[witbrock]

Outcome

Fix the live JVNAUTOSCI-2592 failure where a model-valid reuse_existing plan referenced deterministic future concept IDs that did not yet exist, causing every first-write parent relationship to fail with source_concept_not_found.

Change

• Preview the deterministic spreadsheet KR guard and collect only its authorised stable concept IDs.
• Resolve all candidates in one access-controlled, projection-only, indexed and 3-second-bounded canonical read.
• Bind each slot before model planning:
  • absent ID -> create only;
  • accessible existing ID -> reuse_existing only.
• Fail closed with a typed aggregate error if the canonical store/read is unavailable.
• Preserve the generic model-driven KR plan; the existing guard rejects any model decision that conflicts with canonical existence.
• Reuse the existing Atlas-efficient concept-existence helper rather than introducing per-concept reads.

Authority and privacy

This is a deterministic idempotency/write boundary. Workbook contents remain untrusted evidence; actor/org access filtering remains applied by the repository path. No workbook names, paths, hashes, cell values, person names, or private operational IDs are logged by the added checks.

Validation

• 114 focused spreadsheet, KR workflow, control-flow and metadata tests pass.
• Ruff and diff checks pass.
• Seed-version guard passes; no represented workflow authority changed.
• A privacy-safe read-only probe against the actual failed record now yields 10 create-only slots, 0 reuse-only slots and 0 ambiguous decisions.
• Independent review found no blocking issue; the unrelated untracked uv.lock is excluded.

Jira: JVNAUTOSCI-2592

[Copilot]

Pull request overview

This PR fixes a spreadsheet-record ingestion failure mode where reuse_existing could be planned for deterministic concept IDs that don’t actually exist yet, by binding each concept slot’s allowed decision to canonical concept existence before model planning/materialisation.

Changes:

• Add a bounded, access-controlled canonical existence lookup to resolve which deterministic concept IDs are actually present and reusable.
• Bind spreadsheet KR materialisation guard slots to create-only vs reuse_existing-only based on that existence read-back.
• Add focused tests covering the new binding behaviour, guard enforcement, and fail-closed handling when the concept store/read is unavailable.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/backend/test_spreadsheet_record_ingestion_service.py	Adds an end-to-end test asserting slot decisions are bound to canonical existence (create-only vs reuse-only).
tests/backend/test_source_processing_marker_service.py	Adds a regression test ensuring existence reads fail closed when the concept store is unavailable.
tests/backend/test_kr_materialisation_guard_service.py	Adds tests asserting the guard rejects plans that violate the bound decisions.
tests/backend/test_internal_mcp_spreadsheet_record_tools.py	Adds tests for single bounded existence read behaviour and fail-closed MCP tool responses.
src/backend/services/spreadsheet_record_ingestion_service.py	Threads reusable existing concept IDs into request construction and clarifies prompt authority language.
src/backend/services/spreadsheet_materialisation_guard_service.py	Implements slot decision binding based on reusable existing concept IDs.
src/backend/services/source_processing_marker_service.py	Introduces a public helper for bounded, access-controlled concept existence resolution with fail-closed semantics.
src/backend/integrations/internal_mcp/spreadsheet_record_tools.py	Binds reuse authority by previewing the guard, resolving existing IDs in one read, then building the bound request.