path: root/design-documents/013-peer-to-peer-payments.rst

Design Doc 013: Customer-to-Customer Payments
#############################################

Summary
=======

This design document proposes an extension of the Taler protocol that allows
payments from customer-to-customer without a merchant.

Motivation
==========

To be usable as an electronic payment system with cash-like properties,
customers should be able to transfer money between themselves without
needing to setup anything beyond their wallet(s).


Requirements
============

* The protocol must permit transacting arbitrary amounts in any currency,
  as long as both parties trust the exchange involved.
* The control data for customer-to-customer payments should be small
  enough to fit into a QR code or short message (so ideally less than 64 bytes).
* No other direct communication channel between payer and payee should
  be required.
* The customer-to-customer payment must be possible without trusting the other
  party beyond the point where the money has been received by the payee.  Thus,
  sharing of coin private keys is not sufficient, we need transactional semantics
  resulting in exclusive control over the funds by the recipient.
* The customer-to-customer payment protocol must not allow users to circumvent income
  transparency.  That is, each customer-to-customer transaction must be visible
  on a KYCed transaction ledger (such as a bank account).
* The money received via a customer-to-customer payment must be usable for
  further Taler payments with minimal delay (after KYC).
* Two payment scenarios must be possible: (1) one where the payee first
  transmits a proposal to the payer (request-to-pay) that the payer
  accepts by making the payment, and (2) completely uni-directional
  payments where the payer includes a proposal with the payment and the
  payee accepts the proposal by taking the offered payment.
* If the payment fails (i.e. the receiver refuses to accept the money
  or the message is lost), the payer must automatically recover the
  funds (minus applicable fees) without the need for further communication.
* If funds flow back to the payer due to an aborted payment, it must be
  provable for the payer that these funds were not income but merely an aborted
  transaction.  Furthermore, in this case, no KYC should be required from the
  payer.
* If a payment would partially succeed, i.e. because the payer inadvertedly
  used some double-spent coins and some valid coins, this must fail before the
  uni-directional communication and be correctable payer-side.
* The usual properties of Taler (everything auditable, unlinkability,
  high-performance in terms of CPU, bandwidth, latency, storage
  requirements, and the ability to levy fees on every operation that
  is costly for the exchange) need to be preserved.



New Terminology
===============

* An ``account`` is a non-expiring reserve for which entity knowing the
  reserve private key has completed a KYC procedure sufficient to enable
  receiving income under that address.
* A ``purse`` is a public-private key pair where the public key is
  an exchange address from which any owners of an account
  can pick up an amount left at a ``purse`` assuming they know the purse
  private key.
* A ``wad`` is an exchange-to-exchange wire transfer that wires money
  into a group of accounts at the target exchange.


Proposed Solution
=================

A complete solution for customer-to-customer payments consists of three
possible operational flows.


Account creation and withdrawal
-------------------------------

1. The payee generates an account key, which also yields a
   ``payto://taler/$EXCHANGE_BASE_URL/$ACCOUNT_PUB``
   target address (for which the payee knows the corresponding
   account private key).
2. When withdrawing from an account, the exchange first checks if the
   customer has satisfied the KYC requirements. If not, the customer
   is redirected to a Web page where they can perform the necessary
   KYC operation.
3. For this, the exchange wire gateway is extended with a request to
   check the KYC status of a customer based on an ACCOUNT_PUB.
   Possible replies are in-progress and succeeded. An in-progress
   status should be accompanied with
   information how the customer may complete the KYC check.
4. A new exchange endpoint ``/account/$ACCOUNT_PUB/kyc``
   allows wallets to request a KYC for an
   ACCOUNT_PUB.  Such a request may include the requirement to pay
   a **KYC fee**.  The KYC fee may be charged to that account (if it exists),
   or could be waved if the account was established via a wire transfer
   from a partner bank.
5. If the account owner fails to perform the KYC check, the funds
   in an account remain inaccessible.  After a configurable duration,
   the funds may be considered forfeit and become the property of
   the exchange where the account is located.
6. When requesting an account's history (which can get quite long),
   the exchange only returns the last 3 months of data.  Requesting
   the full history requires paying an **account history fee**
   (which is not done via a 402, but simply charged to the account
   when requested; full account histories for accounts with an
   insufficient balance cannot be requested -- except of course
   the wallet could simply top up the account balance first, see below).

7. If the exchange has received a **deposit** or **merge** into an
   account, or received an inbound wire transfer from a **wad**
   matching the account (see below), it adds the respective amount(s)
   to the account's balance, allowing the KYC'ed customer to withdraw
   the funds.
8. The account history endpoint should also allow long-polling.
   Note that long-polling should be limited to short durations,
   as inbound transfers via ``taler-exchange-wirewatch`` cannot cause the long
   polling to be resumed, only transfers within the same exchange can benefit
   from long polling acceleration.



Payment into an unknown account
-------------------------------

1. If the payer does not know the payee's account, they can create
   a **purse** by computing a public-private key pair.
2. The payer then uses the ``/depost`` API to deposit coins into
   the purse, using ``payto://taler/$PURSE_PUB`` as the target
   address. Note that the lack of a hostname indicates that the
   target address is a purse.
3. The payer shares the purse's private key and the base URL
   of the exchange where the purse was created with the payee.
   This can be done using a ``taler://purse/BASE_URL/$PURSE_PRIV`` URL.
4. The payee uses the new ``/purse/$PURSE_PUB`` endpoint to retrieve
   the purse history, which includes all deposits and withdrawals
   involving the purse.
5. The payee can then POST to ``/purse/$PURSE_PUB/merge`` a
   request signed by the purse's private key to **merge** the
   funds into an account.  The signature is only over a salted
   hash of the account public key, thus avoiding disclosure of
   the account public key in the purse history.  The exchange
   processes the merge request akin to the logic for payments
   into known accounts, as detailed below, except that no
   **deposit fees** are charged at this time.  The exchange
   confirms the merge, allowing the payee to instantly affirm
   to the user that the amount is inbound (even if it may not
   be instantly available).


Payment directly into a known account at the same exchange
----------------------------------------------------------

1. If the payer knows the payee's account and uses the same exchange,
   they may perform an ordinary ``/deposit`` operation, paying
   the usual **deposit fee**.
2. The exchange detects the use of the ``taler`` wire method, and
   directly credits the target account.


Payment into known account at a remote exchange
-----------------------------------------------

1. If the payer knows the payee's account and uses a different exchange,
   they POST the coins to a new ``/wad/MASTER_PUB`` endpoint using the master
   public key of the target exchange, including a deposit permission,
   the base URL of the target exchange, and the target account,
   paying a **deposit fee** and a new **wad fee**.  The **wad fees** can
   be used to cover the cost of the exchange-to-exchange wire transfer.
2. The payer's exchange creates a **wad** grouping all wad requests, executing
   the transaction when either the **wad threshold** (maximum number
   of transactons aggregated per wad) or the **wad delay** (maximum
   delay for transfers) has been reached.
3. If the (aggregated) wire transfer fails (say the
   ``/wire`` endpoint of the payee exchange does not
   resolve to a valid bank account), the
   originating exchange automatically creates a full refund for
   all aggregated deposits and **purses** (**refund fees** apply).

   .. note::

      While the **refund fee** amount can be reused, these types of refunds
      are not approved by a merchant's signature. Thus, we will need
      a new message type in the coin history to represent these events.

4. The payee's exchange observes the wire transfer and uses a GET
   ``/wad/{MASTER_PUB}/{WTID}`` request to obtain details about the target
   accounts.
5. When the payer's exchange is requested to provide information about
   aggregated transfers under the WTID, it provides a signed list of
   account public keys and associated amounts that must add up to an
   amount below the total amount transferred. If they do not, the
   payee's exchange does not credit any of the accounts and instead
   preserves the bogus reply (to justify its inaction with its own
   auditor) and reports the issue to the auditor of the payer's exchange
   (keeping the received funds for future manual resolution).
6. ``taler-exchange-wirewatch`` and the Taler wire gateway API will
   need to be extended to allow passing inbound wire transfers with WTID
   and exchange base URL to the exchange. Furthermore, another tool
   is needed to lookup the **wad** data at remote exchanges.


C2C Payment Metadata
--------------------

The standard Taler Customer-to-Merchant payments always use a contract terms
JSON object to record the modalities of the payment and the resulting
obligations of a successful payment.

The contract terms concept does not directly carry over to
C2C payments, because:

* Either party may initiate the payment
* The payee does not have a merchant public key, and
  at the time of payment initiation, the payee account might not yet
  be known.
* There is no nonce that the customer generates and uses to prove that they
  uniquely "own" the contract terms.
* There is no negotiation of trusted auditors / exchanges possible or necessary.

Metadata for C2C payments can be exchanged in two ways:

1. Inline, as part of the payment request / payment offer.
2. As an annotation on the purse, encrypted with the purse's public key.  The
   exchange may charge for storing the annotation.
   In case of a "payment request purse", the fee should be
   taken from an account provided by the payee (which may even
   offer a limited number of free P2P payment initiations per time period).
   In case of a "payment offer purse", the fee should be taken from the
   coins deposited into the purse.


Examples
---------

Cross-exchange C2C payment request:

* Bob borrowed 15 Eur from Alice to buy a train ticket.  A few days later,
  Alice wants her money back.  She creates a request for payment in her wallet.
  The wallet creates a pouch for 15 Eur at the only exchange that Alice is currently
  using.  The wallet shows her a
  ``taler://payment-request/{EXCHANGE_URL}/{POUCH_PRIV}?i=Please+pay+me+Bob``
  link that she can share with Bob.  Bob receives the link and opens it with
  his Taler wallet.  Bob is using a different EUR exchange than Alice.  Bob's
  wallet makes a ``/wad`` request to his own exchange.  Shortly after, Alice's
  exchange receives the wad from Bob's exchange, and credits the money into
  Alice's pouch.


  * How does Bob find out if Alices exchange supports a wad transfer from
    Bob's exchange?
  * How does Bob get a "receipt" to prove that he paid Alice?  Is some signature
    from Bob's exchange enough, or should Bob also request a signature for Alice's
    exchange once the money has arrived?  Bob's wallet could poll
    the purse at Alice's exchange for a receipt.

Cross-exchange payment offer:

* Carol wants to send some money to Dave as a birthday gift.  Carol knows that
  Dave is using Taler, but she doesn't know which exchange he is using.  She
  opens her Taler wallet and initiates a P2P payment.  She sends the
  ``taler://payment-offer/{EXCHANGE_URL}/{POUCH_PRIV}?i=Happy+Birthday+from+Alice`` in an e-mail to Dave.
  Dave opens they link in the e-mail with his Taler wallet.
  Since Dave is using a different exchange than Alice, Dave's wallet
  issues a ``/wad`` request to Alice's exchange.  Shortly after,
  Dave's exchange receives the wad, and credits Dave's account with the money.


State machine for Purses
------------------------

.. code-block:: none

   // The "OPEN" starte state requires funding
   // the purse either from an account or from
   // a coin.
   -> OPEN

   // "Filled" means that it is filled with as many coins
   // as indicated in the creation request.
   OPEN -> FILLED

   // Canceled, all coins put into the purse are refunded
   FILLED -> CLOSED_REFUNDED

   // Sent via a wad to a different exchange
   FILLED -> PENDING_WAD

   // The associated wad transfer succeeded.
   PENDING_WAD -> CLOSED_WAD

   // Merged into an account at the same exchange
   FILLED -> CLOSED_MERGE


Additional considerations
-------------------------

* Creation of additional accounts per customer can
  be discouraged by asking for higher fees.
* The global transaction volume of one customer can be easily
  determined by authorities, which can then trigger further audits
  of the customer
* As a technically expensive but more water-tight measure, normal
  withdrawals from reserves could be disallowed.  Instead,
  a modified refresh protocol could ensure that whoever has knowledge
  of the account private key can also learn the private keys
  of coins withdrawn from that account, thereby removing
  Taler's "one-hop withdrawal loohole".


Alternatives
============

* The payer could directly give deposit permissions to the payee.
  This has two problems:

  1. The payer doesn't know the wire details of the payee.
     Thus we would need to introduce some "wildcard deposit permission",
     where the exchange allows any wire details on ``/deposit``.
  2. The payment information would be rather large, making it difficult
     to transfer via a QR code or short text message.

* Account history exceeding a configurable time limit (like 6 years)
  could be subject to garbage collection. However, doing so may be difficult to
  square with onboarding new auditors in the presence of existing
  accounts, as the auditors could then not reconstruct the account
  balances from cryptographic proofs.

* Accounts without KYC check could be eventually closed. However,
  even if the coins used to fill the account are refunded, it
  would be difficult to inform the originating wallet that the
  coins have received a refund.  Thus, it is likely simpler to
  declare the funds forfeit.


Drawbacks
=========

The overall changes required are not small:

* New **KYC fee**, **wad fee** and **account history fee**
  required in ``/keys`` endpoint (similar to closing and wire fees),
  requires some work across toolchain (offline signature, etc.)
* New ``taler`` wire method needs special case to possibly bypass
  (same exchange scenario, with long-poll trigger) the usual aggregation logic.
* New exchange table(s) required to store inbound amounts by account.
  Likely two tables, one for local exchange p2p and one for remote exchange p2p
  payments.
* New exchange table for purses required (for remote p2p payments).
* New exchange logic required to make ``transfers`` requests for purses
  (another separate process).
* New ``/account/$ACCOUNT_PUB/kyc`` endpoint required.
* New ``/purse/$PURSE_PUB/merge`` endpoint required.
* Additional tables to be verified by the auditor.
* ``taler-exchange-wirewatch`` needs to support receiving purses closures
  and exchange-to-exchange wire transfers with WTIDs.

Aside from implementation complexity, the solution has the following drawbacks:

* If a p2p payment failed (say the receiver lost the account private key),
  the customer's money can be forfeit.


Q / A
=====

* Q: Why are direct payments into accounts allowed?

  * A: Direct payments into accounts are used by the customer
    to fund the expenses for using the account.  They should not
    be used for payments between customers, as contract terms for
    the ``/deposit`` of coins can't be negotiated.  Furthermore,
    the sender of the payment can't be sure that the account of
    the sender is still valid.

* Q: Who "owns" a purse?  The payer of payee?

  * Both.  Ownership is shared.  Either the payer issues
    a refund on the purse, or the payee claims it by merging
    it with one of their accounts.

* Q: Are purses created with a pre-determined "capacity"?

  * A: Yes.  Otherwise there would be weird failure modes when the payee
    merges the purse before the payer fully deposited into it.

* Q: Are account public keys considered private or public data?

  * A: Private.  Making it public would make it too easy
    to accidentally send payments that nobody can receive, because
    the account is wrong/lost.  Furthermore, making the account
    public makes it easy to receive unsolicited payments without
    consent from the payee.  Since Taler should be as cash-like
    as possible, we might want to avoid that, especially
    for payees who aren't merchant.

* Q: Why do merchant payments not use purses?

  * A: Refunds aren't possible with purses, the customer can't prove that
    they own the contract terms (contract terms claiming isn't interactive).

    * All these issues could be addressed though.  But that
      would be a major "Taler 2.0 / 3.0" protocol change