Report Spec — what we show from a statement
What the analyzer extracts and displays for a lender. Machine schema:
app/lib/report-types.ts(canonical:server/src/statements/report.py). Sample data:app/lib/sample-report.ts. Live sample:app.obsrv.in/report. ★ = MVP (show first).
| # | Section | Fields | MVP |
|---|---|---|---|
| A | Statement & account header | Bank, branch, holder, masked a/c, type, IFSC, period, opening/closing balance, source type (digital/scanned/password), txns parsed, extraction confidence | ★ |
| B | Verdict | Plain-English summary, positives, concerns, indicative risk band (Low/Med/High) + score, recommended next checks | ★ |
| C | Income | Detected income sources (source, amount, regularity, class salary/business/rental/interest/government, tier core/supplementary), avg monthly income split into core (drives FOIR) + supplementary, stability score, trend, last income date | ★ |
| D | Balance | Avg monthly balance, avg daily balance, min/max (with dates), negative-balance days, median end-of-day balance, liquidity buffer ratio (median month-end ÷ monthly income), days below 10% of monthly income | ★ |
| E | Cash flow | Total credits/debits, net flow, inflow/outflow ratio, surplus vs deficit months, month-by-month table (credits, debits, net, AMB, closing) | ★ |
| F | Obligations & capacity | EMIs/loans, rent, insurance, SIPs (each flagged whether it counts toward FOIR per lender policy); total monthly obligations; FOIR (debt-to-income); estimated disposable income | ★ |
| G | Risk flags | NACH/ECS bounces (recency-weighted), cheque returns, penal charges, negative/overdraft events, circular transfers, sudden large credits (window-dressing/pass-through), high cash deposits, loan stacking, paycheck-to-paycheck, revolving credit-card, service-failure charges, inactive income, speculative activity (gambling/crypto), joint-account caveat | ★ |
| H | Document integrity | Balance-continuity math check, duplicate detection, page-sequence, PDF metadata anomalies, font/alignment, authenticity confidence | basic ★ / forensics later |
| I | Categorization & counterparties | Spend by category; top 5 money received-from + top 5 money sent-to (amount, txn count, share); channel mix (UPI/NEFT/RTGS/IMPS/NACH/SI/AePS/BBPS/wallet/cash/card); unique counterparties; largest single credit/debit; total cash deposits; salary share of income | ★ |
| J | Transactions ledger | date, narration, ref, debit, credit, balance, mode, category, counterparty, flags | ★ |
| K | Trust footer | Overall confidence, fields needing human review, reconciliation pass/fail, disclaimer | ★ |
| L | Ask-the-statement | Natural-language Q&A over the data | later |
Principles
- Never silently guess. Uncertain fields carry confidence and go to
reviewFlagsfor a human. - Indicative, not a decision. The verdict assists underwriting; it is not a credit decision.
- Show the math. Aggregates trace back to the transaction ledger so a lender can trust them.