Counterfactual and decision
Fork the ledger, then difference it.
The FDCS layer answers what-if questions by branching the ledger and executing every branch under the same audit. The answers are conditional on what the document declared — and the runtime never pretends otherwise.
Interventions
- Hard
- A value is fixed. The intervened transaction stops depending on its sources.
- Soft
- A value is nudged. The dependency structure survives, and the effect propagates through it.
The distinction is not cosmetic. A hard intervention cuts edges in the dependency graph; a soft one does not. They answer different questions and they are declared separately.
Intervention conflicts are a validation error, not a runtime surprise. A document that pins the same transaction two incompatible ways is rejected before execution.
Branch differential ledgers
Each branch is a full ledger and gets a full audit. What you read afterwards is the difference between branches, not a single number claiming to summarize them.
A branch that fails its own audit is reported as a failing branch. It is not silently dropped from the comparison, and it does not invalidate the branches that passed.
Fixed points and time
Temporal references let a transaction depend on an earlier state of the ledger. fixed_point_groups iterate to convergence under a declared contract — an iteration bound and a tolerance, both stated in the document.
- Fixed-point convergencea formal proof for arbitrary nonlinear systems
Uncertainty
Discrete uncertainty propagates exactly. Continuous uncertainty uses deterministic Halton sequences and copulas, so a rerun gives the same numbers — which makes it reproducible, and does not make it analytic.
- Deterministic Halton and copula samplesanalytic continuous inference
- A declared branch probabilitylearned or objectively calibrated probability
Decisions
On top of the uncertainty layer sit three finite analyses.
- Policy scoring
- Each declared policy is scored under the declared utility.
- Information value
- What an observation would be worth before you pay for it.
- Finite sequential decisions
- A bounded horizon of decisions, solved exactly rather than approximated.
Finite is the operative word. These are exact answers to bounded problems, not approximations of unbounded ones.
Identifiability
E7 asked whether two different declared structures could be distinguished by the observations available. Sometimes they cannot, and the runtime says so.
- Finite observation equivalencegeneral statistical identifiability
Reporting that two models are observationally equivalent under the available data is a result. Picking one of them and reporting it as the answer would not be.
Where the conditionality lives
Declared probabilities, causal structures, costs, utilities and observation models are model inputs. Every conclusion on this page is conditional on them, and none of them is inferred from the world by the runtime.