Evidence contract — Iris

There is no separate evidence tool. Each of the four tools returns the evidence on the tool result.

Every tool result includes:

• envelope_version: "1"
• status, tool, product (iris), product_version
• route as { engine, path }
• warnings and gaps (arrays; either may be empty)

read_image puts geometry and metadata on twin (source file name, mime, dimensions, metadata, trust warnings). OCR is a sibling ocr object, and only when include_ocr is true or profile is quality.

An OCR line is {text, bbox: {x, y, width, height}, confidence}. The default language is eng.

If Tesseract is missing, status stays ok. ocr.available is false, and gaps gains {code: "OCR_UNAVAILABLE", message}.

image_probe evidence is the camelCase probe object. crop_region evidence is the camelCase region_evidence object (imageBase64 only when requested). compare_images evidence is the snake_case diff object. Inside a read_image region, crop bytes are image_base64 and only when include_region_image is true.

GPS is always redacted. OCR text never replaces dimensions or hashes.