DCG research · Framework + reproducible empirical benchmark

Causal intelligence for decisions that cannot be rehearsed.

DCG Intelligence is a causal operating model for high-stakes intervention: it structures evidence, reconstructs counterfactual worlds, challenges causal claims and carries qualified options through explicit governance. This release introduces the framework, reproduces a 39-state empirical benchmark and maps independent deployment pathways.

source verifieddeterministicfalsifiableaudit ready

Layer 01 · DCG framework

What DCG is and why it exists

The domain-agnostic causal operating model: representations, intervention logic, world branching, validation and governed action.

Layer 02 · Case study 01

What the empirical benchmark establishes

A reproducible evaluation using California Proposition 99, 38 donor states, synthetic controls and explicit stress tests.

Layer 03 · Application portfolio

Where the framework can be deployed

Separate deployment blueprints for defence, cyber, strategy, agriculture, health, energy and sovereign intelligence.

The decision problem

What if you could test a strategy before reality made it irreversible?

Most dashboards describe what happened. Most forecasts extend what already exists. DCG Intelligence models what changes when you intervene—and keeps every assumption, branch and result auditable.

Structure the decision

Define identities, state, mechanisms, constraints, evidence and authority before a model is allowed to recommend action.

Construct missing worlds

Compare action, inaction and alternative interventions through explicit counterfactuals and versioned scenario branches.

Challenge the claim

Use holdouts, placebos, sensitivity tests, falsification and external benchmarks to expose fragility.

Govern the action

Carry qualified evidence into authority, execution, monitoring and learning without losing the decision lineage.

The DCG difference

From data to governed action in one causal operating model

DCG combines capabilities that are usually fragmented across knowledge graphs, forecasting tools, simulators, causal inference packages and governance workflows.

01 · CTPS

Structure reality

Stable identities, state, forces, direction, couplings and provenance.

02 · CAUSAL ENGINE

Estimate impact

Construct counterfactuals and measure the difference an intervention makes.

03 · CLOSURE

Challenge stability

Test whether the model closes consistently before trusting the break.

04 · FSE

Explore futures

Fork versioned worlds while preserving lineage and evidence class.

05 · SDE

Govern decisions

Apply admissibility gates, authority and continuous learning.

06 · SGM

Read power configuration

Map coercion, resources, legitimacy and luminosity when power is decision-relevant. SGM is part of the wider framework, not an input to the Proposition 99 effect estimate.

Intervention-native

The central question is not “what comes next?” but “what changes if we act, wait, redirect or stop?”

Multiple futures, one lineage

Every scenario inherits a traceable parent, evidence state, intervention and causal graph diff.

Confidence without opacity

DCG exposes where evidence is strong, where it is suggestive and what must be learned next.

Estimated annual reduction

—

fewer cigarette packs per capita / year

Relative reduction

—

versus the internal synthetic twin

Spatial placebo rank

—

live model · full published design ranks 1 / 39

Executable integrity

—

data, model and governance checks

Case logic and finding

Everything below this point refers specifically to Proposition 99.

A real intervention

In 1988, California approved Proposition 99: a US$0.25 cigarette tax increase plus a statewide tobacco-control program.

The missing comparison

California after the policy is observable. California without the policy is not. That unobserved trajectory is the case-specific causal question.

A synthetic control

The implementation reconstructs the counterfactual from 38 untreated states, then challenges the result across time, donors and spatial placebos.

Replicated magnitude

The public model estimates 19.7 fewer packs per capita per year, closely reproducing the magnitude reported by the peer-reviewed study.

Case study 01 · Live estimation

California versus two synthetic counterfactuals

The published line applies the five donor weights reported in the 2010 study. The independent line is recalculated here from outcome history alone, with λ = 0.1 and 3,500 deterministic iterations.

Observed CaliforniaSynthetic · published weightsSynthetic · live reproduction

The intervention gap, year by year

Negative values mean observed sales fell below the estimated no-policy world.

What builds the synthetic twin

The live model finds a weighted combination of donor states whose pre-1989 trajectory resembles California.

State	Stable identity	F512*	Weight

*Optional deterministic identity slot; it has no causal or statistical role.

Case study 01 · Scientific validation

A case result becomes decision-grade only after serious attempts to break it

This section tests the Proposition 99 estimate across space, time and composition: 39 state placebos, a temporal holdout, a false intervention date, donor removal and alternative pre-policy windows.

Spatial placebo distribution · post/pre MSPE

Every state is treated as if it had implemented the policy in 1989. California should stand apart if the signal is meaningful.

embedded reference

Core diagnostics

What the numbers say: the compact in-browser estimator ranks California 3rd of 39 (p = 0.077). The full peer-reviewed specification—with covariates and the original donor design—reports 1st of 39 (p = 0.026). The lightweight engine reproduces magnitude more strongly than statistical separation.

Leave-one-donor-out

The model is re-estimated after removing each of its six largest donor weights.

Time windows and a false intervention date

Tests whether the signal survives shorter histories and whether a similar break appears before the real policy.

Bootstrap/closure as a stability layer

Applied to the pre-1989 gap, the closure operator is an estimated AR(1). Contraction means small discrepancies tend to return toward a local fixed point—giving the post-policy break a stable baseline against which to be interpreted.

DCG framework · Architecture

One decision object. Five specialized intelligence layers.

This section defines the general architecture. Proposition 99 appears as a clearly identified mapping between the framework and case-specific evidence.

Framework definition · The universal causal unit

U = ⟨id, state, drivers, mechanisms, outcomes, couplings, intervention, evidence, provenance⟩

Layer	Core fields	Role in this demonstration
CTPS	State · Force · Direction · Coupling	Represents observed state, policy force, expected direction and empirical couplings.
Causal engine	treatment · donors · counterfactual · effect	Builds the synthetic California and measures the intervention gap.
Bootstrap/Closure	operator · fixed point · stability	Tests closure of pre-policy error and characterizes the structural break.
FSE	Context · Agents · Mechanisms · Outcomes	Versions observed, counterfactual and exploratory worlds with lineage.
SDE/SLP	gates · admissibility · authority	Converts analysis into a governed decision process rather than a naked ranking.

Case 01 illustration · Evidence boundaries

This graph maps the general unit into Proposition 99 solely to show how framework fields receive case-specific evidence.

Interventions are typed—not treated as generic events

Type	Meaning	In this case
hard_do	sets a variable directly	outside the current case design
parameter	changes a structural parameter	US$0.25 tax increase per pack
policy	changes institutional rules and allocation	statewide tobacco-control program
topology	changes causal relationships	future mechanism study
observation	updates evidence without acting	annual cigarette sales

SDE/SLP decision gates

DCG application portfolio

One causal operating model. Mission-critical decisions across industries.

These are independent deployment pathways. Each establishes its own evidence pack, domain model, validation protocol and governance policy.

DCG domain transfer mapone causal kernel · eight institutional evidence packs

Defence & national security

Contested logistics and readiness

deployment blueprint

Which force posture preserves mission readiness when a port, corridor, supplier or communication layer is disrupted?

DCG moveModel bases, assets, suppliers and routes as coupled units; branch disruption, rerouting, stockpiling and non-action; stress-test cascading failure and reversibility.

Auditable dataNATO defence expenditure, U.S. DoD budget and contract data, AIS/port data, UCDP conflict events, national logistics statistics.

Decision outputResilient intervention portfolio with readiness, latency, cost, escalation and optionality gates.

Cybersecurity

Attack-path intervention intelligence

high-value pilot

What should be patched, isolated or monitored first to reduce real blast radius—not merely the highest vulnerability score?

DCG moveCombine asset topology, exploitability, business criticality and identity paths; compare patch, isolate, compensate and wait interventions.

Auditable dataCISA Known Exploited Vulnerabilities, NVD/CVE, FIRST EPSS, MITRE ATT&CK and organizational telemetry.

Decision outputCausal remediation sequence ranked by expected risk reduction, service impact and reversibility.

Corporate intelligence

Competitor, supply-chain and geopolitical moves

public-data ready

If a competitor changes price, capacity or geography—or a sanction hits a supplier—where does the impact propagate next?

DCG moveBuild an evidence graph across companies, products, suppliers, jurisdictions and mechanisms; fork response and non-response strategies.

Auditable dataSEC EDGAR, UN Comtrade, World Bank, WIPO/USPTO, procurement records, earnings calls and verified corporate disclosures.

Decision outputEarly-warning causal map and response portfolio with confidence, trigger and monitoring rules.

Complex strategic decisions

Capital allocation, market entry and policy

executive use case

Which strategy remains governable across demand shocks, regulation, competitor adaptation and execution constraints?

DCG moveConnect strategic assumptions to causal mechanisms; evaluate action and inaction across multiple futures; preserve the decision lineage.

Auditable dataInternal finance and operations, IMF, OECD, World Bank, national statistics, market filings and policy documents.

Decision outputAdmissible strategy portfolio, leading indicators, kill criteria and decision-memory record.

Agriculture & food security

Climate-resilient production

international data

Which combination of irrigation, planting window, cultivar and fertilizer protects yield under weather and price volatility?

DCG moveFuse field state, weather, soil, input prices and management interventions; estimate historical effects and branch seasonal scenarios.

Auditable dataFAOSTAT, NASA POWER, Copernicus/Sentinel-2, USDA NASS, soil maps and farm telemetry.

Decision outputField- and region-level intervention plan with yield, water, cost and resilience trade-offs.

Health systems

Policy and care-pathway impact

empirical extension

Which intervention improves outcomes without shifting risk, cost or congestion elsewhere in the health system?

DCG moveEstimate effects from natural experiments or phased rollouts; model patient flow, capacity and downstream mechanisms; monitor drift.

Auditable dataWHO Global Health Observatory, CDC WONDER, openFDA, claims/EHR under governance, hospital operations and public registries.

Decision outputEvidence-graded intervention with capacity, equity, safety and authority gates.

Energy & critical infrastructure

Grid resilience and transition sequencing

systems pilot

Where should storage, redundancy, demand response or maintenance be deployed to prevent cascading failure?

DCG moveRepresent assets and dependencies as a dynamic causal network; test shocks, maintenance timing and investment portfolios.

Auditable dataU.S. EIA, ENTSO-E transparency data, NOAA weather, grid-operator reports and asset telemetry.

Decision outputResilience portfolio with reliability, cost, recovery time and critical-service gates.

Sovereign intelligence · UAE & USA

Smart-city and national portfolio decisions

public showcase

How should mobility, water, health, energy and economic interventions be sequenced when they share resources and create cross-sector effects?

DCG moveCreate a sovereign causal twin; evaluate portfolios across ministries and agencies; maintain authority, provenance and public accountability.

Auditable dataUAE Open Data, Dubai Pulse, Federal Competitiveness and Statistics Centre, Data.gov, U.S. Census, DOT and NOAA.

Decision outputCross-sector intervention roadmap with national KPIs, resilience, equity, legitimacy and execution gates.

Official-data launchpad

Representative primary sources for internationally auditable demonstrations.

NATO defence data CISA KEV SEC EDGAR UN Comtrade FAOSTAT NASA POWER WHO GHO U.S. EIA UAE Open Data Data.gov

The repeatable pattern

Every deployment follows the same scientific spine

Domain expertise changes the graph and evidence. The operating discipline remains stable.

Evidence pack

Version sources, identities, measurements, assumptions and known gaps.

Causal model

Encode mechanisms, couplings, interventions and plausible interference.

Empirical challenge

Use holdouts, placebos, sensitivity, falsification and external benchmarks.

Governed action

Authorize, monitor, update and learn through an auditable decision cycle.

Case study 01 · Exploratory scenarios

Explore alternative Proposition 99 worlds—without losing scientific lineage

These controls extend the case model only. Adjust policy strength and timing to see how a scenario branch is generated while preserving which trajectory is observed, estimated or exploratory.

Counterfactual base

Intervention strength α1.00×

Implementation delay d0 years

No-policy worldObserved worldExploratory branch

Versioned world-state tree

Exploration scope: this dataset observes one policy level. Alternative strength and delay settings are transparent scenario transformations designed for future domain calibration.

Case study 01 · Evidence and audit

Proposition 99 data, methods, sources and claims—open for inspection

This record documents the public benchmark case. Every future DCG deployment receives its own evidence and decision record. The full CSV and computations live inside this file.

Dataset integrity

Implementation integrity suite

Scientific claim ledger

ID	Claim	Status	Evidence	Evidence scope

Primary sources and provenance chain

Source	Use	Provenance / note
Abadie, Diamond & Hainmueller, JASA 2010	Method, case, published weights and placebos	Peer-reviewed article; the original specification includes additional covariates.
Full paper · MIT Economics	Complete method, findings and limitations	Published result: nearly −20 packs per year on average and p = 1/39.
california_prop99.csv · synthdid	Panel executed in this file	Derived from Abadie et al.; 1,209 rows, 39 states, 1970–2000.
synthdid documentation	Public dataset distribution chain	Academic implementation and replication materials.
California Department of Public Health	Intervention and legislative mandate	Government source for approval, tax increase and program allocation.

Public research provenance

This edition publishes the evidence chain required to inspect and reproduce the benchmark. Controlled research artifacts, internal document identifiers and development records remain in the governed DCG research archive.

Release scope: framework narrative, public benchmark data, executable estimator, diagnostics, claims and third-party sources.

Third-party data notice

The California Proposition 99 panel is redistributed from the synthdid research package and derives from Abadie, Diamond and Hainmueller (2010).

Copyright	Copyright 2019, Stanford University
License	BSD 3-Clause; redistribution conditions and warranty disclaimer apply.
Terms	Official license notice and disclaimer