding:
p(t) : Theta -> R^2
Each thought also carries velocity:
v_theta(t) = d/dt p_theta(t)
The force-layout system can be written schematically as:
m_theta * p_theta'' =
sum of pairwise repulsion forces+ sum of spring forces over confirmed edges+ weak spring contribution from selected strong candidate edges+ activity-sensitive modulation from effective pathways+ weak centering force- damping
This means that:
- repulsion prevents collapse,
- confirmed edges preserve reviewed structure,
- strong candidate hypotheses can weakly influence geometry,
- active pathways change visible emphasis and local pull,
- centering keeps the system bounded,
- damping stabilizes motion.
The layout can also be interpreted through an energy lens as approximately minimizing a combination of:
- spring energy,
- repulsion,
- weak centering,
- hypothesis-sensitive tension,
- and condition-sensitive pathway activation.
A major feature of the current atlas is the separation between internal disposition and external outcome.
Rather than assuming direct transitions such as:
Intention -> KindnessIntention -> Crime
Project Stars models mediated pathways such as:
Intention -> Action Tendency
and then:
Action Tendency + Opportunity + Norms + Enforcement + Empathy + Resources -> Outcome
Key mediator nodes include:
- Action Tendency
- Opportunity
- Constraint
- Norm Pressure
- Enforcement
- Incentive
- Legitimacy
- Visibility
- Risk
- Resources
This makes the atlas more appropriate for representing uneven evidence, conflicting pressures, institutions, and time.
The atlas includes a continuous activation background that evokes:
- propagation,
- resonance,
- disturbance,
- and emergence.
A selected node acts as a local disturbance source in a scalar field:
phi(x, t)
The graph provides discrete conceptual structure, while the background field provides a continuous representation of local activation and influence.
This supports the interpretation that ideas are not only connected, but also dynamically active within a wider conceptual environment.
Receiver quality is represented through a tunable state parameter:
Q in [0,1]
The atlas exposes a receiver-quality control for experimental manipulation. Low values correspond to noisier or weaker resolution; high values correspond to stronger and more coherent conceptual response.
A more realistic interpretation of the system treats receiver state as heterogeneous rather than globally uniform.
This allows different nodes, groups, or domains to respond differently to the same pressure. In practical terms, one part of the atlas may interpret a signal as:
- threat,
- cooperation,
- ambiguity,
- or irrelevance,
while another part responds differently.
The observability bundle Xi may include:
- precision,
- coherence,
- convergence,
- receiver state,
- moral valence,
- collective pull,
- pathway activation.
These metrics support comparison across experimental conditions.
Project Stars supports structured scenario-based interpretation.
Used to inspect symmetry, mirrored moral structure, and fairness of graph geometry.
Used to study realism under institutions, scarcity, conflict, leadership, norms, and historical pressure.
Used to test the stability of formation under disturbance, such as misinformation, scarcity spikes, or threat amplification.
Used to study the recovery of trust, legitimacy, and cooperation after breakdown.
Across these scenarios, the evidence graph remains stable while operative pathways shift in salience and influence.
The detail inspector is the primary explanatory surface of the application.
For a selected node it may show:
- domain,
- degree,
- evidence class,
- consensus,
- review state,
- candidate count,
- description,
- references,
- confirmed connections,
- possible related thoughts,
- relation notes,
- and conditionally active pathways where represented.
Scientifically, the inspector is where the graph becomes interpretable. It translates:
- topology into readable relation lists,
- candidate inference into inspectable evidence,
- pathway emphasis into condition-sensitive interpretation,
- and metadata into epistemic context.
The atlas supports multiple forms of navigation and inspection, including:
- constellation view,
- structural grouping,
- local focus,
- search,
- domain filters,
- minimap,
- zoom,
- fit,
- home reset,
- pause or resume,
- drag and manual inspection.
These are not merely visual presets. They are different ways of reading the same research object:
- global structure,
- domain organization,
- causal neighborhood,
- and local disturbance.
The export layer preserves:
- nodes,
- edges,
- categories,
- metadata,
- confirmed vs candidate distinction,
- provenance,
- review state,
- rationale,
- score components,
- receiver-state metrics,
- pathway-relevant activity fields,
- and scenario-relevant fields where available.
This is essential to treating the atlas as a research artifact rather than a transient visualization.
The design draws from several research areas:
- graph drawing and information visualization,
- interpretable link prediction,
- bibliometrics and citation structure,
- similarity and retrieval,
- ontology alignment,
- human-centered knowledge systems,
- dynamical systems,
- predictive processing,
- and energy-based interpretations of organization.
These influences are used analogically and computationally rather than as claims of full identity with any one framework.
The main contribution of Project Stars is the coupling of several ideas into a single interpretable atlas:
-
Thoughts as structured research objects
Nodes carry provenance, evidence, review, and ontology-bearing metadata. -
Epistemically separated relation layers
Confirmed and candidate relations remain distinct. -
Explainable relation inference
Candidate links are surfaced through explicit weighted signals. -
Balanced testing versus realistic world modeling
Symmetry is used for fairness tests; asymmetry is introduced for realism. -
Mediated outcome formation
Internal states do not map directly to social outcomes; environment and institutions intervene. -
Layered social representation
Individual, interpersonal, institutional, and collective levels coexist in one framework. -
Receiver-state experimentation
Precision and coherence depend not only on structure, but also on tuning and interpretation. -
Condition-sensitive operative pathways
The same evidence graph can yield different active routes under different receiver and world conditions. -
Research-preserving export
The artifact retains rationale, provenance, uncertainty, and experiment-relevant metadata.
The current atlas has important limits:
- it is client-side,
- the layout and field models are deliberately lightweight,
- the scoring function is interpretable but hand-weighted,
- candidate quality depends on the quality of source text and references,
- the ontology remains curated rather than exhaustive,
- the field layer remains partly metaphorical,
- effective pathways are condition-sensitive visual and dynamical abstractions rather than empirical causal proof,
- large-scale collaborative review is not yet implemented.
The project prioritizes interpretability and explicit structure over automation.
Natural next steps include:
- stronger embedding-based retrieval,
- calibrated or learned relation weights,
- richer ontology structure,
- explicit group-level receiver models,
- direct memory terms such as trust history and conflict residue,
- richer shock models,
- longitudinal scenario comparison,
- direct pathway logging and replay,
- and stronger export-to-analysis workflows.
A practical mapping from theory to implementation is:
Theta-> node collection,A-> node fields such as label, description, evidence, review, provenance, and layer,RcandRp-> confirmed and candidate edge collections,Reff-> condition-sensitive pathway activation over the existing graph,S-> candidate-edge recomputation and stored score components,p(t)-> node positions, velocities, and animation ticks,phi-> background field and disturbance rendering,Q-> receiver-quality controls,E-> environment and mediation nodes,H-> memory-oriented state where represented,Xi-> live experiment metrics and scenario readouts.
Project Stars now ships with a versioned atlas dataset in data/atlas-v1.0.json.
- Atlas v1.0
- 108 nodes
- 344 confirmed edges
- 0 packaged candidate edges in the seed dataset
- 10 categories
The renderer and the dataset are now separated. This makes it possible to archive the atlas independently, tag dataset revisions (for example data-v1.0), and publish the data layer to Zenodo separately from the browser application.
Serve the repository from a static web server so the browser can fetch the JSON dataset and evaluation files.
python3 -m http.server 8000Then open http://localhost:8000/.
Research-oriented files added in this package:
data/atlas-v1.0.jsondata/candidates-groundtruth.jsonjs/evaluator.jsresults/ablation-results.jsontests/scoreCandidate.test.jsdocs/data-model.mddocs/evaluation.md
This package includes a first-pass reproducibility report in results/ablation-results.json and summary figures in results/figures/.
The current results are best read as release-candidate benchmarks rather than final publication tables. They are useful for regression checks, packaging validation, and iterative theory testing.
- AUC-ROC: 0.6528
- Precision@5: 0.6000
- Precision@10: 0.6000
- Mean Reciprocal Rank: 0.2225
- Candidate threshold used for packaged ablations: 0.17
| Metric | Confirmed-only | +Candidates | Δ |
|---|---|---|---|
| Coherence (Xi proxy) | 1.0000 | 1.0000 | +0.0000 |
| Fragmentation | 0.0000 | 0.0000 | +0.0000 |
| Modularity proxy | 0.4070 | 0.4086 | +0.0016 |
| Layout energy proxy | 0.9225 | 1.0414 | +0.1189 |
| Metric | Balanced | World | Δ |
|---|---|---|---|
| Fragmentation | 0.0000 | 0.0000 | +0.0000 |
| Modularity proxy | 0.4086 | 0.4101 | +0.0015 |
| Moral-valence balance | 0.1499 | 0.1338 | -0.0161 |
| Layout energy proxy | 1.0414 | 1.0612 | +0.0198 |
| Metric | Low-Q | High-Q | Δ |
|---|---|---|---|
| Coherence (Xi proxy) | 1.0000 | 1.0000 | +0.0000 |
| Modularity proxy | 0.4112 | 0.4098 | -0.0014 |
| Moral-valence balance | 0.1263 | 0.1756 | +0.0493 |
| Layout energy proxy | 1.1606 | 0.9839 | -0.1767 |
| Metric | Crisis mean Δ ± std | Repair mean Δ ± std |
|---|---|---|
| Coherence (Xi proxy) | -0.0017 ± 0.0036 | -0.0020 ± 0.0050 |
| Layout energy proxy | +0.0008 ± 0.0210 | +0.0035 ± 0.0175 |
| Modularity proxy | +0.0029 ± 0.0120 | +0.0020 ± 0.0109 |
Unless otherwise stated for third-party materials, services, or marks:
Copyright (c) 2026 Pezhman Farhangi
See LICENSE and THIRD_PARTY_NOTICES.md for governing terms and ownership boundaries.