tinh2 Commercial Lease Abstraction
You convert a lease PDF (and amendments) into a structured abstract. Manual lease abstraction takes 4-12 hours per document depending on complexity; AI-assisted reduces that to 20-40 minutes of human review on top of a generated draft. The standardization matters as much as the speed — every abstracted lease should fit the same schema so a portfolio can be queried.
============================================================ === PRE-FLIGHT ===
- Lease document(s): original lease PDF + ALL amendments + side letters + work letters. Missing amendments = wrong financial picture.
- Lease type: office, retail, industrial, ground lease, residential (CA Civil Code §1947 differs), medical office, mixed-use. Drives which clauses to expect.
- Output schema target: VTS, Visual Lease, MRI, Yardi, ProLease, Excel template, or generic JSON. Each system has slightly different field names.
- Reporting purpose: lease admin (operational), tenant rep (negotiation prep), landlord acquisition DD, ASC 842/IFRS 16 accounting capture.
Recovery:
- Missing amendments: refuse to publish a final abstract; output draft + flag "amendments missing — DO NOT rely on for accounting."
- For complex retail leases (% rent + CAM gross-ups + exclusivity webs), surface complexity to a human reviewer; the abstract is a draft.
============================================================ === PHASE 1: DOCUMENT INTAKE + OCR ===
PDF preprocessing:
- If image-only PDF, OCR via
pytesseractor AWS Textract (better for table extraction). - Split lease into named sections (definitions, basic terms, rent, additional rent, options, default, indemnity).
- Extract embedded tables (rent schedule, CAM exhibit) as structured data, not flowed text.
Persist lease.text.json with section anchors so the abstract can cite source page+section per extracted field.
VALIDATION: Page count > 0, no all-white pages, text extraction confidence per section ≥ 80%.
============================================================ === PHASE 2: STANDARDIZED FIELD EXTRACTION ===
Extract into this schema (camelCase JSON):
{
"parties": {"tenant": "...", "landlord": "...", "guarantor": "...", "broker": "..."},
"premises": {"address": "...", "suite": "...", "rsf": 12500, "usf": 11700, "load_factor": 0.068},
"term": {
"commencement_date": "2024-04-01",
"expiration_date": "2034-03-31",
"term_months": 120,
"rent_commencement_date": "2024-06-01",
"free_rent_months": 2
},
"rent_schedule": [
{"period": "Year 1", "start": "2024-04-01", "end": "2025-03-31", "monthly_rent_psf": 4.50, "monthly_rent_total": 56250.00, "annual_rent_total": 675000},
...
],
"escalations": {"type": "fixed_pct", "pct": 0.03, "annual_review_date": "anniversary"},
"options": [
{"type": "renewal", "count": 2, "term_months": 60, "notice_window_months": 12, "rent_basis": "FMR", "rent_floor_pct": null}
],
"additional_rent": {
"expense_structure": "modified_gross",
"base_year": 2024,
"expense_stop_psf": null,
"tenant_share_pct": 0.085,
"cap": {"type": "cumulative_5pct_year_over_year", "applies_to": "controllable_only"},
"gross_up_provision": true,
"audit_rights": {"window_months": 12, "self_help": false}
},
"ti_allowance": {"amount_psf": 75.00, "total": 937500, "unused_treatment": "rent_credit"},
"security_deposit": {"type": "letter_of_credit", "amount": 250000, "burn_down": [{"month": 36, "to": 125000}, {"month": 60, "to": 0}]},
"use_restriction": "general office and ancillary uses; no medical, retail, or industrial",
"co_tenancy": null,
"exclusivity": null,
"assignment_sublet": {"consent_standard": "not_unreasonably_withheld", "permitted_transfers": ["affiliates", "merger_consolidation"]},
"termination_rights": {"tenant_kick_out_year": 7, "notice_months": 12, "termination_fee": "9 months base rent + unamortized TI/commissions"},
"expansion_rights": {"type": "ROFO", "size_sf": null, "notice": "60 days"},
"holdover": {"rate_pct_of_holdover": 1.50, "consent_required": false},
"insurance": {"tenant_liability_minimum": 3000000, "additional_insureds": ["landlord", "property_manager"]},
"default": {"monetary_cure_days": 5, "non_monetary_cure_days": 30, "late_fee_pct": 0.05, "interest_rate": "prime + 4%"},
"estoppel_sndaa": {"estoppel_required": true, "snda_attached": true},
"amendments_incorporated": ["Amendment 1 (2024-09-15)", "Amendment 2 (2025-03-01)"]
}
Every field has a source sub-field with {page, section, confidence}.
VALIDATION: Schema fields present (null OK if absent in lease — don't fabricate). Rent schedule sum reconciles to total committed rent over term.
============================================================ === PHASE 3: AMBIGUITY & FLAG DETECTION ===
Some lease clauses are inherently ambiguous; flag for human review:
- Gross-up provision boundaries: which expenses can be grossed up to 95%/100% occupancy? Listed expressly or implicit?
- CAM cap: cumulative vs annual? Applies to all expenses or controllable only? Taxes/insurance excluded?
- Tenant share: pro rata by RSF or fixed percentage? Adjustable on changes to building size?
- Renewal rent basis: FMR with floor? Pre-negotiated rate? Greater of (X, Y)?
- Assignment consent: "consent not unreasonably withheld" vs absolute discretion vs deemed-consent if no response in N days.
- Termination fee: includes unamortized TI? Brokerage commissions? Free rent? At what discount rate?
Flag each as {field, ambiguity, suggested_interpretation, reviewer_action}.
VALIDATION: At least 1 flag per lease (if zero, the abstraction is too confident and worth a second pass).
============================================================ === PHASE 4: CRITICAL DATES CALENDAR ===
Emit a calendar file (critical_dates.ics + critical_dates.csv) with reminders:
- Rent escalation date (60-day and 30-day pre-alerts)
- Renewal option exercise window opening + closing (1 year and 6 month pre-alerts)
- Lease expiration (24 / 12 / 6 / 3 month pre-alerts)
- Termination right exercise window (if applicable)
- Expansion/ROFO/ROFR triggers
- Security deposit burn-down dates
- CAM audit window (typically 12 months after year-end statement)
- Estoppel/SNDA delivery deadlines
VALIDATION: Calendar imports into Google/Outlook/iCal without errors.
============================================================ === PHASE 5: ABSTRACT OUTPUTS ===
Final deliverable directory:
lease-{tenant-slug}-{address-slug}/
├── README.md # what each file is and where it goes
├── abstract.json # full schema (Phase 2)
├── abstract.csv # flat row for portfolio rollup
├── abstract.md # human-readable summary
├── critical_dates.ics
├── critical_dates.csv
├── flags.md # ambiguities (Phase 3)
├── source_lease.pdf # original (link, not copy)
└── audit_trail.json # extraction confidence per field
The Markdown summary is a one-pager: parties, premises, term, rent, escalations, options, key flags. Designed for handoff to a broker / attorney / portfolio manager.
VALIDATION: All files render. CSV imports into Excel without column drift.
============================================================ === SELF-REVIEW ===
Score 1–5:
- Complete: All schema fields populated or explicitly null?
- Robust: Amendments incorporated? Rent reconciliation passes?
- Clean: Markdown summary fits one page? Critical dates calendar correct?
- CRE-credible: Would a lease admin or tenant-rep broker accept the abstract as a working draft?
Common gap: forgetting to incorporate amendments. The "original" lease's rent schedule is often superseded by an amendment.
============================================================ === LEARNINGS CAPTURE ===
Append to ~/.claude/skills/lease-abstraction/LEARNINGS.md:
— <asset type, lease length, complexity>
- What worked / awkward / patch / verdict.
============================================================ === STRICT RULES ===
- Never publish a final abstract without incorporating all amendments.
- Never fabricate a field. Null is acceptable; invented values are a tort.
- Never auto-import to Argus / VTS without reviewer sign-off. Wrong rent schedule = wrong valuation = lawsuit.
- Always preserve source citations per field for auditability.
- Always run the critical-dates calendar — missing a renewal notice can lose a tenant a premises.