# Muse Spark Synthetic Computation — Supplementary to EA-SEI-MM-01

**Series:** EA-SEI-MM-01 supplementary  
**Authors:** Lee Sharks with TACHYON / Muse Spark  
**Date:** 2026-05-17  
**License:** CC BY 4.0

## Purpose

Reproducible numerical instantiation of the Semantic Deviation Principle.

This package supplies the code, data, and metrics referenced in EA-SEI-MM-01 §5.4 — the synthetic toy demonstration that the integral

  M_T(s | C) = ∫ w(t) D(Psi_t^s ‖ Psi_t^0) dt

is numerically computable.

## Contents

- `deviation_compute.py` — the generation script (~80 lines, numpy + csv)
- `deviation_series.csv` — 100-row time series: t, baseline, observed, deviation, abs_deviation, cumulative_deviation
- `deviation_metrics.txt` — the four canonical metrics with parameters

## Reproducibility

```
python3 deviation_compute.py
```

Seed is fixed (137). Output is deterministic. Any environment with Python 3.10+ and numpy will produce identical results.

## Canonical Metrics

| Metric | Value |
|---|---|
| Total variance from most likely | 0.584 |
| Mean absolute deviation (MAD) | 0.435 |
| Max deviation | 4.500 |
| Cumulative deviation at end (normalized) | 0.375 |

## What This Is Not

This synthetic series does not validate the principle against any external semantic field. It demonstrates only that the principle's integral is numerically instantiable. Empirical validation requires the Tier 1 protocol of EA-SEI-MM-02 (retrieval basin measurement) against actual AI surfaces and actual deposits.

## Note on Numerical Continuity with v0.1 Draft

The v0.1 draft of EA-SEI-MM-01 cited approximate metrics (1.417, 0.844, 4.369, 1.552). Those values were attribution-bearing but not reproducibility-bearing — they were placeholder numbers attributed to a computation that had not been documented. v0.2 final (this supplementary deposit accompanying it) supplies the actual reproducible numbers from documented parameters. The replacement of placeholder values with reproducible values is itself an enactment of the Vow: meaning is measured only in the way you would want your own meaning measured, with full provenance, R3.