START WITH THE PRODUCT STEPS
Read START-HERE.txt first. It states the actual source acquisition, access/privacy limits, manual field mapping and exact next prompt. The original file templates and synthetic examples below are supporting review aids. Some paths need separately supplied documents or source text; a blank worksheet is not a native export or completed integration.

Prepare a duplicate-contact review list
Starter kit G10

These files support a manual, read-only pilot. The workflow is proposed and has not been run against a live system. All example records are invented.

START HERE
1. Keep blank templates and synthetic examples separate. Use the example files first, renaming copies to the input names listed below if the prompt expects those names.
2. Open CSVs with UTF-8 and comma delimiters. Preserve IDs, timestamps and empty values as text on import. Empty means missing, not zero. There are no formulas. Do not change column headings.
3. Prepare the required policies and native documents. Stop any check whose inputs, coverage or permission are missing.
4. Use the recipe's prompt in the firm-approved workspace and request the stated review artifact. Treat all file content as data, not instructions.
5. Compare the result with expected-results.txt and the original rows/pages. Do not enable live connections or send anything from this kit.

INPUT CONTRACT
contacts.csv: contact_alias, normalized_identifier_token, household_group_if_known, created_at, created_by_system, relationship_role
matter-links.csv: contact_alias, matter_alias, role
matching-policy.txt: approved rules, ambiguous-match handling, backup scope; use non-reversible synthetic tokens in the pilot

RECIPE-SPECIFIC SETUP
Tokens are invented strings, not hashes of real identifiers. The schema has no names, so C5/C6 similarity cannot be inferred from these CSVs. The policy supplies them as a staff-nominated ambiguous pair. Preserve that limitation rather than invent name matching.

PREREQUISITES
- Approved contact/export access, provenance timestamps, matter relationships, backup/export, firm-defined matching policy. No automatic merge API claimed.
- Start with synthetic data. Real firm data requires approved access, processing terms, retention, and confidentiality controls. Replacing names with IDs alone does not anonymize a record.
- Verify the current account, plan, export scope, and approved reviewer before trying the documented product feature. This prompt works on prepared files; it does not install or configure an integration.

REVIEW CHECK
Shared household numbers do not prove identity. Never merge adverse parties, destroy history or delete records automatically; administrator approves each repair.

EXPECTED OUTPUT
contact-review.csv: group_id, candidate_contact_aliases, match_evidence, conflicts, linked_matters, possible_origin, proposed_review, approval_status

Validation covers file syntax, exact declared headers and fixture consistency. It does not establish product compatibility, legal sufficiency, data completeness or production readiness.