QSE Engine Version: v1.0•Deterministic•Probabilistic•Confidence-Based

QSE Pipeline

How propositions flow through the Quantum State Engine.

This page documents the seven-stage execution pipeline that every QSE analysis follows: from input validation and module routing through factor scoring, quality assurance, and final confidence allocation. The same pipeline powers both the Proposition Console and the automated Edge Spotter signal generation.

One module is different. The Chemical Exposure module does not use this seven-stage flow. It runs a dedicated, safety-first pipeline with a disclaimer acknowledgment gate, structured input validation, safety intercepts, dose normalization, and a strict four-key public payload. See the Chemical Exposure Pipeline.

Pipeline Overview

The system processes markets through a gated pipeline designed to filter noise and enforce reliability.

Every QSE analysis follows a strict seven-stage pipeline. Each stage validates its inputs and produces well-defined outputs. If any stage fails validation, the system returns a "FAIL" status rather than fabricating data (fail-closed principle).

Input → Router → Executor → QA → Repair → Engine → Serializer

Pipeline Stages

Input Validation

Validates and normalizes the incoming proposition

•Trims and validates proposition text
•Rejects interrogatives (questions)
•Normalizes claims to canonical form
•Generates unique Run ID

Output: Validated proposition + Run ID

Router

Determines the appropriate analysis module

•Analyzes proposition content
•Detects sports-specific signals (NBA, NFL, etc.)
•Routes to General or specialized module
•Records routing decision for transparency

Output: Selected module (General / NBA)

Executor

Generates factor scores based on evidence

•Applies module-specific factor taxonomy
•Scores each factor: -1, -0.5, 0, +0.5, +1
•Assigns weights (1-10) per factor
•Pads to exactly 30 factors if needed

Output: 30 scored factors with weights

QA Gate

Validates executor output integrity

•Verifies factor count = 30
•Validates score values in allowed set
•Checks weight bounds [1-10]
•Confirms polarity alignment

Output: Pass / Fail + validation report

Repair (if needed)

Attempts to fix validation failures

•Triggered only if QA fails
•Snaps out-of-range scores to nearest valid
•Clamps weights to [1-10]
•Re-validates after repair

Output: Repaired factors or FAIL

Engine

Computes confidence allocation

•Calculates raw sum R = Σ(score × weight)
•Applies sigmoid: p = 1 / (1 + e^(-αR))
•Applies suppression for uncertain signals
•Maps to quantum state probabilities via Gaussian kernel

Output: State probabilities + final label

Serializer

Formats output for display

•Selects top 5 factors for public view
•Hides internal weights and raw scores
•Generates caveats and flip conditions
•Adds metadata (timestamp, run ID)

Output: PublicOutput JSON

Fail-Closed Behavior

On Success

All stages pass validation → Complete PublicOutput with confidence allocation, factors, and metadata.

On Failure

Any stage fails validation → Returns { "status": "FAIL" } with no partial data. No fabrication, no guessing.

Key Principles

Deterministic

Same inputs always produce the same outputs. No hidden randomness.

Probabilistic

Distributes confidence across states rather than producing point-estimate forecasts.

Confidence-Based

Outputs reflect probabilistic confidence across a structured state lattice.

Fail-Closed

Returns FAIL rather than fabricating data on error.