# Foil Defect Simulation v2: public method and data notes

**PUBLIC DATASET — SIMULATED — NOT FACTORY TEST DATA**

This package is the public companion data for the article [Foil Label Defects: Causes, Checks and Practical Fixes](https://www.labelsagent.com/common-foil-label-defects/). It illustrates how a simulated record set can be summarized and interpreted. It does not report physical testing, factory production results, customer orders, or confirmed material performance.

## What this extract contains

`foil-defects-simulated-records-v2.csv` contains 1,020 simulated unit-level records generated on 2026-08-28 by the model `foil-literature-virtual-v2.0`, using random seed `260822`.

The virtual design represents 30 virtual target sheets, each with 34 units. It spans five model material categories, two process categories, and three virtual process stages. The records are useful only as an illustrative classification and data-reading example.

Every CSV row is marked `DATA_SOURCE_TYPE=SIMULATED` and carries this disclaimer: `SIMULATED - NOT FACTORY TEST DATA. Illustrative only; not measured, not a production result, and not factory validated.`

## What the extract does not contain

There were **zero physical measurements, zero factory-validated results, zero production batches, and zero image observations** in this data set. It must not be used to select a material, predict a defect rate, set a specification, compare suppliers, or make a production claim.

The public extract intentionally excludes internal model inputs and fields that could be mistaken for measured performance:

- normalized material priors, risk variables, model weights, probabilities, scores, and pass/fail labels;
- generated reason strings and primary-risk labels;
- internal evidence IDs and the complete evidence registry;
- source workbooks, photos, equipment details, and internal run notes.

The package is therefore auditable for its visible record count and code totals, but it is **not independently reproducible as a physical experiment** and does not provide a validated production model.

## CSV field guide

| Field | Meaning |
| --- | --- |
| `PUBLIC_RECORD_SCHEMA_VERSION` | Public extract schema version; currently `1.0`. |
| `RECORD_ID`, `SAMPLE_ID`, `UNIT_ID` | Virtual record identifiers. They do not identify a factory sample or production lot. |
| `FEATURE_TYPE`, `SPEC`, `POLARITY` | Virtual feature descriptors used for this simulated example. |
| `MATERIAL_CODE`, `PROCESS_CODE`, `REPLICATE` | Model scenario labels only; they are not availability, performance, or supplier claims. |
| `SIMULATED_DEFECT` | The model-assigned defect code, or blank when no code was assigned. |
| `DATA_SOURCE_TYPE` | Always `SIMULATED`. |
| `MODEL_VERSION`, `RANDOM_SEED` | Run identifiers for this exact simulated extract. |
| `DISCLAIMER` | Row-level statement of the data limitation. |

## Defect-code summary

| Code | Simulated records | Share of 1,020 | Interpretation in this model |
| --- | ---: | ---: | --- |
| `BK` | 149 | 14.6% | Broken foil line or stroke |
| `FI` | 141 | 13.8% | Fill-in in reversed or fine detail |
| `PH` | 116 | 11.4% | Pinholes or missing foil, combined code |
| `SR` | 20 | 2.0% | Print-to-foil registration offset |
| `RG` | 145 | 14.2% | Rough or jagged edges |
| `EG` | 83 | 8.1% | Edge feathering or bleed-like spread |
| `DB` | 1 | 0.1% | Mottled or uneven coverage |
| blank | 365 | 35.8% | No defect code assigned by the model |
| **Total** | **1,020** | **100.0%** | — |

The shares are calculated as `code count ÷ 1,020`, rounded to one decimal place. They describe this one seeded simulation only; they are not observed defect frequencies.

## Method boundary and limitations

The underlying virtual model used literature-informed assumptions to create a teaching dataset. Its results depend on its scenarios and seed. A physical follow-up would need dated protocols, sample and configuration IDs, raw observations, photos where appropriate, decision rules, sample size, deviations, and production or technical review before any result could be described as a test finding.

Do not combine this v2 extract with a later simulation version or with measured records. Any future measured dataset should be released separately and labeled with its own method, date, reviewer, and limitations.

## Integrity checks

- Original internal v2 simulated source SHA-256: `d1608bf345b96c77b4778060917d16aec59919f5a15fe1e966e86d591e89e519`
- Public CSV SHA-256: `e795e199558d2e8d02a21368b4f9de804d65774f3319155019c018e6f9b7a37c`
- Expected CSV record count: `1,020`
- Expected `DATA_SOURCE_TYPE` value: `SIMULATED` only

## Suggested article disclosure

> Download the simulated record extract and method notes. Both files are labeled **SIMULATED — NOT FACTORY TEST DATA** and are provided only to show the example’s inputs and limitations.