Observe
Observational intake
Retrieval lanes surface traceable evidence, score each source, and bind every claim to an auditable substrate.
RetrievalOnline
Source-bound
WU-WEISM · CAUSAL SCIENCE
Causal Science.
Automated.
From observation to counterfactual — Pearl's causal ladder, automated with institutional rigor.
The Laboratory
Two instruments. One discipline.
For Researchers
Your hypotheses. Causally governed.
Connect your own API keys. Index your corpora. Every claim is contradiction-tested, every inference is traced back through the causal graph before it becomes a conclusion.
- Rejection-aware RAG with sovereign memory
- Hong-inspired hypothesis recombination
- Novelty proofing against indexed literature
For Institutions
Scientific governance at scale.
Multi-agent critique, causal provenance trails, and falsifiability gates baked into every synthesis cycle — so your research output meets institutional-grade epistemic standards.
- Epistemologist · Skeptic · Architect agents
- End-to-end causal provenance audit
- Falsifiability gate before memory commit
Manifesto
Institutional‑grade rigor. Neural‑speed synthesis.
Science has always moved forward through disciplined contradiction. MASA automates the infrastructure of causal discovery — so researchers spend their time at the frontier, not the pipeline. Every hypothesis is generated, critiqued, and validated against physical reality before it becomes a claim.
Causal Pipeline
From Question to Proof.
A closed-loop scientific process. Each stage is governed, traced, and auditable before MASA commits a conclusion to action or memory.
Evidence enters with source, time, and scoring intact.
Contradiction is preserved until a mechanism survives it.
Interventions are simulated on structure, not guessed from correlation.
Only refuted-safe findings reach memory or action.
Monitored Sequence
Question to Proof Runtime
Hypothesize
Mechanism drafting
Contradiction pressure forces mechanism candidates to survive conflict before MASA treats them as causal structure.
SynthesisOnline
Contradiction-tested
Intervene
Interventional runtime
Counterfactual interventions are simulated on the structural graph rather than guessed from correlation or intuition.
Do-calculusPrimary
Action-simulated
Validate
Refutation gate
Only refuted-safe findings clear provenance, falsifiability, and audit thresholds before they reach memory or action.
CommitOnline
Refuted-safe
Research Surfaces
Three purpose-built instruments.
Each surface is purpose-built for a distinct mode of causal inquiry. All share the same underlying SCM engine and provenance trail.
Causal Workbench
/chatPearl SCM
Structured causal dialogue with intervention framing. Every exchange moves you up Pearl's ladder — from observation, through intervention, to counterfactual reasoning.
Open →Hybrid Synthesis
/hybridNovelty Proofing
Surface genuine novelty by reconciling conflicting claims across your literature corpus. Hong-inspired recombination builds bridges where consensus fails.
Open →Legal Causation
/legalBut-For · Proximate
Causation analysis that meets legal epistemological standards. Applies but-for and proximate cause frameworks to complex multi-factor causal chains.
Open →The Shift
The world changed. The method didn't.
Until now. MASA replaces the informal scientific pipeline with a governed causal architecture — without sacrificing researcher autonomy.
Today's Reality
Hypothesis drift
Months of literature review with no causal structure — observations pile up, mechanisms remain opaque.
Opaque methods
No provenance, no reproducibility. Conclusions emerge from intuition, not traceable inference chains.
Bottleneck science
Weeks per synthesis cycle. Contradictions are avoided, not reconciled. Novelty is claimed, not proved.
The MASA Layer
Pearl SCM
Structural causal models built in real time. DAGs constructed from your corpus, not your assumptions.
Causal provenance
Every inference traced end-to-end. The full reasoning chain is auditable at every rung of the ladder.
Real-time synthesis
Minutes, not months. Contradictions are the engine of discovery — MASA reconciles them systematically.
Architecture
Five Pillars. One Causal Engine.
Every capability is purpose-built for causal governance. Together they form a closed-loop automated scientist: one governed runtime, five operational disciplines, no ungoverned path to output.
Runtime Ledger
Each lane monitors one discipline. Focus a pillar to inspect its role, signal strength, and immediate dependencies inside the same governed board.
Governed Runtime Board
The Mechanism
Sequence of Events.
How a research question becomes a validated causal conclusion — every stage governed, every transition auditable.
01
INGEST
Index your corpus
Research papers, PDFs, web sources — MASA extracts entities, relationships, and raw causal claims. Every document becomes a node in the knowledge graph.
02
RETRIEVE
Confidence-weighted evidence
Rejection-aware RAG surfaces only the evidence that survives scrutiny. Each retrieved fragment carries a source score, a temporal decay factor, and a contradiction flag.
03
SYNTHESIZE
Contradict and recombine
Contradiction detection drives hypothesis generation. MASA builds structural causal models from conflict, not consensus — then tests their logical coherence.
04
VALIDATE
Pearl do-calculus
Counterfactual simulation. Intervention framing. Every candidate conclusion is interrogated through the causal graph before it leaves the synthesis engine.
05
COMMIT
Sovereign memory update
Only claims that pass the falsifiability gate enter long-term memory. The full provenance trail — sources, reasoning steps, confidence deltas — is persisted alongside.
Common Questions
What researchers ask.
A standard AI assistant generates plausible text. MASA generates falsifiable hypotheses. Every claim passes through Pearl's do-calculus before it's committed to memory — meaning causal structure, not correlation, drives the output. The system explicitly tracks what it doesn't know, flags contradictions, and refuses to paper over them with confident-sounding prose.