Skip to content

Relationship review evidence — 4 October 2026 ​

This is a bounded preflight for the remaining browsing choice in the relationship review, not a complete schema or deployed-system audit.

Sources recovered ​

SubjectSourceUse and limit
Product source_data/review/content-mechanism.json, P-09; decisions/family-reconciliation.mdHistorical question preserved. Its example does not establish inside-app collection.
Whole web page_data/review/content-retail.json, R-04Historical surface-only recommendation conflicts with the recorded content rule that a landing page is a piece. Type-versus-page-purpose wording remains separate.
Later approvalsdecisions/decided.md, document parts and photographs (3 October), strategy recovery, useful differences and versions (4 October)Carry forward the meanings; no implementation certification.
Current guidancepillars/products.md, pillars/groupings.mdProduct-subject and capture-source descriptions need reconciliation after the residual choice.
Original answersdocs/reference/2026-10/2026-10-03/sitting-14-joe-response.txt; docs/reference/2026-10/2026-10-04/sitting-10-joe-response.txtExact answer sources remain in the existing answer bank.

System evidence ​

The literal selected SQL results are saved in docs/reference/2026-10/2026-10-04/relationship-boundaries-live-evidence.json. They inspect content_products roles/provenance, selected links and products columns, and three content_item records. No broad absence claim follows from these selections.

The Operator record has a TikTok source and separate product associations. The two Claude on Mars records have fullshot/hero source suffixes and stored page/type labels. A null analysis_source_id in those selected records does not prove that no relationship exists elsewhere. Stored classification is evidence of storage, not proof of correct meaning.

Local code inspected: backend/services/post_processor.py product-resolution and write sections; backend/services/classification_schema.py; frontend/src/lib/build-content-query.ts product-category filter; frontend/src/components/detail/ContentIntelCard.tsx product display. The detail component retains role information. Do not generalize the category-query finding to all displays.

The writer can derive a link's product association from the analyzer's primary product and uses a shared content/product conflict key for associations. The category filter traverses product associations without a role-specific condition in the inspected branch. Before changing either, trace original link assignment, downstream reads, corrections, multiplicity, uniqueness and reanalysis. No deployed writer comparison or full consumer proof was completed in this batch.

The Anthropic source article was read. The TikTok source URL was checked in the live record; the video was not watched, and no motion classification is proposed here. The full and hero capture images are not used as newly inspected visual proof in this review.

Follow-through boundary ​

After the browsing answer, update Content, Products and Groupings together where their descriptions change and bank its exact scope. Record the recovered whole-page rule without requesting a new answer; keep Product Detail Page type-versus-purpose wording partial. The comprehensive pass owns storage design, live constraint checks, writer/reader compatibility, manual-correction preservation and repair/rollback contracts. Publication of this review authorizes none of those mutations.

The BrandTrackers domain model. Source: git markdown, drift-checked against the live DB.