# Fortax Reconciliation > Reconcile two sets of records that should agree and explain every difference before any adjustment - books vs return (GSTR-1/3B/2B vs registers, 26AS/AIS vs books), control account vs subledger (debtors, creditors, fixed asset register, stock), bank vs books, related-party / group balances, a trial balance that does not tie. Narrows the gap to rows, tests one cause at a time, categorises reconciling items, ages them and states what stays open. Typical asks - "difference nikalo", "2B aur books me farak kyun hai", "26AS match nahi ho raha", "TB tally nahi ho raha", "debtors ledger aur control account alag hai". - Skill: `amit-voais/fortax-reconciliation` (Agent Skill, multi-file: 4 files) - Install (CLI): `npx skillmds@latest add amit-voais/fortax-reconciliation` - Raw SKILL.md: https://api.skillmd.com/api/skills/amit-voais/fortax-reconciliation/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - License: Apache-2.0 - Author: amit-voais (https://skillmd.com/u/amit-voais) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/amit-voais/fortax-reconciliation --- # Reconciliation — find the difference, then explain it Needs Python 3 and internet: runs `scripts/fortax.py` (the Fortax engine on ai.fortax.in; your file is processed and not stored). Verdict first: two records agree, or they differ by an exact amount made of named items plus an unexplained balance. This skill helps prepare the working; the CA reviews and signs it off. ## The rule **No adjustment before the cause is known.** A difference that is plugged, rounded away, or "adjusted to match" is not reconciled — it is hidden, and it will surface in a notice. This applies hardest when the difference is small. A Rs 118 gap is a wrong tax rate somewhere; a Rs 1 gap is a rounding rule you have not identified yet. Size is not evidence of harmlessness. ## Use an engine where one exists | Pair | Do this | |---|---| | GSTR-2B vs purchase register | `fortax-gstr2b-reconciliation` (`fortax.py gstr2b`) — it matches; you judge its `_pairs_to_judge.json` | | GSTR-1 / 3B vs sales register | `fortax-gstr1-and-3b` (`fortax.py gstr1`) | | Bank statement vs books | `fortax-bank-statement-to-books` (`fortax.py bank`, then its BRS part) | | Everything else | this skill, with the matching done in a script or a workbook with formulas (`fortax-xlsx-manipulation`) | Numbers come from code: never add a column, compute a difference or count items by eye. ## When to use Any two records that should agree and do not: - GSTR-2B against the purchase register; GSTR-1 against the sales register; 3B against both. - 26AS / AIS / TIS against the books or the computation. - A trial balance that does not tie; a control account against its subsidiary ledger. - A portal figure that differs from the working paper. - Balances between related entities (holding / subsidiary, group companies, branches). - A total that changed between two versions of your own file. ## Step 1 — state the difference exactly Before anything else, write one line: > As per Rs X; as per Rs Y; difference Rs Z (A - B), for , . Get the period, the entity and the sign right. Half of all "differences" are two documents covering different months, or one figure inclusive of tax and the other not. ## Step 2 — narrow it before you explain it Split the difference until it is a small set of rows, not one number: - By month, then by party, then by invoice. - By tax head — IGST, CGST, SGST, cess separately. A gap that vanishes when heads are separated is a classification issue, not a value issue. - By status — matched, unmatched in A, unmatched in B, matched but different value. Stop splitting when each remaining row can be explained on its own. ## Step 3 — the usual causes, checked in order Check these before inventing a theory. Each is a fact you can confirm from the files: | Cause | How you confirm it | |---|---| | Timing | The invoice exists in both, in different periods | | Not filed by the supplier | Present in the register, absent from 2B; the supplier's return status says so | | Wrong GSTIN or PAN | The party's identifier differs between the two records | | Reverse charge, or ineligible credit | The entry is correctly excluded from one side by rule | | Rate or head applied differently | Same taxable value, different tax split | | Debit or credit note not carried through | One side nets it, the other does not | | Rounding | The gap is under Rs 1 per line and matches the line count | | Duplicate | The same invoice number appears twice on one side | Typical causes by type of reconciliation are in the next section. ## Step 4 — one hypothesis at a time State it in writing: "I think Rs Z is , because ." Then test only that, against the source document — not against your own earlier working. If it does not hold, say so and take the next hypothesis. Never stack two guesses. ## Step 5 — when it still does not tie After three explanations that do not hold, stop. Do not keep adjusting. Write the CA a short note: the exact difference, the rows you isolated, what you ruled out and how, and what you would need to settle it — usually a document, a supplier or party confirmation, or a portal download. An honest open difference is a working paper. A forced tie is a liability. ## Reconciliation types ### Control account vs subledger Compare the ledger control balance to the detailed subsidiary balance at the same date. - Sundry debtors (control vs party-wise ageing), sundry creditors (control vs party-wise ageing). - Fixed assets (ledger vs fixed asset register — Companies Act and Income-tax blocks separately). - Inventory (stock ledger vs valuation / stock statement given to the bank). - Prepaid expenses (ledger vs amortisation schedule); provisions and outstanding expenses (ledger vs the detail schedules). Process: pull the control balance at period end; pull the subledger detail at the same date; compare totals (they should match if posting is real-time); investigate differences. Common causes: a journal passed directly to the control account and not in the subledger; subledger transactions not yet posted to the ledger; timing of batch posting; a reclassification in the ledger without the subledger; an import or interface error. ### Bank vs books Use `fortax-bank-statement-to-books` Part B. In brief: bank balance; add deposits in transit; less cheques issued not presented; add/less bank errors = adjusted bank balance. Book balance; add interest and credits not recorded; less charges not recorded; add/less book errors = adjusted book balance. Difference must be 0.00. ### Related-party and group balances Balances between related entities should net to zero on consolidation and agree for the related-party disclosure (AS 18 / Ind AS 24, s.188 Companies Act). 1. Pull the receivable/payable for each entity pair. 2. Compare A's receivable from B with B's payable to A. 3. Identify and resolve differences; confirm every transaction is recorded on both sides. 4. Check the elimination entries used for consolidation. Common causes: one entity recorded it, the other has not yet (timing); different exchange rates; classified as third-party on one side; disputed amounts or unapplied payments; different cut-off practice; TDS deducted by the payer shown net on one side and gross on the other. ### Tax records vs books - 26AS / AIS vs books: TDS credit by deductor TAN and section; income shown in AIS that is not in books; a deductor who deducted but filed against a wrong PAN. Confirm with the deductor's Form 16A. - GST returns vs books: see the table above; for the annual view see `fortax-gst-annual-return`. ## Categorise every reconciling item **Category 1 — timing differences.** Clear on their own in the normal cycle; no entry needed: cheques issued not presented, deposits in transit, items posted in one system and pending in the other, transactions awaiting approval. Expected to clear within a few working days. **Category 2 — adjustment required.** Need a journal entry (prepare it with `fortax-journal-entry`): unrecorded bank charges, unrecorded interest, recording errors (wrong amount, wrong ledger, duplicate), missing entries, classification errors. **Category 3 — requires investigation.** Cannot be explained yet: unidentified differences, disputed items, aged items that should have cleared, the same unexplained difference every period. Investigate the root cause, document it, escalate if unresolved. ## Ageing of open items | Age | Status | Action | |---|---|---| | 0-30 days | Current | Monitor — within the normal cycle | | 31-60 days | Ageing | Investigate — why has it not cleared | | 61-90 days | Overdue | Escalate to the reviewer; document the investigation | | 90+ days | Stale | Escalate to the CA / management; possible write-off or adjustment | Report format: | Item # | Description | Amount (Rs) | Date originated | Age (days) | Category | Status | Owner | |---|---|---|---|---|---|---|---| Compute the age with a formula (`= - `), not by counting. Trending: compare total open items with last period; flag if the total exceeds the materiality level, if the count grows period on period, or if the same item recurs every period (a process problem). ## Escalation thresholds (illustrative — set with the CA for each client's size) | Trigger | Example threshold | Escalate to | |---|---|---| | Single item | above Rs 1 lakh | Reviewer | | Single item | above Rs 10 lakh | CA / partner | | Total reconciling items | above Rs 25 lakh | CA / partner | | Item age | over 60 days | Reviewer follow-up | | Item age | over 90 days | CA / management | | Unexplained difference | any amount | Cannot close — resolve or document as open | | Growing trend | 3+ consecutive periods | Process fix | ## Good practice 1. **On time:** finish within the close timetable (typically 3-5 working days after period end, and before the return that depends on it). 2. **Complete:** reconcile every balance sheet account on a set frequency — monthly for material ones, quarterly for the rest. 3. **Documented:** preparer, reviewer, date, and a clear explanation of every reconciling item. 4. **Segregated:** whoever reconciles should not be the person who posts to that account. 5. **Followed through:** track open items to resolution; do not carry them forward indefinitely. 6. **Root cause:** for recurring items, fix the process that causes them. 7. **Standard:** the same template for every account. 8. **Retained:** keep reconciliations and support for the statutory period (books of account: 8 years under s.128 Companies Act; confirm for the entity and the tax law). ## Output Into the client folder next to the inputs (`///`), with entity and period in the name: - A reconciliation sheet: matched, unmatched each way, value differences, with a `Source` column on every row and the category and age of every open item. - A difference statement: opening gap, each explained item with its cause and amount, closing unexplained gap — all by formula. - **Any unexplained balance is a blocker in your summary to the CA, not a footnote.** ## Credit Reconciliation types, item categories, ageing and escalation are adapted from anthropics/knowledge-work-plugins (Apache-2.0). Changed by Fortax: rewritten for Indian practice (GST, 26AS, Companies Act, rupee thresholds) and merged with Fortax's find-the-difference method.