add fees dd to index

1 files changed, 120 insertions, 0 deletions

diff --git a/design-documents/008-fees.rst b/design-documents/008-fees.rst
new file mode 100644
index 00000000..331d6cd7
--- /dev/null
+++ b/design-documents/008-fees.rst
@@ -0,0 +1,120 @@
+Design Doc 008: Fee Structure Metrics
+#####################################
+
+.. note::
+
+  This design document is currently a draft, it
+  does not reflect any implementation decisions yet.
+
+Summary
+=======
+
+We discuss criterea for evaluating an exchange's denomination and fee structure.
+
+Motivation
+==========
+
+Currently, users of a Taler wallet can't easily judge the fee structure of an
+exchange and what it will mean for them.  Thus we want to define some metrics
+that allow a user to make more informed decisions.
+
+Similarly, the fee structure metrics might be used by exchange operators
+as a senity check.
+
+An auditor might also enforce ranges on these metrics as a condition for
+auditing a denomination structure.
+
+Requirements
+============
+
+* Privacy:  A denomination structure that has too many denominations
+  will result in reduced anonymity.
+* Efficiency:  A denomination structure that has too few denominations
+  is inefficient.
+* Predictability:
+
+  * The amount of fees for an operation should not come
+    as a total surprise for the user.
+
+  * The user should know beforehand when they have to be online
+    to refresh their balance, and they should know how much
+    this will cost them.
+
+The metrics for the exchange should be advisory, i.e. an informed user should
+be able to accept the withdrawal anyway.
+  
+The exchange's business interests may be in conflict with (1) a transparency of
+cost for the user and and (2) restrictions on denomination/fee structures.
+Thus the metrics should still allow some degree of variability between
+exchanges.
+
+We make the assumption that wallet always prefer operations with better
+privacy.  For example, a single coin should only be used for at most one
+spend operation and at most one refresh operation.
+
+Proposed Solution
+=================
+
+The following yes/no criteria are applied to determine if a denomination/fee structure
+is *reasonable*:
+
+* For a denomination of value ``v``, either
+
+  (a) ``v`` must be the the smallest offered denomination, or 
+  (b) there must be another denomination with a value of at least ``v / 10``.
+
+Is there a relationship between the smallest denomination and the size of fees?
+
+Idea: When when only doing spends on a coin that are a multiple of the smallest spending amount,
+we constrain the number of coins that are refreshed into.
+
+When evaluating the e2e cost of a denomiation, look at:
+ * the cost of withdrawing the coin by itself and spending it fully, directly
+ * the cost of withdrawing the coin, directly refreshing it into the next smallest
+   fully fitting currency (or use greedy fit!) and add up the withdraw, refresh and re-withdraw fees.
+ * percentage is always computed with sum against the full coin's value
+ * smallest coins are an exception
+
+Actually, smallest spendable amount should be a multiple of the smallest coins, since
+we still need to deal with fees even at that level
+
+
+Refund fees should not be higher than if the user got a refund through
+a different channel than Taler and then purchased the product again.
+
+
+
+The following metrics are also defined:
+
+* Shortest time from withdrawal expiration to deposit expiration.
+* Upkeep, i.e. how much does it cost the customer to keep coins in their wallet
+  per time period, on average over a long period.
+* Assurance period, i.e. the time span for which the exchange has already announced
+  its fees.
+* Spending amount range.
+* End-to-end fee range.  This is the minimum/maximum cost to withdraw electronic cash
+  (within the spending range), spend some of it and then obtain change for the transaction.
+* Merchant contribution range.  This is the percentage of the end-to-end fees that the merchant
+  can cover if they choose to do so.
+
+
+Alternatives
+============
+
+* Users can manually study the fee structure.
+* The auditor could impose a fee structure and not
+  allow any variability between exchanges
+
+Drawbacks
+=========
+
+* The approach does not work well in some special-purpose deployments,
+  where the coin structure is tailored to the products of merchants,
+  and refreshing would never even happen.
+
+* The approach also does not consider more "creative" fee structures,
+  such as ones where coins that are valid for a longer time have higher
+  fees.
+
+Discussion / Q&A
+================