diff --git a/docs/plans/2026-02-10-spending-analysis-app-design.md b/docs/plans/2026-02-10-spending-analysis-app-design.md
index 67bcfbe..a223ec8 100644
--- a/docs/plans/2026-02-10-spending-analysis-app-design.md
+++ b/docs/plans/2026-02-10-spending-analysis-app-design.md
@@ -35,6 +35,16 @@ src/
 
 - id, name, relationship
 - Seeded with the primary user on first run
+- Initial household: Andrew (self), Donna (wife), son
+
+### Known income sources
+
+Based on real data:
+- **Andrew's salary** — CIM Techniques via QuickBooks direct deposit (biweekly, ~$2,314)
+- **Donna's salary** — Oasis Batch payroll (biweekly, varies ~$998–$1,185)
+- **Other deposits** — Mobile deposits, Capital One transfers in
+
+These are auto-detected as income and attributed to the corresponding household member via categorization rules.
 
 ### accounts
 
@@ -65,18 +75,60 @@ src/
 
 ## CSV Import & Mapping
 
+### Known Source Formats
+
+Based on real sample data:
+
+**Chase credit card** (e.g., `Chase0372_Activity*.CSV`):
+- Has headers: Transaction Date, Post Date, Description, Category, Type, Amount, Memo
+- Amount is signed (negative=purchase, positive=payment/return)
+- Type column: Sale, Payment, Return
+- Includes Chase's own category assignments — import as initial suggestions
+
+**Wells Fargo checking** (e.g., `Checking1.csv`):
+- No headers — five positional columns: Date, Amount, Flag ("*"), Check Number, Description
+- Amount is signed (negative=debit, positive=credit)
+- Descriptions are very verbose with embedded authorization dates, card numbers, and reference IDs
+- Requires description normalization to extract core merchant/payee
+
+### Headerless CSV support
+
+The mapping wizard handles files with no headers by showing column previews with positional indices. User assigns meaning by column position rather than header name.
+
+### Description normalization
+
+The service layer strips noise from transaction descriptions before storage:
+- Authorization dates ("PURCHASE AUTHORIZED ON 02/06")
+- Card numbers ("CARD 5360")
+- Reference IDs ("S586033096695382", "REF #OP0WS99NKQ")
+- Redundant location/phone details
+
+The cleaned description is stored for display and rule matching. The raw description is preserved in a separate field for reference.
+
+### Source category hints
+
+When a CSV includes pre-assigned categories (like Chase does), the import process can use these as initial category suggestions. The user can accept, override, or ignore them.
+
+### Cross-account transfer detection
+
+The import service detects matching transfer pairs across accounts. For example:
+- Checking: "CHASE CREDIT CRD EPAY 260128 ... -$1,461.35"
+- Chase: "Payment Thank You - Web ... +$1,461.35"
+
+These are linked as transfers so they don't double-count as spending or income.
+
 ### First import from a new source
 
-1. App reads CSV, detects delimiter and headers
+1. App reads CSV, detects delimiter, and checks for headers vs. headerless format
 2. Mapping wizard shows a preview of the first few rows
-3. User maps columns to normalized fields: date, amount, description, and optionally reference number, balance
+3. User maps columns to normalized fields: date, amount, description, and optionally reference number, balance, source category
 4. User specifies amount logic (single signed column, separate debit/credit columns, type column)
 5. User selects or creates the account this CSV belongs to
-6. Mapping saved to `csv_mappings` keyed on column header fingerprint
+6. Mapping saved to `csv_mappings` keyed on column header fingerprint (or column count + sample patterns for headerless files)
 
 ### Subsequent imports
 
-1. App matches CSV headers to a saved mapping
+1. App matches CSV headers (or structure) to a saved mapping
 2. Confirmation screen shows: transaction count, date range, duplicates detected
 3. User confirms before committing
 
@@ -96,12 +148,23 @@ Categorization engine runs on new transactions. Unmatched transactions flagged f
 
 ### Rule-based categorization
 
-Rules match patterns against transaction descriptions and assign: category, optional tag override, optional person attribution. Examples:
+Rules match patterns against transaction descriptions and assign: category, optional tag override, optional person attribution. Examples from real data:
 
-- `NETFLIX` -> Entertainment, Wants
-- `KROGER` -> Groceries, Needs
-- `TRANSFER WIFE VISA` -> Transfer, attributed to Wife
-- `ALLOWANCE` -> Family, Wants, attributed to Son
+- `CIMTECHNIQUES` -> Income, attributed to Andrew
+- `OASISBATCH PAYROLL` -> Income, attributed to Donna
+- `CHASE CREDIT CRD EPAY` -> Transfer
+- `CAPITAL ONE TRANSFER` -> Transfer
+- `AMEX EPAYMENT` -> Transfer, attributed to Donna
+- `FREEDOM MTG PYMTS` -> Housing, Needs (mortgage)
+- `DOMINION ENERGY` -> Utilities, Needs
+- `HELLOFRESH` -> Groceries, Needs
+- `Netflix.com` -> Subscriptions, Wants
+- `PUBLIX|ALDI|PIGGLY WIGGLY|WAL-MART` -> Groceries, Needs
+- `CHICK-FIL-A|KFC|MCDONALD` -> Dining Out, Wants
+- `VW CREDIT` -> Transportation, Needs (auto loan)
+- `FARM BUREAU INS` -> Insurance, Needs
+- `WSFS LOAN` -> Debt Payment, Needs
+- `WAY2SAVE SAVINGS` -> Transfer (savings)
 
 ### Auto-rule creation