Compare variants
Two design variants of Enterprise Document Processing, delivering the same unit event by different designs. The comparison states what differs; it does not rank the designs.
Common unit event
Business Document
BUSINESS_DOCUMENTOne completed execution of the document-processing workflow for one inbound business document.
Left
CompleteHigh Accuracy Processing
A reasoning document-understanding pass, classification, two extraction passes, enterprise-knowledge grounding with reranking, and a grounded final generation. The same business unit event as Standard Processing, implemented differently.
Direct cost per business document
Reference
Open scenario
dv-high-accuracy-processingRight
CompleteStandard Processing
A single document-understanding pass, classification, structured extraction and one final generation pass. No enterprise retrieval grounding.
Direct cost per business document
Reference
Open scenario
dv-standard-processingNo difference figure is shown
A difference between two prices is itself a price, and this surface derives none. The read model carries each variant’s own reference result and no comparison of them, so both figures are shown as they were produced and are read against each other by eye.
Shared assumptions
Inputs both variants carry with the same meaning. Everything compared below stands on these; where a shared input holds a different value in each variant, the row says so.
40 shared
| Input | High Accuracy Processing | Standard Processing |
|---|---|---|
Cached Tokensau-classify-document.cached_tokens | 0TOKEN | 0TOKEN |
Classify Document — offeringau-classify-document.commercial_offering | ctxdex:offering:agent-makers:ctxdex-intelligence-4-fast | ctxdex:offering:agent-makers:ctxdex-intelligence-4-fast |
Document Summary TokensDiffersau-classify-document.document_summary_tokens | 1400TOKEN | 700TOKEN |
Expected Classification Tokensau-classify-document.expected_classification_tokens | 80TOKEN | 80TOKEN |
Classify Document — modeau-classify-document.implementation_mode | im-classify-document-model | im-classify-document-model |
Classify Document — occurrencesau-classify-document.occurrence_count | 1UNIT | 1UNIT |
System Instruction Tokensau-classify-document.system_instruction_tokens | 900TOKEN | 900TOKEN |
Cached Tokensau-extract-fields.cached_tokens | 0TOKEN | 0TOKEN |
Class Hint Tokensau-extract-fields.class_hint_tokens | 400TOKEN | 400TOKEN |
Extract Structured Fields — offeringau-extract-fields.commercial_offering | ctxdex:offering:agent-makers:ctxdex-intelligence-4-fast | ctxdex:offering:agent-makers:ctxdex-intelligence-4-fast |
Document Text Tokensau-extract-fields.document_text_tokens | 9000TOKEN | 9000TOKEN |
Expected Extraction Tokensau-extract-fields.expected_extraction_tokens | 500TOKEN | 500TOKEN |
Extracted Field Countau-extract-fields.extracted_field_count | 18UNIT | 18UNIT |
Extract Structured Fields — modeau-extract-fields.implementation_mode | im-extract-fields-model | im-extract-fields-model |
Extract Structured Fields — occurrencesDiffersau-extract-fields.occurrence_count | 2UNIT | 1UNIT |
System Instruction Tokensau-extract-fields.system_instruction_tokens | 600TOKEN | 600TOKEN |
Cached Tokensau-generate-structured-result.cached_tokens | 0TOKEN | 0TOKEN |
Generate Structured Result — offeringDiffersau-generate-structured-result.commercial_offering | ctxdex:offering:agent-makers:ctxdex-intelligence-4-pro | ctxdex:offering:agent-makers:ctxdex-intelligence-4-fast |
Document Summary TokensDiffersau-generate-structured-result.document_summary_tokens | 1400TOKEN | 700TOKEN |
Expected Result TokensDiffersau-generate-structured-result.expected_result_tokens | 900TOKEN | 600TOKEN |
Generate Structured Result — modeau-generate-structured-result.implementation_mode | im-generate-structured-result-model | im-generate-structured-result-model |
Normalized Fields Tokensau-generate-structured-result.normalized_fields_tokens | 450TOKEN | 450TOKEN |
Generate Structured Result — occurrencesau-generate-structured-result.occurrence_count | 1UNIT | 1UNIT |
Retrieved Context TokensDiffersau-generate-structured-result.retrieved_context_tokens | 4000TOKEN | 0TOKEN |
System Instruction Tokensau-generate-structured-result.system_instruction_tokens | 700TOKEN | 700TOKEN |
Ingest and Parse Document — offeringau-ingest-document.commercial_offering | ctxdex:offering:agent-makers:ctxdex-enterprise-document-1 | ctxdex:offering:agent-makers:ctxdex-enterprise-document-1 |
Document Pagesau-ingest-document.document_pages | 12PAGE | 12PAGE |
Extracted Text Tokensau-ingest-document.extracted_text_tokens | 9000TOKEN | 9000TOKEN |
Ingest and Parse Document — modeau-ingest-document.implementation_mode | im-ingest-document-service | im-ingest-document-service |
Ingest and Parse Document — occurrencesau-ingest-document.occurrence_count | 1UNIT | 1UNIT |
Normalize Extracted Result — modeau-normalize-extracted-result.implementation_mode | im-normalize-extracted-deterministic | im-normalize-extracted-deterministic |
Normalized Fields Tokensau-normalize-extracted-result.normalized_fields_tokens | 450TOKEN | 450TOKEN |
Normalize Extracted Result — occurrencesau-normalize-extracted-result.occurrence_count | 1UNIT | 1UNIT |
Cached Tokensau-understand-document.cached_tokens | 0TOKEN | 0TOKEN |
Understand Document — offeringDiffersau-understand-document.commercial_offering | ctxdex:offering:agent-makers:ctxdex-intelligence-4-pro | ctxdex:offering:agent-makers:ctxdex-intelligence-4-fast |
Document Summary TokensDiffersau-understand-document.document_summary_tokens | 1400TOKEN | 700TOKEN |
Document Text Tokensau-understand-document.document_text_tokens | 9000TOKEN | 9000TOKEN |
Understand Document — modeDiffersau-understand-document.implementation_mode | im-understand-document-reasoning | im-understand-document-model |
Understand Document — occurrencesau-understand-document.occurrence_count | 1UNIT | 1UNIT |
System Instruction Tokensau-understand-document.system_instruction_tokens | 600TOKEN | 600TOKEN |
Only in High Accuracy Processing
- Assemble Grounded Context — mode
au-assemble-grounded-context.implementation_modeim-assemble-context-deterministic - Assemble Grounded Context — occurrences
au-assemble-grounded-context.occurrence_count1UNIT - Retrieved Context Tokens
au-assemble-grounded-context.retrieved_context_tokens4000TOKEN - Cached Tokens
au-embed-query.cached_tokens0TOKEN - Embed Query — offering
au-embed-query.commercial_offeringctxdex:offering:agent-makers:ctxdex-retrieval-3-embed - Embed Query — mode
au-embed-query.implementation_modeim-embed-model - Embed Query — occurrences
au-embed-query.occurrence_count1UNIT - Retrieval Query Tokens
au-embed-query.retrieval_query_tokens200TOKEN - Prepare Document Retrieval Query — mode
au-prepare-document-retrieval-query.implementation_modeim-prepare-document-query-deterministic - Prepare Document Retrieval Query — occurrences
au-prepare-document-retrieval-query.occurrence_count1UNIT - Retrieval Query Tokens
au-prepare-document-retrieval-query.retrieval_query_tokens200TOKEN - Candidate Document Count
au-rerank-candidates.candidate_document_count20UNIT - Candidate Document Tokens
au-rerank-candidates.candidate_document_tokens16000TOKEN - Rerank Candidates — offering
au-rerank-candidates.commercial_offeringctxdex:offering:agent-makers:ctxdex-retrieval-2-rerank - Rerank Candidates — mode
au-rerank-candidates.implementation_modeim-rerank-model - Rerank Candidates — occurrences
au-rerank-candidates.occurrence_count1UNIT - Query Count
au-rerank-candidates.query_count1UNIT - Request Output Tokens
au-rerank-candidates.request_output_tokens0TOKEN - Retrieval Query Tokens
au-rerank-candidates.retrieval_query_tokens200TOKEN - Candidate Document Count
au-search-enterprise-knowledge.candidate_document_count20UNIT - Candidate Document Tokens
au-search-enterprise-knowledge.candidate_document_tokens16000TOKEN - Search Enterprise Knowledge — mode
au-search-enterprise-knowledge.implementation_modeim-search-local-unpriced - Search Enterprise Knowledge — occurrences
au-search-enterprise-knowledge.occurrence_count1UNIT
Only in Standard Processing
No input unique to this variantDesign differences
What the two designs do differently, before any money is involved.
Extract Structured Fields
How many times it runs
au-extract-fieldsUnderstand Document
Implementation mode
au-understand-documentOperations
Aligned on canonical identity, never on a display name.
In both4
- Classify Document
op-classify-document - Extract Structured Information
op-extract-structured-information - Ingest & Understand Document
op-ingest-understand-document - Produce Final Result
op-produce-final-result
Only in High Accuracy Processing1
- Retrieve Supporting Knowledge
op-retrieve-supporting-knowledge
Only in Standard Processing0
None
Activities
Aligned on canonical identity, never on a display name.
In both6
- Classify Document
au-classify-document - Extract Structured Fields
au-extract-fields - Generate Structured Result
au-generate-structured-result - Ingest and Parse Document
au-ingest-document - Normalize Extracted Result
au-normalize-extracted-result - Understand Document
au-understand-document
Only in High Accuracy Processing5
- Assemble Grounded Context
au-assemble-grounded-context - Embed Query
au-embed-query - Prepare Document Retrieval Query
au-prepare-document-retrieval-query - Rerank Candidates
au-rerank-candidates - Search Enterprise Knowledge
au-search-enterprise-knowledge
Only in Standard Processing0
None