Quick-reference document for resuming work after context compaction.
Contains: project overview, current state, spec summary, phases,
key decisions, file locations, and resumption instructions.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add chk_dedup_scoped constraint enforcing tenant-scoped dedup format
- Filter location_type='owned' in populate_facts() for 'ALL' rollup
- Document competitor exclusion from 'ALL' sentinel rollups
- Add explicit comments in aggregation code for maintainability
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Two micro-risk mitigations documented:
1. dedup_group_id: Format "{business_id}:{hash}" to prevent
cross-tenant collision on similar reviews.
2. Sentinel conventions: 'ALL' (spatial) vs 'all' (semantic).
Case matters — do not normalize.
Spec frozen as v3.1.2.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Final fixes for production-ready spec:
1. locations.location_type: Added 'owned'|'competitor' flag.
Competitors now inserted into locations (preserves FK integrity).
2. Competitor fact query: Added business_id filter to prevent
cross-tenant contamination when same competitor tracked by
multiple customers.
3. issue_events versioning: Added source + review_version columns
for complete review reference in audit log.
4. Enrichment tenant-scoping: business_id now passed from ingest
job (not looked up). Validates place_id exists under tenant.
5. Footer: Fixed version string v3.1.1 → v3.1.2.
Status: Ship-ready specification.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Three final fixes applied:
1. issue_spans versioning: Added source + review_version columns
with FK to reviews_enriched(source, review_id, review_version).
Spans now correctly reference the exact review version.
2. Competitor business_id rule: Clarified that competitor reviews
use customer's business_id + competitor's place_id (not NULL).
Keeps facts and joins working without special-case logic.
3. Trust-weighted facts: Clarified trust_weighted_* columns are
reserved but not populated in v3.1. Trust scoring applies to
issue priority only. Aggregation deferred to v3.2.
Status: Production-grade architecture specification.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Complete pipeline architecture for Google Reviews intelligence:
- Versioned reviews_enriched with (source, review_id, version) PK
- Tenant-scoped locations with (business_id, place_id) PK
- Relational issue_spans replacing array aggregation
- Unified fact_timeseries spine with 'ALL' sentinel for rollups
- Clean competitor model (separate table, no fake business_ids)
- Trust scoring and dedup support
- KPI-ready join keys
Reviewed and fixed: PK for edited reviews, multi-tenant overlap,
param ordering bugs, fact population scope, entity field deferral.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- reviewiq-pipeline-v1-final.md: Earlier pipeline specification
- test_metadata_extraction.py: Test script for metadata extraction
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
