82 lines
3.5 KiB
Python
82 lines
3.5 KiB
Python
"""Shared orchestration for the contract reconstruction engine.
|
|
|
|
This module deliberately accepts the batch's persistence-backed extractor and
|
|
dossier builder as dependencies. Both the browser workflow and the JSON
|
|
integration boundary therefore execute the same business flow in-process,
|
|
while tests can replace either dependency without constructing HTTP calls or
|
|
making network requests.
|
|
"""
|
|
|
|
|
|
def analyze_batch(batch_id, documents, extractor, dossier_builder,
|
|
progress_callback=None, llm_client=None):
|
|
"""Extract document passports, assemble dossiers, and shape the API result.
|
|
|
|
`extractor` owns LLM interaction and persistence of each extraction;
|
|
`dossier_builder` consumes those saved passports and persists the resulting
|
|
graph. Keeping the calls in this order is an invariant: grouping must never
|
|
observe a partially extracted batch. The callback remains optional because
|
|
small batches run synchronously, while the existing large-batch route uses
|
|
it to update persistent progress.
|
|
"""
|
|
extraction = extractor(
|
|
batch_id,
|
|
llm_client=llm_client,
|
|
progress_callback=progress_callback,
|
|
documents=documents,
|
|
)
|
|
# Extractor-level `ok` means every input succeeded; dossier construction only
|
|
# requires at least one persisted passport and can proceed for that subset.
|
|
extraction_results = extraction.get("results") or []
|
|
total = extraction.get("total", len(documents))
|
|
done = extraction.get("done")
|
|
if done is None:
|
|
done = sum(1 for result in extraction_results if result.get("ok"))
|
|
failed = extraction.get("failed")
|
|
if failed is None:
|
|
failed = max(total - done, 0)
|
|
dossier_result = dossier_builder(batch_id) if done > 0 else {"ok": False}
|
|
dossier_ok = bool(dossier_result.get("ok"))
|
|
# Keep result usability separate from completeness so clients can continue
|
|
# with valid dossiers while still reporting failures explicitly.
|
|
status = (
|
|
"failed" if done == 0 or not dossier_ok
|
|
else "partial" if failed > 0
|
|
else "completed"
|
|
)
|
|
documents_by_id = {document.get("id"): document for document in documents}
|
|
failed_documents = []
|
|
for result in extraction_results:
|
|
if result.get("ok"):
|
|
continue
|
|
document_id = result.get("document_id")
|
|
source_document = documents_by_id.get(document_id, {})
|
|
failure = {
|
|
"document_id": document_id,
|
|
"filename": source_document.get("filename"),
|
|
"error": result.get("error") or "extraction failed",
|
|
}
|
|
if result.get("persistence_error"):
|
|
failure["persistence_error"] = result["persistence_error"]
|
|
failed_documents.append(failure)
|
|
classes = extraction.get("classes", {})
|
|
math_checks_passed = sum(
|
|
1 for result in extraction_results
|
|
if (result.get("extraction") or {}).get("math_validation", {}).get("status") == "VERIFIED"
|
|
)
|
|
return {
|
|
"ok": status != "failed",
|
|
"status": status,
|
|
"partial": status == "partial",
|
|
"v3": True,
|
|
"done": done,
|
|
"total": total,
|
|
"failed": failed,
|
|
"failed_documents": failed_documents,
|
|
"classes": classes,
|
|
"ignored_accounting": classes.get("ignored_accounting", 0),
|
|
"math_checks_passed": math_checks_passed,
|
|
"dossier_count": len(dossier_result.get("dossiers", [])),
|
|
"ambiguous": dossier_result.get("ambiguous", 0),
|
|
"unlinked": dossier_result.get("unlinked", []),
|
|
} |