Phase 4: cross-format Compare orchestration

application/single_app/config.py

@@ -97,7 +97,7 @@
 EXECUTOR_TYPE = 'thread'
 EXECUTOR_MAX_WORKERS = 30
 SESSION_TYPE = 'filesystem'
-VERSION = "0.250.066"
+VERSION = "0.250.067"
 IS_DEVELOPMENT = is_development_env_enabled()
 
 SESSION_COOKIE_SAMESITE = os.getenv('SESSION_COOKIE_SAMESITE', 'Lax')

application/single_app/functions_document_comparison.py

@@ -224,6 +224,117 @@ def _format_comparison_coverage_summary(coverage, left_document_name, right_docu
  return '\n'.join(lines)
 
 
+def run_evidence_document_comparison(
+ comparison_prompt,
+ left_source,
+ right_sources,
+ invoke_prompt,
+ activity_callback=None,
+):
+ """Run the established one-left-to-many comparison over native evidence envelopes."""
+ normalized_prompt = str(comparison_prompt or '').strip()
+ if not normalized_prompt or not callable(invoke_prompt):
+ raise ValueError('A comparison prompt and callable invoke_prompt handler are required.')
+
+ left_source = left_source if isinstance(left_source, dict) else {}
+ right_sources = [source for source in list(right_sources or []) if isinstance(source, dict)]
+ left_name = str(left_source.get('document_name') or 'Source').strip() or 'Source'
+ left_summary = str(left_source.get('summary') or '').strip()
+ comparison_items = []
+ compared_targets = []
+ failed_targets = []
+
+ for comparison_index, right_source in enumerate(right_sources, start=1):
+ right_name = str(right_source.get('document_name') or f'Target {comparison_index}').strip() or f'Target {comparison_index}'
+ right_status = str(right_source.get('status') or '').strip().lower()
+ if right_status not in {'completed', 'partial'} or not left_summary:
+ failed_targets.append(right_name)
+ continue
+
+ if callable(activity_callback):
+ activity_callback({
+ 'type': 'comparison_started',
+ 'left_document_id': left_source.get('document_id'),
+ 'left_document_name': left_name,
+ 'right_document_id': right_source.get('document_id'),
+ 'right_document_name': right_name,
+ 'comparison_index': comparison_index,
+ 'comparison_count': len(right_sources),
+ })
+ pairwise_text = str(invoke_prompt(
+ _build_pairwise_comparison_prompt(
+ normalized_prompt,
+ left_name,
+ right_name,
+ left_summary,
+ str(right_source.get('summary') or ''),
+ ),
+ stage='comparison',
+ metadata={
+ 'comparison_index': comparison_index,
+ 'comparison_count': len(right_sources),
+ 'left_document_id': left_source.get('document_id'),
+ 'right_document_id': right_source.get('document_id'),
+ },
+ ) or '').strip()
+ if not pairwise_text:
+ failed_targets.append(right_name)
+ continue
+ compared_targets.append(right_name)
+ comparison_items.append({
+ 'right_document_id': right_source.get('document_id'),
+ 'right_document_name': right_name,
+ 'text': pairwise_text,
+ })
+ if callable(activity_callback):
+ activity_callback({
+ 'type': 'comparison_completed',
+ 'left_document_id': left_source.get('document_id'),
+ 'left_document_name': left_name,
+ 'right_document_id': right_source.get('document_id'),
+ 'right_document_name': right_name,
+ 'comparison_index': comparison_index,
+ 'comparison_count': len(right_sources),
+ })
+
+ if not comparison_items:
+ final_reply = 'No target comparison could be completed from the available source evidence.'
+ elif len(comparison_items) == 1:
+ final_reply = comparison_items[0]['text']
+ else:
+ final_reply = str(invoke_prompt(
+ _build_comparison_reduction_prompt(normalized_prompt, left_name, comparison_items),
+ stage='comparison_reduction',
+ metadata={'comparison_count': len(comparison_items), 'left_document_id': left_source.get('document_id')},
+ ) or '').strip() or 'The completed target comparisons could not be reduced into a final response.'
+
+ evidence_engines = sorted({
+ str(source.get('engine') or 'unknown')
+ for source in [left_source, *right_sources]
+ })
+ conclusion_level = 'aggregate or narrative'
+ if all(str(source.get('source_kind') or '') == 'tabular' for source in [left_source, *right_sources]):
+ conclusion_level = 'aggregate; row-level conclusions require validated structured table operations'
+ coverage_note = (
+ f'\n\n## Comparison Coverage\n- Targets compared: {len(compared_targets)}\n'
+ f'- Failed or partial targets: {len(failed_targets)}\n'
+ f'- Evidence engines: {", ".join(evidence_engines)}\n'
+ f'- Conclusion level: {conclusion_level}'
+ )
+ return {
+ 'reply': f'{final_reply}{coverage_note}',
+ 'analysis_reply': final_reply,
+ 'coverage': {'document_count': 1 + len(right_sources), 'partial_coverage': bool(failed_targets), 'failed_targets': failed_targets},
+ 'documents': [left_source, *right_sources],
+ 'left_document': {'document_id': left_source.get('document_id'), 'document_name': left_name},
+ 'right_documents': [
+ {'document_id': source.get('document_id'), 'document_name': source.get('document_name')}
+ for source in right_sources
+ ],
+ 'comparison_items': comparison_items,
+ }
+
+
 def run_document_comparison(
  user_id,
  comparison_prompt,

application/single_app/functions_settings.py

@@ -233,6 +233,19 @@ def is_mixed_source_analyze_all_enabled(settings):
  )
 
 
+def is_cross_format_compare_enabled(settings):
+ """Return whether Phase 4 native mixed-source Compare behavior is enabled."""
+ return bool((settings or {}).get('enable_cross_format_compare', False))
+
+
+def is_cross_format_compare_one_to_many_enabled(settings):
+ """Return whether the separately staged one-to-many mixed-target rollout is enabled."""
+ return bool(
+ (settings or {}).get('enable_cross_format_compare', False)
+ and (settings or {}).get('enable_cross_format_compare_one_to_many', False)
+ )
+
+
 def is_mixed_source_relevance_candidates_enabled(settings):
  """Return whether Phase 2 relevance-mode table candidates are enabled."""
  return bool(
@@ -795,6 +808,8 @@ def get_settings(use_cosmos=False, include_source=False):
  'enable_mixed_source_relevance_candidates': False,
  'enable_mixed_source_analyze': False,
  'enable_mixed_source_analyze_all': False,
+ 'enable_cross_format_compare': False,
+ 'enable_cross_format_compare_one_to_many': False,
  'max_rounds_per_agent': 1,
  'workflow_max_auto_invoke_attempts': 60,
  'enable_semantic_kernel': False,