sertantai-legal

Build a LAT parse session from a customer's applicable laws. Applies all necessary filters (in-force, not-revoked, missing LAT, making classification) before creating the session to avoid wasting parse effort on laws that don't need processing.

Safe procedures for modifying the database schema. Covers the uk_lrt view + INSTEAD OF trigger pattern, column type changes on partitioned tables, and migration best practices.

Create a scrape session programmatically from data (CSV, list of law names, query results). Handles session + record creation with correct schema constraints so records appear in the admin UI.

Sync a customer's legal register, duties, and actor tuples to Baserow. Covers the three-table model, significance filters, sync config, and common issues.

Run `mix customer.pipeline_status` to report a customer's law corpus readiness — from applicability through LAT parsing, fractalaw enrichment, to Baserow-ready governed duties.

Generate a data quality report comparing a customer's legacy vendor register (Enhesa, Nimonik, etc.) against SertantAI's enriched legal database. Produces confusion matrix, revoked law analysis, coverage gaps, duty density, and family distribution. Outputs .md report + .csv appendices. Designed for customer senior management.

Import a customer's legacy legal register CSV (Enhesa, Nimonik, etc.), transform and match against LRT, group results into matched/scrapeable/not-handled, and create a scrape session for missing laws. First phase of customer onboarding.

Human-AI partnered workflow for LAT (Legal Article Text) parsing sessions. Guides making-classification review, monitors parse progress, runs post-parse QA (LAT shape, hierarchy integrity, annotation sanity), triggers taxa enrichment via Zenoh, then promotes data through NAS sync and production sync with QA gates at each stage.

QA legacy vendor applicability data (Enhesa Yes/No) against SertantAI L2. Per-family analysis comparing vendor selections to LRT making laws, identifying revoked errors, coverage gaps, and Making classification issues. Produces actionable findings and patterns for automated screening.

Human-AI partnered workflow for monthly LRT scrape sessions. Guides scope definition, monitors scrape progress, runs post-scrape QA (data completeness, family sense-check, relationship integrity), then promotes data through NAS sync and production sync with QA gates at each stage.

Managing database snapshots on the office NAS (UGREEN DXP2800). Export, import, verify, and troubleshoot dev database snapshots stored on the LAN NAS via SMB.

Syncing dev database changes to production. Covers incremental delta export/import, bulk pg_restore for empty tables, and the SSH pipeline to reach prod PostgreSQL inside Docker.

Deploy sertantai-legal to production on Hetzner. Build Docker images, push to GHCR, deploy via SSH, and manage ElectricSQL/Nginx/PostgreSQL infrastructure.

AI-driven family classification QA using amends graph relationships. Analyses laws where amendment targets disagree on family, presents batch recommendations, applies confirmed changes, and rebuilds edges.

AI-driven family classification QA using enacted_by graph relationships. Analyses a family population for suspect assignments, presents batch recommendations, applies confirmed changes, and rebuilds edges.

AI-driven QA of enacted_by parser results. Fetches introduction XML from legislation.gov.uk, verifies parsed parent law links, classifies errors, and identifies parser improvement opportunities.

Safe procedures for restarting Docker development services (PostgreSQL, ElectricSQL) after shutdown or reboot. Prevents accidental database data loss.

How stale/broken ElectricSQL shapes are detected and recovered after Electric restarts. Covers the root cause, client-side fix, and server-side prerequisites.

How sertantai-legal publishes LRT, LAT, and AmendmentAnnotation tables to fractalaw over Zenoh P2P mesh. Covers architecture, configuration, adding new queryables, and troubleshooting.

Using the GET /api/ai/sync/lat and /api/ai/sync/annotations endpoints for incremental pull-based sync to the AI service.

Using the GET /api/ai/drrp/clause/queue endpoint to pull DRRP entries needing AI clause refinement.