Make state, assumptions, and uncertainty inspectable.

Design goals

Five operating principles

1. 01

   Evidence before inference

   Retain raw source records and capture metadata before normalization or materialization.

2. 02

   Gaps are data

   Represent a missing interval explicitly. Never make an incomplete stream appear continuous.

3. 03

   Deterministic transitions

   Given identical inputs and an engine version, reconstruction must produce identical state.

4. 04

   Assumptions travel with outputs

   Analytics results remain attached to portfolio snapshots, market state, and user-supplied scenario assumptions.

5. 05

   Views, not prescriptions

   The system describes exposures and conditional outcomes; it does not choose a course of action.

Order book reconstruction

An order book is a state derived from an ordered event stream. Expreci's intended reconstruction process begins from a validated snapshot, applies normalized deltas in source-sequence order, and records the exact reconstruction-engine version. Price and size use decimal strings at interfaces to avoid introducing floating-point ambiguity.

Determinism matters because two researchers should not obtain different book states from the same archived inputs. It also makes corrections auditable: a source correction or engine change produces a new version rather than silently altering the interpretation of prior results.

Sequence and gap handling

1. Partition streams by the narrowest source scope with a valid ordering guarantee.
2. Compare each incoming sequence against the last accepted sequence for that partition.
3. Reject exact duplicates idempotently and retain conflicting duplicates for investigation.
4. On a forward gap, stop declaring reconstructed state valid from that point.
5. Attempt recovery from retained raw events or a later validated snapshot; record the affected interval either way.

When a source does not expose a strict global sequence, the applicable ordering domain and tie-break rules must be documented rather than overstated. Receive timestamps can describe capture order but cannot retroactively create exchange ordering.

Event contracts as binary claims

For a simple YES claim with unit payoff, terminal value can be represented as an indicator of the documented resolution condition. A NO claim is the complementary payoff only when the market's settlement rules make it so. Contract definitions, invalid-market behavior, and settlement sources therefore belong in the data model, not only in display copy.

A portfolio is a collection of those claims plus quantities, costs, cash balances, and settlement relationships. Expreci aims to support multiple valuation views—market mark, terminal state, and depth-constrained liquidation—without presenting any one view as the correct decision basis.

Dependence and joint scenarios

Related event contracts cannot always be analyzed by perturbing each implied probability independently. Events may be mutually exclusive, nested, conditionally related, or linked through a common underlying driver. The intended scenario model supports explicit joint states and logical constraints first; more statistical dependence models would require named methodology, fitted inputs, and validation evidence.

Scenario probabilities, when present, are user inputs or model outputs—not objective truths. Results should distinguish a conditional payoff calculation from an estimated likelihood.

Current limitations

• No production data coverage is claimed yet.
• No uptime or completeness guarantees.
• No predictive models.
• No execution, custody, or advisory services.

This methodology is published so prospective users can evaluate the design.