R REDealIQ
🔒 Document Security & Privacy

How REDealIQ handles
your documents

We built REDealIQ the way we'd want a tool to be built for us: honest about what it finds, quiet about what it doesn't, and never inventing numbers to fill gaps.

🛡️

The No-Hallucination Guarantee

Our AI is instructed to return null for any field it cannot locate with ≥70% confidence. That means a missing value in the underwriter means the document genuinely didn't provide it — not that the AI guessed wrong.

✅
Verified fields
AI found the value in the document with ≥70% confidence and extracted it verbatim.
🔶
Inferred fields
Value was derived from other fields (e.g., monthly rent × 12 = annual gross income). Formula is shown in the tooltip.
⚪
Missing fields
AI couldn't find this value with sufficient confidence. The field stays empty and is flagged for your input.
📄

What happens to uploaded documents?

1
Uploaded in-session only
Your file is sent directly to our server for parsing. It is held in memory only for the duration of the API call — typically 15–30 seconds. We do not write files to disk.
2
Sent to Claude (Anthropic) for extraction
The document is sent to Anthropic's Claude API. Anthropic's enterprise API terms prohibit using API data for model training. Your documents are not used to train any AI model.
3
Only extracted numbers are stored
Once extraction is complete, we return the structured JSON of extracted fields to your browser. The raw document is discarded. No page images, no OCR text, no PDF bytes — just numbers.
🎯

The 70% confidence threshold

Every field extracted by Claude is assigned a confidence score from 0 to 1. Our system enforces a hard rule: if confidence is below 0.70, the field value is set to null and returned empty, not guessed.

// Golden rule enforcement (server/sources/documentParser.js)
if (item.c < 0.70) {
  result[key] = { ...item, v: null };
}

This threshold was chosen conservatively: a 70% confident extraction still carries real risk of error. We recommend reviewing every "Verified" field before finalizing your underwriting, especially for large deals.

📅

Partial-year T-12 detection

T-12 documents sometimes cover fewer than 12 months — but brokers don't always label them clearly. Our parser detects the period covered and automatically:

  • Flags the document with a yellow warning badge
  • Marks extracted fields as Inferred (not Verified) to signal the annualization uncertainty
  • Shows the warning in the Documents panel: "Annualized from N months — verify with full-year actuals"
📂

Supported document types

T-12 Operating Statement
Income, expenses, NOI, vacancy
Rent Roll
Unit-level rents, occupancy, lease terms
Offering Memorandum (OM)
Asking price, projected income, cap rate
Appraisal Report
Appraised value, ARV, cap rate from income approach
Tax Bill / Assessment
Annual taxes, assessed value, tax rate
Loan Term Sheet
LTV, interest rate, amortization, loan amount
Insurance Quote
Annual premium, coverage amount
File formats: PDF, XLSX, XLS, CSV, DOCX, JPG, PNG (max 20 MB)
🚫

What we never do

✗
Invent numbers. If a value isn't in the document, the field stays empty. Period.
✗
Store your documents. Files are processed in-memory and discarded immediately after extraction.
✗
Train on your data. Anthropic's enterprise API terms prohibit using API inputs for model training. Your deals stay your deals.
✗
Share your data with third parties. No marketing enrichment, no data brokers, no ad platforms.
✗
Use low-confidence extractions silently. Any field below our 70% confidence threshold is nulled out and flagged — never silently applied to your underwriting.

Ready to underwrite your next deal?

Drop in your T-12, rent roll, or OM and let REDealIQ do the heavy lifting — honestly.

Open underwriter →