# Three inspectable work samples

Andreas Ehstand · prepared with AI assistance and editorial review · 7 September 2026

**All case details and observation counts are synthetic.** This package contains no client records, actual evaluator responses, or observed training and learning outcomes. The sample narratives are in German; this guide explains their purpose and data structure.

| Sample | What to inspect | Readable version |
|---|---|---|
| F03 · AI output evaluator | Twelve statements, six references, and a separate answer key. | [Twelve reference cases](f03/ARBEITSPROBE.md) |
| F07 · Training observation | A completed observation sheet with invented counts and explicit denominators. | [One exercise, separate observations](f07/ARBEITSPROBE.md) |
| F08 · Learning material | Four statements checked against three source cards. | [Short learning-package excerpt](f08/ARBEITSPROBE.md) |

## Reading and editing

Start with the readable sample. CSV files provide editable tables; JSON files provide structured data. CSV uses UTF-8, commas, and quoted text where needed. Data files use decimal points. Units and denominators are explained in the German sample.

For F03 and F08, read the sources and answer the tasks before opening the separate answer key. In F03, only the assigned reference and case text belong in the evaluator input. Do not supply the answer key. F03 also keeps its JSON input and solution files separate. Use the solution files only for the subsequent comparison.

For F07, compare only compatible phases and task conditions. The new task variant has no baseline of its own, so its result cannot establish a transfer gain. These invented numbers illustrate a reporting decision; they do not describe a real participant.

## Separate purposes and limits

F03 examines a defined evaluation criterion. F07 structures observations of one training task. F08 supplies learning materials for an internal trainer. The four-statement F08 excerpt is not the full proposed 60-minute package or a client adaptation.

These samples do not establish evaluator accuracy, causal training effects, learning gains, certification, or a pass threshold. They include no commitment to personal delivery or availability.

The separate paid offers are described on the [pilot page](https://augmanitai.com/pilotprojekt/). Prices, contracting and any further use of underlying works are separate matters.

[Sources and use](SOURCES_AND_USE.md) · [Deutscher Lesehinweis](README_DE.md) · [File manifest and checksums](manifest.json)