I. Data Triangulation

Every visual reconstruction begins with Primary Evidence. We cross-reference three distinct data streams before generating a single pixel:

• Textual Evidence: Epigraphic surveys of the Wadi el-Hôl inscriptions.
• Topographical Data: Latitude/Longitude and elevation data from the Theban Desert Road Survey.
• Archaeological Reports: Ceramic and osteological remains (e.g., the presence of donkeys vs. the historical absence of camels in the Middle Kingdom).

II. Prompt Engineering as Curation

Prompts are constructed using controlled archaeological vocabulary. To prevent "generic" desert imagery, the following parameters are strictly enforced:

• Geological Specificity: Using terms like "limestone scarps," "desert varnish," and "wadi outwash" to reflect the specific Qena Bend environment.
• Negative Constraints: Explicitly forbidding modern infrastructure, camels (pre-Ptolemaic), and inaccurate vegetation.
• Atmospheric Realism: Adjusting for the harsh, high-contrast lighting characteristic of the Western Desert.

III. The Iterative Loop

AI generation is an iterative process. Initial outputs are audited against historical maps (e.g., the Farshût Road path). If a reconstruction fails to align with the Theban Plateau's actual topography, the prompt is refined and the image is regenerated. This ensures that the visuals serve the narrative without sacrificing geographical truth.