Contact Manager

Step 2: Create normalized schema

Output format (per account): contacts_normalized.csv

email,first_name,last_name,company,category,touchbase_cadence,notes

For SKEstates (expanded schema with department/phone):

email,first_name,last_name,title,company,department,category,touchbase_cadence,phone,notes

Step 3: Category classification

Categories:

• client — current/past clients (EPCI operators, engineering firms)
• colleague — current/former colleagues, same-company contacts
• vendor — service providers, suppliers
• recruiter — staffing, job-related contacts
• industry — industry organizations, standards bodies
• alumni — university contacts (Rice, TAMU, UT McCombs)
• personal — friends, family (gmail/yahoo with known names)
• financial — banks, insurance, investment
• government — tax, legal, regulatory
• newsletter — subscriptions, marketing lists
• spam — junk, craigslist, unknown bulk
• tenant — tenant contacts (for real estate accounts)
• vendor-functional — shared functional inboxes (no touchbase)
• tenant-functional — shared tenant inboxes

Engineered domain mappings (from iteration on 2,400+ contacts)

Ace client domains:

ril.com, dorisgroup.com, mcdermott.com, shell.com, kbr.com,
technip.com, technipfmc.com, subsea7.com, nov.com, aker.com,
bp.com, awilcodrilling.com, eagle.org, vulcanoffshore.com,
boptechnologies.com, risersinc.com, sandsig.com,
engineeredcustomsolutions.com, mecorparada.com.ve

Ace colleague domains:

trendsetterengineering.com, spire-engineers.com,
2hoffshoreinc.com, 2hoffshore.com, prospricing.com

Ace vendor domains:

disys.com, winworldinfo.com, partneresi.com, ansys.com,
akselos.com, engys.com, dnvgl.com, tescocorp.com,
deccaconsulting.com, flooranddecor.com, pulse-monitoring.com,
quantumep.com, acematrix.com

Ace recruiter domains:

stepstoprogress.com, thejukesgroup.com, apexsystems.com,
indianeagle.com

Ace industry domains:

km.kongsberg.com, ceesol.com

Personal alumni domains:

rice.edu, mccombs.utexas.edu, neo.tamu.edu, tamu.edu, houstonisd.org

Personal financial domains:

aaa-texas.com, colehealth.com, harkandgroup.com, aol.com,
sbcglobal.net, constellation.com

Spam domains (auto-remove):

sale.craigslist.org, talkmatch.com

Spam patterns (regex):

unsubscribe, no.?reply, noreply, do.?not.?reply
mailer-daemon, postmaster@, bounce@
@unsubscribe2\., \.unsubscribe\., @sailthru\., @mcsv\., @rsys5\., @customer\.io

Spam name fragments:

craigslist, mailer-daemon, unsubscribe, academia.edu, 123greetings, machinemetrics

Step 4: Intra-file deduplication

Rules:

• Same primary email → keep first occurrence
• Same name, different email → keep both, flag for review
• No name, only email → try to derive name from email prefix

Step 5: Cross-file deduplication

132 overlaps found between ace and personal files. Decision rules (applied in order):

1. aceengineer.com → ace only (internal domain)
2. achanta in email prefix → personal only (family)
3. Company domain (not gmail/yahoo/hotmail) → ace only (professional)
4. Personal email with company tag in source → ace only
5. Personal email with company in ace record → ace only
6. Both unknown, personal email → personal only
7. Ace has real classification → ace wins

Result: 122 moved from personal to ace, 10 kept in personal. Zero remaining overlaps.

Report template: reports/email/contact-dedup-report.md

Scripts

Contact normalizer (run with uv)

uv run scripts/email/contact-normalizer.py

Input: raw Outlook CSV exports from both accounts. Output: aceengineer_normalized.csv (1,281 contacts) + achantav_normalized.csv (994 contacts).

SKEstates contact builder

Built from two markdown files:

• key_contacts.md (3 PCA vendor contacts)
• fd_corporate_contact_maintenance.md (22 Family Dollar/Dollar Tree staff)

Output: sabithaandkrishnaestates/admin/contacts/skestates_contacts.csv (25 contacts).

Count unknown domains to discover missing mappings

unknown_domains = Counter(r["email"].split("@")[-1] for r in contacts if r["category"] == "unknown") for domain, count in unknown_domains.most_common(20): print(f" {domain:40s} {count}")

Then expand the domain classification sets (ACE_CLIENT_DOMAINS, ACE_VENDOR_DOMAINS, etc.)
in the normalizer script and re-run. Second pass typically reaches 80%+ classification.

### Step 4: Cross-file deduplication

When contacts appear in multiple account files (email addresses imported into
multiple Outlook exports), resolve overlaps using these rules:

| Signal | Canonical Home | Rationale |
|---|---|---|
| aceengineer.com domain | ace | Internal domain |
| achantav* in email prefix | personal | Family email |
| Company domain (non-gmail/yahoo) | ace | Professional contact |
| Personal email with company in record | ace | Cross-referenced professional |
| Both unknown, personal email | personal | No business value |
| Ace has classification | ace | Already categorized |

After dedup: annotate the canonical file with cross-ref note, remove from duplicate file.

### Step 5: Write output files

## Scripts

### Contact normalizer (run with uv)
```bash
uv run scripts/email/contact-normalizer.py

The normalizer processes ALL accounts in one run. It reads both raw CSVs and outputs normalized files alongside them. No CLI args needed — paths are hardcoded for the workspace-hub convention.

Re-running after adding domain mappings

Edit the domain sets in scripts/email/contact-normalizer.py (ACE_CLIENT_DOMAINS, ACE_VENDOR_DOMAINS, etc.) then re-run. The script is idempotent.

File Locations (post-normalization)

Pitfalls

1. Outlook CSV export has 64 columns with verbose names ("E-mail Address", "E-mail 2 Address", etc.)
2. Email addresses often wrapped in angle brackets: <[email protected]>
3. Some entries have email stored in "First Name" field when name is unknown
4. Windows BOM (utf-8-sig) — always open with encoding="utf-8-sig"
5. Empty first/last name rows are common (~30-36% of contacts) — use email prefix for name inference
6. Cross-file email overlap is guaranteed (~132 of ~2,400) — always dedup after normalizing
7. External repos (aceengineer-admin, sabithaandkrishnaestates) must be committed from within their own git dirs
8. Legal scan: contact CSVs may contain client names — run .legal-deny-list.yaml scan before committing
9. Don't delete raw Outlook exports — keep originals, create _normalized.csv alongside
10. Domain inference from email prefix is heuristic-only — tag uncertain results for manual review
11. Don't delete raw CSVs — keep originals, create normalized copies alongside
12. Legal: contact data is PII — never commit to public repos, .gitignore if needed
13. The personal CSV has craigslist/talkmatch spam entries — auto-classify as spam
14. Company field often empty — try to infer from email domain
15. External repos (aceengineer-admin, sabithaandkrishnaestates) are separate git repos — cd into them to commit, don't use git add from workspace-hub root
16. First-pass domain classification yields ~20% — ALWAYS run the unknown-domain-frequency analysis and expand mappings before claiming done
17. The contact-normalizer.py script lives in workspace-hub root (scripts/email/) but reads/writes in repo subdirectories — it finds the root by walking up from its own path