Table of Contents

Back to Articles
Finance Engineering & ERP

The Architecture of Deterministic BRS: Why Vertis Uses Exact-Decimal Code Beside Tally Prime

Editorial Architecture:Published by Misora Labs Engineering. Explores deterministic systems, financial reconciliation, and operational AI.

Why probabilistic LLMs fail at accounting math, and how Finora’s deterministic Vertis engine matches complex UTR narrations, fee splits, and TDS schedules with zero general ledger risk.

2026-11-14

Abhranshu Bagchi (Founder & Principal Systems Architect, Misora Labs)

6 min read

Provenance notice — pending founder confirmation

Metrics and timelines on this page describe an illustrative engineering deployment narrative used to explain Finora's deterministic matching behaviour. This is not a named customer endorsement unless explicitly labelled elsewhere with verified provenance.

The Fatal Flaw of "AI-Powered" Accounting

Over the last 18 months, dozens of "AI accounting startups" launched with the premise that large language models could replace the manual grind of Bank Reconciliation Statements (BRS). In practice, almost all of them failed when exposed to real-world corporate ledgers.

The reason is mathematical: Large language models are probabilistic token predictors, not arithmetic calculators. An LLM does not compute exact balance derivations—it predicts the most likely next string. In finance, being 99.4% accurate is not a feature; it is an audit catastrophe. If a system reconciles ₹48,00,000 and misses ₹280 on a TDS withholding line, the books do not balance, the auditor rejects the schedule, and senior accountants spend days manually tracking the discrepancy.

The Deterministic Boundary: Enter the Vertis Engine

At Misora Labs, we built Finora around a non-negotiable architectural rule: Money is deterministic, exact-decimal code.

All reconciliation math, multi-way matching, and balance derivations in Finora are executed by Vertis—our proprietary, 100% deterministic matching engine written in Python and compiled exact-decimal logic. Language models are strictly firewalled outside the calculation core.

// THE VERTIS DETERMINISTIC BOUNDARY

[Bank Feeds / MT940] ──┐

[Gateway Payout CSVs] ──┼──► [VERTIS DETERMINISTIC ENGINE] ──► [Exact-Decimal Match Queue]

[Tally / NetSuite Daybook] ┘ (Exact Arithmetic, Zero LLM) │

▼

[LLM Synthesis Layer]

(Plain-English Controller Brief)

How Vertis Handles Complex UTR Narration Breaks

Corporate bank statements rarely contain clean metadata. A typical NEFT or RTGS credit line arrives with messy abbreviations: NEFT/AXIS0029182/INFOSYS_LTD/INV9821_LESS_TDS. Standard rule-based scripts fail because vendor names and invoice numbers are truncated.

Vertis solves this through a multi-pass pipeline:

  1. Deterministic Normalization: Strips bank routing noise, isolates standardized 12-to-16 digit UTR strings, and extracts date/amount pairs.
  2. One-to-Many & Fee-Adjusted Pass: Matches bundled gateway payouts where one lump-sum bank credit represents 42 individual invoices minus 1.85% MDR fees.
  3. TDS Schedule Correlation: Automatically derives whether a 2% or 10% withholding explains an exact variance between invoice value and received cash.
  4. Voucher-Level Provenance: Unresolved breaks do not disappear into a black box; they surface in a ranked controller queue showing the exact bank line and matching candidates.

The "Do No Harm" Guarantee: Read-Only Beside Tally Prime

The single greatest hesitation for CFOs and Financial Controllers is the fear of third-party software corrupting their General Ledger. Finora guarantees 100% data safety:

  • Read-Only Ingestion: Finora operates exclusively on exported XML/CSV daybooks, bank statement files, and read-replica databases.
  • Zero Autonomous Ledger Writes: Finora never writes back into Tally Prime or Oracle NetSuite without explicit voucher-by-voucher approval from an authorized controller.
  • Audit-Proof Trail: Every matched line carries immutable cryptographic timestamps and matching rule IDs, ready for Big-4 statutory audit inspection.

Summary: Proof Over Hype

In our recent deployment with a mid-market manufacturing enterprise running on Tally Prime, Finora ingested 30 days of multi-bank feeds and cleared 21 open reconciliation exceptions totaling ₹48,22,500 within 48 hours—with zero engineering dependency after kickoff.

If your finance team is losing the first week of every month to manual VLOOKUP detective work, explore our Finora Subsystem or Scope a 14-Day Diagnostic Sprint.

Topic

Finance & deterministic reconciliation

Bank reconciliation, ERP-adjacent matching, and controller-safe automation beside systems of record — without autonomous ledger writes.

Scope a project conversation

Eliminate month-end reconciliation toil on your infrastructure

Our principal systems architects deploy Finora beside your existing Tally Prime or NetSuite ERP read-only, resolving multi-bank and unapplied cash exceptions with exact-decimal Vertis math.

Actionable Operations Excellence

Autonmis helps modern teams own their entire operations and data workflow — fast, simple, and cost-effective.