[Shivam Sharma (ShivamSharma43)]

Description

Fixes #3742.

JsonConversationPrinter._build_piece_async exported each piece with a single data_type, set to original_value_data_type, next to both original_value and converted_value. The converted value's type was lost. The HTML scenario report switched on that original data_type for both values, so a converted value was labelled with the wrong type.

Since #3771 merged, the report shows "Original:" and "Converted:" side by side when the two values differ. With this change each of them gets its own type:

Piece (original → converted)	Converted value on main	Converted value with this change
text "draw a cat" → image_path media/cat.png	<div class="piece">media/cat.png</div>	<div class="piece media">[image_path] media/cat.png</div>
image_path media/dog.png → text "a dog on a log"	<div class="piece media">[image_path] a dog on a log</div>	<div class="piece">a dog on a log</div>

The original value keeps its own type label (draw a cat stays plain text, [image_path] media/dog.png stays media).

• pyrit/output/conversation/json.py: each non-reasoning piece now also carries original_value_data_type and converted_value_data_type. data_type stays as it was (the original type), so existing consumers of the JSON keep working, and reasoning pieces are unchanged.
• pyrit/output/scenario_result/html.py: the template labels the original value with original_value_data_type and the converted value with converted_value_data_type. When only one value is shown (no converted value, or it matches the original), it uses the type of the value it shows. Both fall back to data_type, so older JSON documents without the new fields render as before.

Tests and Documentation

New tests:

• tests/unit/output/conversation/test_json.py::test_build_async_exports_original_and_converted_data_types: a text → image_path piece exports both types, and data_type stays text.
• tests/unit/output/scenario_result/test_html.py::test_html_report_labels_converted_media_by_converted_type: a text → image_path piece renders the original as plain text and the converted value as [image_path] media/cat.png.
• tests/unit/output/scenario_result/test_html.py::test_html_report_renders_converted_text_of_media_original_as_text: an image_path → text piece renders the original as [image_path] media/cat.png and the converted value as plain text.
• tests/unit/output/scenario_result/test_html.py::test_html_report_labels_unchanged_value_by_converted_type: when the value is unchanged but the converted type differs, the single shown value uses the converted type.
• tests/unit/output/scenario_result/test_html.py::test_html_report_labels_converted_types_from_real_builders: both directions built from real MessagePieces by JsonConversationPrinter and rendered to HTML.

tests/unit/output passes (410 tests, including the #3771 tests), and so do the JSON export consumers in tests/unit/cli/test_output.py, test_scenario_report_parity.py, test_results.py and tests/unit/analytics/test_scenario_statistics_parity.py. ruff check, ruff format --check and ty check are clean on the changed files.

No documentation changes: the docs don't list the per-piece JSON fields. I didn't run JupyText because no notebooks are affected.

This change was prepared with the help of Claude Code (an AI coding assistant). I reviewed the change and the tests.

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

[Shivam Sharma (ShivamSharma43)]

The only red check is build-book / Build latest, which hit the 6h job timeout without failing a step. The unit tests and the frontend tests pass. Could someone rerun it when convenient? Thanks!