Check GST invoice number series for gaps, repeats and bad formats

What the rule says

Rule 46 of the CGST Rules, 2017 lists the particulars a tax invoice must contain. Clause (b) requires a consecutive serial number not exceeding sixteen characters, in one or multiple series, containing letters, numerals or the special characters hyphen or dash and slash, and unique for a financial year. That gives three machine-checkable conditions. Treat the rule text as the reference and your tax adviser as the authority on how it applies to your business; the check below only tests the register against the wording.

Typical ways a series breaks

Invoice numbers come from billing software, from a manual book at a small branch, and from an Excel template someone built years ago. Common failures:

• a new format introduced mid-year that includes the full year and a document type, pushing the number past 16 characters;
• spaces instead of slashes, typed by hand;
• the series not reset on 1 April, so an April invoice still carries last year's code;
• the same number saved twice when a bill was edited and re-saved;
• numbers skipped when a draft was discarded, with no record of why.

The check

The register below is synthetic: two branch series, Noida (NOI) and Ghaziabad (GZB), across the change of financial year on 1 April 2026. The cancellation register records one cancelled number.

import re
import pandas as pd

# Synthetic sales register: two branch series, crossing the 1 April 2026 year boundary.
reg = pd.DataFrame({
    "inv_no": ["NOI/25-26/0141", "NOI/25-26/0142", "NOI/25-26/0144", "NOI/25-26/0145",
               "NOI/26-27/0001", "NOI/26-27/0002", "NOI/26-27/0002",
               "GZB/25-26/0088", "GZB/25-26/0089", "GZB/26-27/0001", "GZB/26-27/0003",
               "GZB/2026-27/INV/0004", "NOI 26-27 0003", "NOI/25-26/0146"],
    "inv_date": pd.to_datetime(["2026-03-28", "2026-03-30", "2026-03-31", "2026-03-31",
                                "2026-04-01", "2026-04-02", "2026-04-02",
                                "2026-03-30", "2026-03-31", "2026-04-01", "2026-04-03",
                                "2026-04-04", "2026-04-04", "2026-04-01"]),
    "taxable": [42000, 18500, 7600, 12999, 31000, 9800, 9800, 56000, 4400, 22000, 15000,
                8100, 6000, 11200],
})
cancelled = {"NOI/25-26/0143"}             # from the cancellation register, with reasons kept there

# 1. Format: at most 16 characters; only letters, digits, hyphen/dash and slash.
ok_chars = reg["inv_no"].str.fullmatch(r"[A-Za-z0-9/\-]+")
reg["format_issue"] = ""
reg.loc[reg["inv_no"].str.len() > 16, "format_issue"] = "longer than 16 characters"
reg.loc[~ok_chars, "format_issue"] = "character outside letters, digits, - and /"

# 2. Financial year of the invoice date (April to March), and the year the number claims.
fy_start = reg["inv_date"].dt.year - (reg["inv_date"].dt.month < 4)
reg["fy"] = fy_start.astype(str) + "-" + ((fy_start + 1) % 100).astype(str).str.zfill(2)
parts = reg["inv_no"].str.extract(r"^(?P<series>[A-Z]+)/(?P<fy_code>\d{2}-\d{2})/(?P<seq>\d+)$")
reg = reg.join(parts)
reg["seq"] = pd.to_numeric(reg["seq"])
mismatch = reg["fy_code"].notna() & (reg["fy_code"] != reg["fy"].str[2:])
reg.loc[mismatch, "format_issue"] = "FY in number does not match invoice date"

print(reg.loc[reg["format_issue"] != "", ["inv_no", "inv_date", "format_issue"]].to_string(index=False))

# 3. Uniqueness within the financial year.
dups = reg[reg.duplicated(["inv_no", "fy"], keep=False)]
print("\nrepeated within FY:\n", dups[["inv_no", "inv_date", "taxable"]].to_string(index=False))

# 4. Gaps per series and year, net of documented cancellations.
good = reg.dropna(subset=["seq"]).drop_duplicates(["inv_no", "fy"])
gaps = []
for (series, fy_code), g in good.groupby(["series", "fy_code"]):
    seen = set(g["seq"].astype(int))
    for n in range(min(seen), max(seen) + 1):
        num = f"{series}/{fy_code}/{n:04d}"
        if n not in seen and num not in cancelled:
            gaps.append(num)
print("\nunexplained gaps:", gaps)
assert gaps == ["GZB/26-27/0002"] and len(dups) == 2

Reading the findings

NOI/25-26/0143 is missing too, but it is in the cancellation register, so it is not reported. That is the difference between a gap check that produces noise and one that gets used: explained gaps drop out, and only the ones needing attention remain.

The financial year is computed from the invoice date, April to March, and compared with the year code inside the number. A 25-26 code on 1 April 2026 is exactly the reset failure. Because the rule makes uniqueness a per-year condition, the duplicate check also groups by financial year, so NOI/25-26/0001 and NOI/26-27/0001 do not count as a repeat.

Why the check should not fix anything

It is tempting to renumber the long or badly formatted invoices in the register so that the report looks clean. Don't. The invoice already issued to the customer carries the number printed on it, and the register must match that document. The check's job is to find problems early, ideally within a day of the invoice being raised, so that the format is corrected before many more invoices are issued in it. Any correction to issued documents is an accounting and tax decision.

For the repeated number, look at the two rows before deciding. Two identical rows with the same date and amount usually mean the register imported the bill twice; two different amounts under the same number mean two documents were issued with one number, which is more serious. For removing genuine duplicate rows without losing legitimate line items, see detect duplicate invoices without deleting legitimate line items.

Excel version for a small register

With invoice numbers in column A, dates in column B and the FY formula in column D:

Too long?    =LEN(A2)>16
Bad chars?   =SUMPRODUCT(--ISERROR(FIND(MID(A2,SEQUENCE(LEN(A2)),1),
                 "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789-/")))>0
FY           =IF(MONTH(B2)>=4, YEAR(B2), YEAR(B2)-1)
Repeated?    =COUNTIFS(A:A, A2, D:D, D2)>1

For gaps, sort by series and number, extract the numeric part into a column, and flag any row where it is not the previous row's number plus one within the same series.

Run it on a schedule

Run the check daily or weekly on the register exported from your billing system, and once more before the month's returns are prepared. Keep the output for each run. The same register also feeds the purchase-side matching in reconcile GSTR-2B with the purchase register. Controls like these, which catch problems before a return is filed rather than after, are a practical part of the finance analytics work in our Data Analyst programme.