lat-session-build

name	LAT Session Build
description	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.

LAT Session Build

Overview

Build a LAT parse session for a customer's laws that need LAT parsing. The key is filtering before building — parsing is expensive (legislation.gov.uk fetches, XML processing, fractalaw enrichment) so every record in the session should genuinely need work.

Pre-Build Filters

Apply these filters in order before creating session records. Each filter reduces the set.

1. Customer Applicability (required)

Only include laws the customer has marked as applicable.

SELECT oa.law_name
FROM org_applicabilities oa
WHERE oa.organization_id = '{customer_id}'
AND oa.status = 'yes'

2. Missing LAT (required)

Only include laws that don't already have LAT records.

AND (lr.lat_count IS NULL OR lr.lat_count = 0)

3. In-Force Status (required)

Exclude fully revoked/repealed laws. No point parsing a law that's been completely abolished.

AND (lr.live IS NULL OR lr.live NOT LIKE '❌%')

The three statuses:

• ✔ In force — parse it
• ⭕ Part Revocation / Repeal — parse it (still has active provisions)
• ❌ Revoked / Repealed / Abolished — skip it
• NULL — parse it (not yet enriched, assume in force)

4. Making Classification (optional)

Optionally filter to only making laws (laws that create duties). Non-making laws (housekeeping, empowering) don't produce obligations for the compliance register.

AND lr.is_making = true

Use with caution: some laws haven't been classified yet (is_making IS NULL). If you filter strictly on is_making = true you'll miss unclassified laws. Consider including NULLs:

AND (lr.is_making = true OR lr.is_making IS NULL)

Or skip this filter entirely and let the LAT parse + fractalaw triage determine making status.

Build Query

The full query combining all required filters:

SELECT oa.law_name, lr.title_en, lr.type_code, lr.live, lr.is_making
FROM org_applicabilities oa
JOIN legal_register lr ON lr.name = oa.law_name
WHERE oa.organization_id = '{customer_id}'
AND oa.status = 'yes'
AND (lr.lat_count IS NULL OR lr.lat_count = 0)
AND (lr.live IS NULL OR lr.live NOT LIKE '❌%')
ORDER BY lr.type_code, oa.law_name;

Pre-Build Summary

Before creating the session, show the user a summary so they can sanity-check:

-- Count by type_code
SELECT lr.type_code, COUNT(*)
FROM org_applicabilities oa
JOIN legal_register lr ON lr.name = oa.law_name
WHERE oa.organization_id = '{customer_id}'
AND oa.status = 'yes'
AND (lr.lat_count IS NULL OR lr.lat_count = 0)
AND (lr.live IS NULL OR lr.live NOT LIKE '❌%')
GROUP BY lr.type_code ORDER BY COUNT(*) DESC;

-- Count by live status
SELECT
  CASE
    WHEN lr.live LIKE '✔%' THEN 'In force'
    WHEN lr.live LIKE '⭕%' THEN 'Part revoked'
    WHEN lr.live IS NULL OR lr.live = '' THEN 'Unknown (not enriched)'
    ELSE lr.live
  END as status,
  COUNT(*)
FROM org_applicabilities oa
JOIN legal_register lr ON lr.name = oa.law_name
WHERE oa.organization_id = '{customer_id}'
AND oa.status = 'yes'
AND (lr.lat_count IS NULL OR lr.lat_count = 0)
AND (lr.live IS NULL OR lr.live NOT LIKE '❌%')
GROUP BY 1 ORDER BY 2 DESC;

-- Count what was filtered out
SELECT
  CASE
    WHEN lr.live LIKE '❌%' THEN 'Revoked (excluded)'
    WHEN lr.lat_count > 0 THEN 'Already has LAT (excluded)'
    ELSE 'Other'
  END as reason,
  COUNT(*)
FROM org_applicabilities oa
JOIN legal_register lr ON lr.name = oa.law_name
WHERE oa.organization_id = '{customer_id}'
AND oa.status = 'yes'
AND (lr.lat_count > 0 OR lr.live LIKE '❌%')
GROUP BY 1;

Session Creation

Use the lrt-create-session skill patterns. Key points:

• session_type = 'lat_parse'
• Don't put title_en in parsed_data (or do — the staged_parser now correctly overwrites with the canonical XML title if different)
• Session naming convention: lat-parse-{purpose}-{date}

session_id = 'lat-parse-qq-missing-YYYY-MM-DD'

# Create session
psql(f"""
INSERT INTO scrape_sessions (session_id, year, month, day_from, day_to, status, session_type)
VALUES ('{session_id}', {year}, {month}, {day}, {day}, 'pending', 'lat_parse')
""")

# Insert filtered records
for name, title in filtered_laws:
    parsed = json.dumps({"name": name, "title": title})
    psql(f"""
    INSERT INTO scrape_session_records (id, session_id, law_name, "group", status, selected, parsed_data, inserted_at, updated_at)
    VALUES ('{uuid4()}', '{session_id}', $n${name}$n$, 'group1', 'pending', false,
            $j${parsed}$j$::jsonb, NOW(), NOW())
    """)

# Finalize
psql(f"""
UPDATE scrape_sessions SET
  status = 'categorized',
  group1_count = (SELECT COUNT(*) FROM scrape_session_records WHERE session_id = '{session_id}' AND "group" = 'group1')
WHERE session_id = '{session_id}'
""")

Common Mistakes

Mistake	Consequence	Prevention
Not filtering revoked laws	Wasted parse effort on dead legislation	Always apply live NOT LIKE '❌%' filter
Strict is_making = true filter	Misses unclassified laws (NULL)	Include NULLs or skip the filter
Not checking existing LAT	Re-parsing laws that already have provisions	Always filter lat_count = 0 OR NULL
Including status = 'no' laws	Parsing laws the customer marked not applicable	Always filter oa.status = 'yes'
Large sessions without summary	User can't sanity-check before committing	Always show type_code + live breakdown first

Customer IDs

Look up customer UUIDs via the organizations table:

SELECT id, name FROM organizations;

Customer	UUID
QQ	c075d56b-8420-4408-b695-ccfbc1ba15ec
Demo	aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee

Related Skills

• lrt-create-session — session creation mechanics (schema constraints, UI requirements)
• lat-parse-session — the human-AI parse workflow after the session is built
• db-schema-changes — if session schema needs modification