Strip the technology talk away. Bank deposits have been digital since the 1970s; reserves for decades. "Digital money" is not new. A central bank digital currency is new in exactly one respect: a CBDC is a liability of the central bank, in digital form, held directly by a class of holder that previously could not hold central-bank money at all. Everything else (tokens, ledgers, programmability, offline capability) is implementation.
What are retail and wholesale CBDCs?
That framing produces the taxonomy. Retail (general-purpose) CBDC extends central-bank money to households and firms, who today can hold it only as physical banknotes; a retail CBDC is a digital banknote. India's is the Digital Rupee (e₹), described in the RBI's Concept Note on Central Bank Digital Currency of 7 October 2022, with the retail pilot (e₹-R) launched 1 December 2022 and the wholesale pilot (e₹-W) 1 November 2022. Three design facts from the RBI's own FAQ determine everything downstream: e₹ "is legal tender and is the liability of Reserve Bank of India"; it is distributed two-tier: "RBI creates e₹ and issues them to banks and non-banks electronically. Banks and non-banks, thereafter, are responsible for facilitating onboarding and opening e₹ wallets for their customers"; and, decisively, "Given the cash-like features of e₹, there is no interest payable on the wallet balances" (RBI, Digital Rupee (e₹) – FAQs).
Advertisement
Wholesale CBDC extends tokenised central-bank money to financial institutions, which already hold central-bank money as reserves. Hence the sharpest criticism (it is not new money but a new format) and the honest defence: the format is the point, because tokenised reserves can settle atomically against tokenised assets on a shared ledger, enabling delivery-versus-payment and payment-versus-payment across asset classes and borders in a single conditional transaction. That is the BIS's tokenised unified ledger proposition, "a new type of financial market infrastructure" integrating "tokenised central bank reserves, commercial bank money and financial assets into the same venue" (BIS, Annual Economic Report 2025, Chapter III). Project Agorá is the live test: eight central banks (Bank of England, Federal Reserve Bank of New York, Banque de France for the Eurosystem, Bank of Japan, Bank of Korea, Bank of Mexico, Swiss National Bank, later the Bank of Canada) with more than 40 private financial institutions, prototyping tokenised correspondent banking; the BIS reported on 27 May 2026 that it had shown atomic, all-or-nothing cross-border settlement across jurisdictions with privacy at transaction and balance level, and would advance to real-value testing.
What does a CBDC do to bank balance sheets?
The balance-sheet consequence is the whole policy debate. When a household converts a bank deposit into retail CBDC, three balance sheets move at once. The household swaps a commercial-bank claim for a central-bank claim (total unchanged; the quality of its money rises to the top of the hierarchy). The commercial bank loses a deposit liability and pays by surrendering reserves, it shrinks on both sides, losing cheap funding and high-quality liquid assets simultaneously. The central bank's reserve liability falls and its CBDC-in-circulation liability rises by the same amount, its balance-sheet total is unchanged; only the composition of its liabilities changes. That last fact is what an AI gets wrong most often, and the cleanest test of whether someone understands CBDC.
The bank's problem is quantifiable, and Worked Example 8 quantifies it on a synthetic Indian bank. The two numbers to carry: a conversion of just 2.00% of total deposits costs 13.18 percentage points of liquidity coverage ratio, because the numerator takes the full blow and the denominator only a twentieth of it; and an 11.28% shift out of retail current and savings accounts exhausts that bank's entire regulatory buffer. Those numbers are the entire case for design limits.
The design levers that bound it, in order of power: no remuneration (RBI's choice, an unremunerated e₹ competes with cash, not with a savings account, which is why the FAQ sentence about interest is a policy statement, not a footnote); holding caps per wallet or person; tiered remuneration (a modest balance earns the policy rate, larger balances less or nothing, making CBDC useless as a bulk store of value); and conversion frictions at the two-tier boundary. Each trades usefulness against disintermediation risk; none is free.
Does a retail CBDC solve a real problem?
The honest assessment. Ask what problem a retail CBDC solves in a jurisdiction that already has an instant, ubiquitous, near-free retail payment system. India has UPI. The answers that survive scrutiny are narrow but real: a public payment option not dependent on any private intermediary's solvency or willingness to serve; offline capability where connectivity fails; programmability for targeted transfers with verifiable end-use; a monetary anchor if private tokenised money proliferates; and cross-border improvements at the wholesale layer, where the strongest evidence sits. The answers that do not survive are "faster payments" (UPI is already instant) and "financial inclusion" (a wallet still requires a device and onboarding). Be able to make both the case and the anti-case with sources, and be suspicious of any account that makes only one.
{
"skill": "Getting a fast, correct mechanical account of an unfamiliar piece of monetary plumbing (a CBDC design, a settlement system, a facility)",
"use": "Use a chat assistant to produce a first-pass balance-sheet description and a list of the primary documents that define the arrangement -- then verify the mechanism against those documents before you believe a single sentence of the description.",
"tools": ["Chat assistants — Claude, ChatGPT, Gemini", "Research/browsing assistants that can fetch the central bank's own publication", "Document/PDF AI run over a concept note or rulebook you have already downloaded"],
"prompt": "For <ARRANGEMENT>, produce: (1) a T-account for every party showing which balance sheet expands, shrinks or only changes composition; (2) the exact legal identity of the instrument (whose liability, redeemable into what, on what terms); (3) the settlement mechanism and the moment of finality; (4) the named primary documents — central bank concept note, rulebook, statute, or BIS report — where each of these is stated, with the section. Do NOT assert any mechanical claim you cannot point to a document for; mark anything you are inferring as an inference.",
"verify": "Open every named document and confirm the mechanism paragraph by paragraph — RBI's e-rupee FAQ and Concept Note, the Fed's facility pages, the BIS report, the CLS or CCIL rulebook. Trace each mechanical claim to the source; match entity, date and jurisdiction (a 2021 pilot design is not a 2026 design); recompute any balance-sheet figure yourself; date-check the source against the model's cutoff; cite the paragraph. Here the model's most dangerous output is a fluent WRONG MECHANISM, which no arithmetic check will catch — only the document will (Guardrail 1-5).",
"diy": "You must be able to draw the T-accounts yourself first — that is what section 5.8 and Worked Example 8 build. If you cannot say unaided whose balance sheet expands when a retail CBDC is issued, you cannot audit a model's answer to the same question.",
"market": "IN"
}