import json import os import tempfile import zipfile from typing import Any, Dict, Optional, Tuple, List import gradio as gr from drp.bundle import verify_bundle from drp.diff import diff_bundles from drp.report import render_report_markdown from drp.pdf_report import write_pdf_report from drp.simulate import make_demo_bundle_zip, fork_patch_bundle APP_TITLE = "TimelineDiff — Differential Reproducibility Protocol (DRP)" def _tmp_dir() -> str: return tempfile.mkdtemp(prefix="drp_") def _tmp_path(name: str) -> str: d = _tmp_dir() return os.path.join(d, name) def _as_alignment_rows(alignment: List[Dict[str, Any]]) -> List[List[Any]]: """ Convert alignment dict rows to Dataframe rows. """ rows = [] for r in alignment: rows.append([r.get("i"), r.get("status"), r.get("kind_a"), r.get("step_a"), r.get("kind_b"), r.get("step_b")]) return rows def ui_verify(bundle_file) -> Tuple[str, Dict[str, Any]]: if not bundle_file: return "Upload a bundle zip first.", {"ok": False} ok, summary = verify_bundle(bundle_file.name) msg = "✅ Bundle verifies (hash chain OK)." if ok else "❌ Bundle failed verification." return msg, summary def ui_diff(bundle_a, bundle_b) -> Tuple[str, str, Dict[str, Any], List[List[Any]], str, str, str]: if not bundle_a or not bundle_b: return "Upload two bundles.", "", {"error": "missing input"}, [], "", "", "" diff = diff_bundles(bundle_a.name, bundle_b.name) md = render_report_markdown(diff) # Summary header (quick context above JSON) s = diff.get("summary", {}) parts = [] if s.get("first_divergence_index") is None: parts.append("Timelines identical") else: parts.append(f"Identical until step/index {s.get('first_divergence_index')}") parts.append(f"{s.get('diff_event_count', 0)} events differ") if s.get("missing_event_count", 0): parts.append(f"{s.get('missing_event_count')} missing") if s.get("final_reward_delta") is not None: parts.append(f"Final reward delta: {s.get('final_reward_delta'):.6g}") summary_line = " · ".join(parts) # Export artifacts out_dir = _tmp_dir() out_md = os.path.join(out_dir, "report.md") out_json = os.path.join(out_dir, "diff.json") out_pdf = os.path.join(out_dir, "report.pdf") out_zip = os.path.join(out_dir, "drp-diff-report.zip") with open(out_md, "w", encoding="utf-8") as f: f.write(md) with open(out_json, "w", encoding="utf-8") as f: json.dump(diff, f, ensure_ascii=False, indent=2) # PDF: render from markdown lines (plain, reliable) md_lines = md.splitlines() write_pdf_report(out_pdf, "DRP Differential Report", md_lines) with zipfile.ZipFile(out_zip, "w", compression=zipfile.ZIP_DEFLATED) as z: z.write(out_md, arcname="report.md") z.write(out_pdf, arcname="report.pdf") z.write(out_json, arcname="diff.json") alignment_rows = _as_alignment_rows(diff.get("alignment", [])) status = "✅ Diff complete." return status, summary_line, diff, alignment_rows, out_zip, out_pdf, out_md def ui_generate(seed_a: int, seed_b: int, chaos: float) -> Tuple[str, str, str]: chaos = float(max(0.0, min(1.0, chaos))) path_a = _tmp_path("demo-A.zip") path_b = _tmp_path("demo-B.zip") make_demo_bundle_zip(path_a, seed=int(seed_a), chaos=chaos, label="A") make_demo_bundle_zip(path_b, seed=int(seed_b), chaos=chaos, label="B") return "✅ Generated demo bundles.", path_a, path_b def ui_fork(source_bundle, fork_index: int, patch_kind: str, patch_step: str, patch_payload: str) -> Tuple[str, str]: if not source_bundle: return "Upload a source bundle first.", "" patch_payload_json: Optional[Dict[str, Any]] = None if patch_payload.strip(): try: patch_payload_json = json.loads(patch_payload) except Exception as e: return f"Invalid JSON patch payload: {e}", "" out = _tmp_path("forked.zip") fork_patch_bundle( out, source_zip=source_bundle.name, fork_at_index=int(fork_index), patch_kind=patch_kind.strip() or None, patch_step=patch_step.strip() or None, patch_payload_json=patch_payload_json, ) return "✅ Fork bundle created (patched + re-hash-chained).", out with gr.Blocks(title=APP_TITLE) as demo: gr.Markdown(f"# {APP_TITLE}\nDiff two agent timelines. Find first divergence. Export forensic reports.") with gr.Tab("Diff two bundles"): with gr.Row(): bundle_a = gr.File(label="Bundle A (.zip)", file_types=[".zip"]) bundle_b = gr.File(label="Bundle B (.zip)", file_types=[".zip"]) run = gr.Button("Compute differential report", variant="primary") status = gr.Textbox(label="Status", interactive=False) summary = gr.Textbox(label="Summary (fast context)", interactive=False) alignment_table = gr.Dataframe( headers=["i", "status", "kind_a", "step_a", "kind_b", "step_b"], label="Timeline alignment (scan for first divergence)", interactive=False, wrap=True, ) diff_json = gr.JSON(label="Diff JSON (summary + per-event diffs)") with gr.Row(): report_zip = gr.File(label="Download drp-diff-report.zip (MD + PDF + JSON)") report_pdf = gr.File(label="Download report.pdf") report_md = gr.File(label="Download report.md") run.click( fn=ui_diff, inputs=[bundle_a, bundle_b], outputs=[status, summary, diff_json, alignment_table, report_zip, report_pdf, report_md], ) with gr.Tab("Verify bundle integrity"): bundle_v = gr.File(label="Bundle (.zip)", file_types=[".zip"]) verify_btn = gr.Button("Verify hash chain", variant="secondary") v_msg = gr.Textbox(label="Result", interactive=False) v_json = gr.JSON(label="Verification summary") verify_btn.click(fn=ui_verify, inputs=[bundle_v], outputs=[v_msg, v_json]) with gr.Tab("Generate demo bundles"): gr.Markdown("Use this to test the diff UI without integrating exporters yet.") with gr.Row(): seed_a = gr.Number(value=1, label="Seed A", precision=0) seed_b = gr.Number(value=2, label="Seed B", precision=0) chaos = gr.Slider(0, 1, value=0.35, step=0.01, label="Chaos (divergence likelihood)") gen_btn = gr.Button("Generate", variant="primary") gen_msg = gr.Textbox(label="Status", interactive=False) demo_a = gr.File(label="Demo Bundle A") demo_b = gr.File(label="Demo Bundle B") gen_btn.click(fn=ui_generate, inputs=[seed_a, seed_b, chaos], outputs=[gen_msg, demo_a, demo_b]) with gr.Tab("Fork / patch a bundle"): gr.Markdown("Counterfactual workflow: patch an event at index N, then re-hash-chain into a new bundle.") src = gr.File(label="Source bundle (.zip)", file_types=[".zip"]) fork_index = gr.Number(value=0, label="Fork at event index", precision=0) with gr.Row(): patch_kind = gr.Textbox(label="Patch kind (optional)", placeholder="e.g. tool_result") patch_step = gr.Textbox(label="Patch step (optional)", placeholder="e.g. t12.tool_result") patch_payload = gr.Textbox( label="Patch payload JSON (optional)", lines=8, placeholder='{"ok": true, "value": 42}', ) fork_btn = gr.Button("Create fork bundle", variant="primary") fork_msg = gr.Textbox(label="Status", interactive=False) fork_file = gr.File(label="Forked bundle (.zip)") fork_btn.click(fn=ui_fork, inputs=[src, fork_index, patch_kind, patch_step, patch_payload], outputs=[fork_msg, fork_file]) gr.Markdown( "Exporter tip: emit DRP events at LLM calls, tool calls/results, memory writes, planner steps, guardrails, and state snapshots." ) if __name__ == "__main__": demo.launch()