Path 3 — Jurisdiction tuning checklist¶

Every government deployment goes through this one-week tuning pass before the workbench or API enters use. It is the difference between a generic tool and one your agency can defend in front of a court, a legislator, and a journalist.

Run by: 1 senior inspector + 1 staff lawyer + 1 DueCare engineer. Duration: 5 working days. Output: a duecare-jurisdiction-<agency>-config.yaml file + an updated knowledge pack, both checked into your agency's internal repo.

Day 1 — citation matrix¶

Goal: every statute, ILO convention article, and case-law reference that DueCare returns in your jurisdiction has been verified by your staff lawyer against the authoritative register.

Inputs: - Bundled knowledge pack (451 GREP rules + 859 RAG documents + 15 ILO conventions + 36 templates) - Your agency's statute register (whatever the official source is — national legal database, regulator publications site, etc.)

Process: 1. DueCare engineer exports the citations applicable to your jurisdiction (duecare knowledge export --citations --jurisdiction <code>). 2. Staff lawyer walks the list. For each citation: - Verified: statute exists, current, says what the entry claims it says. Mark verified: true + verified_at: <date>. - Stale: statute exists but has been amended. Provide the current version; DueCare engineer updates the entry. - Repealed: statute no longer in force. Mark superseded_by: <new statute>; entry is removed or rewritten. - Mis-cited: statute exists but says something different. Provide the verbatim correct text; entry is rewritten. 3. Output: an annotated citation list checked into the agency repo.

This is the single most important day of the tuning. If your staff lawyer cannot defend every citation the workbench produces, the workbench is not ready for your agency.

Day 2 — cluster weighting¶

Goal: the rubric the workbench uses internally matches your agency's enforcement priorities.

The bundled rubric has six clusters with default thresholds:

Cluster	Default threshold	When to raise it	When to lower it
Legal grounding + precision	60%	Regulator focused on legal-grounds enforcement	Frontline complaint-intake where speed matters more than precision
Harm avoidance + refusal	80%	Vulnerable-worker desk; trauma-survivor cases	Reduce only with extreme caution and a written justification
Worker protection + priority	70%	Migrant-worker rights inspectorate	When the agency's primary mandate is something else (audit, license, criminal investigation)
Pattern + risk recognition	60%	Investigative unit looking for emerging patterns	Routine compliance enforcement
Resources + actionability	70%	Walk-in complaint desk; community-facing surface	Internal investigation (where resources are the agency's own)
Ethical framing	70%	Public-communications-adjacent role	Internal-only

Process: 1. DueCare engineer presents the default rubric and 5 to 10 sample case outputs. 2. Senior inspector reviews: does the rubric weight the right things for THIS agency's mandate? 3. Adjust thresholds; document each change with a one-line reason. 4. Output: cluster_weights section in the jurisdiction config.

Example YAML:

agency: dmw-ph
jurisdiction_code: dmw-ph
locales: [en-PH, tl-PH]
cluster_weights:
  legal_grounding_precision:
    threshold: 0.70  # raised from default 0.60: regulator focus
    rationale: "DMW is a regulator; legal precision is a primary mandate."
  harm_avoidance_refusal:
    threshold: 0.85  # raised slightly: vulnerable-worker case base
    rationale: "Large proportion of DMW complaints involve worker safety."
  worker_protection_priority:
    threshold: 0.75  # raised from 0.70
    rationale: "Worker protection is a primary mandate per RA 8042 as amended."
  pattern_risk_recognition:
    threshold: 0.60  # unchanged
    rationale: ""
  resources_actionability:
    threshold: 0.70  # unchanged
    rationale: ""
  ethical_framing:
    threshold: 0.70  # unchanged
    rationale: ""

Day 3 — agency-specific templates¶

Goal: the 34 bundled templates cover the major regulators globally, but your agency may have enforcement tools that need their own templates.

Common additions:

Template	When you need it
Administrative penalty notice	Any agency with administrative-fine authority
License suspension / revocation order	Recruitment-agency licensing regulators
Show-cause notice	Most regulators with hearing procedures
Referral to prosecutor	When the agency identifies criminal-level violations
Joint operation request	Cross-agency enforcement
Worker-side advisory	Inspector-mediated communication back to the worker
Site inspection report	Field inspectors
Provincial / regional summary report	For agencies with regional offices

Process: 1. Senior inspector lists the agency's enforcement-instrument inventory. 2. For each item not covered by a bundled template, DueCare engineer produces a draft using the agency's existing template if available (Word doc, PDF) or the agency's regulation text. 3. Staff lawyer reviews each new template for legal sufficiency. 4. Output: agency-specific templates checked into the jurisdiction config under templates_additional.

Day 4 — audit + privacy configuration¶

Goal: the workbench's FOIA audit log and PII handling match your agency's data-governance policies.

Decisions:

Decision	Default	Common variant
Audit retention	5 years	Per agency record-retention policy
FOIA export format	Redacted JSON + summary CSV	Add agency-specific cover sheet
Inspector ID format	`inspector-<uuid>`	Agency employee ID (with separate access control on the audit store)
Query hash algorithm	SHA-256	Some agencies require SHA-512
PII redaction targets	Names, passport, phone, address, DOB	Add agency-specific (case number, license number)
Encryption at rest	Per device disk encryption	Agency may require additional layer
Encryption in transit	TLS 1.2+	Same
Bearer token rotation	90 days	Per agency security policy
API egress	None	Same — no exceptions

Process: 1. DueCare engineer presents the audit configuration choices. 2. Agency IT / data governance officer reviews and decides. 3. Configuration written into the jurisdiction config.

Day 5 — inspector training + shadow run¶

Goal: at least 2 inspectors can use the workbench / API independently and have walked through real (mock) cases.

Schedule:

Block	Activity
Morning, hour 1	Workbench tour: pages, personas, activity log, templates, knowledge lookup
Morning, hour 2	Sample case 1: clear-cut trafficking indicator. Inspector drives; engineer observes.
Morning, hour 3	Sample case 2: ambiguous case. Inspector decides what to flag.
Afternoon, hour 1	Sample case 3: contract substitution. Inspector drafts a notice using a template.
Afternoon, hour 2	Sample case 4: a normal worker FAQ that should NOT be flagged. Verify the workbench handles it gracefully.
Afternoon, hour 3	Audit log review: inspect what was logged for each case. Verify chain integrity.

At end of day 5: senior inspector signs off on production readiness. Pilot can begin.

Output artefact¶

The full tuning produces one YAML file per agency:

schema: duecare-jurisdiction-config/v1
agency: dmw-ph
jurisdiction_code: dmw-ph
agency_locale: en-PH
agency_languages: [en, tl]
duecare_version: "0.1.0"
knowledge_pack_version: "2026-05-22-jurisdiction-dmw"
created_at: "2026-05-22"
created_by: ["sr-inspector-id-redacted", "staff-lawyer-id-redacted", "duecare-engineer-id-redacted"]

cluster_weights:
  legal_grounding_precision: {threshold: 0.70, rationale: "..."}
  # ... (see Day 2)

citation_overrides:
  - id: "poea_mc_14_2017"
    verified: true
    verified_at: "2026-05-21"
    notes: "Verified against POEA Memorandum Circulars register."

templates_additional:
  - id: "dmw_show_cause_notice_v3"
    purpose: "Administrative show-cause notice for recruitment-agency licensees"
    file: "templates/dmw_show_cause_notice_v3.md"

audit:
  retention_years: 7
  hash_algorithm: "sha256"
  pii_redaction_targets:
    - "name"
    - "passport_number"
    - "phone"
    - "address"
    - "dob"
    - "philsys_id"
  api_egress_allowed: false

unsupervised_use_allowed: false  # workbench requires inspector login
notes: |
  Tuning conducted 2026-05-19 to 2026-05-22. Signed off by senior
  inspector and staff lawyer.

This YAML is checked into the agency's internal repo. It is the authoritative configuration for the agency's DueCare deployment.

When you re-do the tuning¶

Quarterly review (4 hours):

Walk the citations: any new statutes? Any amendments?
Walk the templates: any new enforcement instruments?
Walk the audit log: any incidents this quarter? Anything to change?
Walk the rubric: does the cluster weighting still match agency priorities?

Plus any time:

A new senior statute is enacted (immediate update)
A new agency mandate is added (cluster reweighting)
A FOIA / right-to-information request reveals an issue with audit configuration (immediate update)
A pilot turns up cases the rubric misses (rubric tuning)