Skip to main content
Version: v0.47

ADR 050: SIGN_MODE_TEXTUAL: Annex 1 Value Renderers


  • Dec 06, 2021: Initial Draft
  • Feb 07, 2022: Draft read and concept-ACKed by the Ledger team.
  • Dec 01, 2022: Remove Object: prefix on Any header screen.
  • Dec 13, 2022: Sign over bytes hash when bytes length > 32.
  • Mar 27, 2023: Update Any value renderer to omit message header screen.


Accepted. Implementation started. Small value renderers details still need to be polished.


This Annex describes value renderers, which are used for displaying Protobuf values in a human-friendly way using a string array.

Value Renderers

Value Renderers describe how values of different Protobuf types should be encoded as a string array. Value renderers can be formalized as a set of bijective functions func renderT(value T) []string, where T is one of the below Protobuf types for which this spec is defined.

Protobuf number

  • Applies to:
    • protobuf numeric integer types (int{32,64}, uint{32,64}, sint{32,64}, fixed{32,64}, sfixed{32,64})
    • strings whose customtype is or
    • bytes whose customtype is or
  • Trailing decimal zeroes are always removed
  • Formatting with 's for every three integral digits.
  • Usage of . to denote the decimal delimiter.


  • 1000 (uint64) -> 1'000
  • "1000000.00" (string representing a Dec) -> 1'000'000
  • "1000000.10" (string representing a Dec) -> 1'000'000.1


  • Applies to cosmos.base.v1beta1.Coin.
  • Denoms are converted to display denoms using Metadata (if available). This requires a state query. The definition of Metadata can be found in the bank protobuf definition. If the display field is empty or nil, then we do not perform any denom conversion.
  • Amounts are converted to display denom amounts and rendered as numbers above
    • We do not change the capitalization of the denom. In practice, display denoms are stored in lowercase in state (e.g. 10 atom), however they are often showed in UPPERCASE in everyday life (e.g. 10 ATOM). Value renderers keep the case used in state, but we may recommend chains changing the denom metadata to be uppercase for better user display.
  • One space between the denom and amount (e.g. 10 atom).
  • In the future, IBC denoms could maybe be converted to DID/IIDs, if we can find a robust way for doing this (ex. cosmos:cosmos:hub:bank:denom:atom)


  • 1000000000uatom -> ["1'000 atom"], because atom is the metadata's display denom.


  • an array of coin is display as the concatenation of each coin encoded as the specification above, the joined together with the delimiter ", " (a comma and a space, no quotes around).
  • the list of coins is ordered by unicode code point of the display denom: A-Z < a-z. For example, the string aAbBcC would be sorted ABCabc.
    • if the coins list had 0 items in it then it'll be rendered as zero


  • ["3cosm", "2000000uatom"] -> 2 atom, 3 COSM (assuming the display denoms are atom and COSM)
  • ["10atom", "20Acoin"] -> 20 Acoin, 10 atom (assuming the display denoms are atom and Acoin)
  • [] -> zero


  • Applies to all repeated fields, except cosmos.tx.v1beta1.TxBody#Messages, which has a particular encoding (see ADR-050).
  • A repeated type has the following template:
<field_name>: <int> <field_kind>
<field_name> (<index>/<int>): <value rendered 1st line>
<optional value rendered in the next lines>
<field_name> (<index>/<int>): <value rendered 1st line>
<optional value rendered in the next lines>
End of <field_name>.


  • field_name is the Protobuf field name of the repeated field
  • field_kind:
    • if the type of the repeated field is a message, field_kind is the message name
    • if the type of the repeated field is an enum, field_kind is the enum name
    • in any other case, field_kind is the protobuf primitive type (e.g. "string" or "bytes")
  • int is the length of the array
  • index is one based index of the repeated field


Given the proto definition:

message AllowedMsgAllowance {
repeated string allowed_messages = 1;

and initializing with:

x := []AllowedMsgAllowance{"", ""}

we have the following value-rendered encoding:

Allowed messages: 2 strings
Allowed messages (1/2):
Allowed messages (2/2):
End of Allowed messages


  • Applies to all Protobuf messages that do not have a custom encoding.

  • Field names follow sentence case

    • replace each _ with a space
    • capitalize first letter of the sentence
  • Field names are ordered by their Protobuf field number

  • Screen title is the field name, and screen content is the value.

  • Nesting:

    • if a field contains a nested message, we value-render the underlying message using the template:
    <field_name>: <1st line of value-rendered message>
    > <lines 2-n of value-rendered message> // Notice the `>` prefix.
    • > character is used to denote nesting. For each additional level of nesting, add >.


Given the following Protobuf messages:

enum VoteOption {

message WeightedVoteOption {
VoteOption option = 1;
string weight = 2 [(cosmos_proto.scalar) = "cosmos.Dec"];

message Vote {
uint64 proposal_id = 1;
string voter = 2 [(cosmos_proto.scalar) = "cosmos.AddressString"];
reserved 3;
repeated WeightedVoteOption options = 4;

we get the following encoding for the Vote message:

Vote object
> Proposal id: 4
> Voter: cosmos1abc...def
> Options: 2 WeightedVoteOptions
> Options (1/2): WeightedVoteOption object
>> Weight: 0.7
> Options (2/2): WeightedVoteOption object
>> Weight: 0.3
> End of Options


  • Show the enum variant name as string.


See example above with message Vote{}.


  • Applies to google.protobuf.Any
  • Rendered as:
> <value rendered underlying message>

There is however one exception: when the underlying message is a Protobuf message that does not have a custom encoding, then the message header screen is omitted, and one level of indentation is removed.

Messages that have a custom encoding, including google.protobuf.Timestamp, google.protobuf.Duration, google.protobuf.Any, cosmos.base.v1beta1.Coin, and messages that have an app-defined custom encoding, will preserve their header and indentation level.


Message header screen is stripped, one-level of indentation removed:

> Proposal id: 4
> Vote: cosmos1abc...def
> Options: 2 WeightedVoteOptions
> Options (1/2): WeightedVoteOption object
>> Option: Yes
>> Weight: 0.7
> Options (2/2): WeightedVoteOption object
>> Option: No
>> Weight: 0.3
> End of Options

Message with custom encoding:

> 10uatom


Rendered using RFC 3339 (a simplification of ISO 8601), which is the current recommendation for portable time values. The rendering always uses "Z" (UTC) as the timezone. It uses only the necessary fractional digits of a second, omitting the fractional part entirely if the timestamp has no fractional seconds. (The resulting timestamps are not automatically sortable by standard lexicographic order, but we favor the legibility of the shorter string.)


The timestamp with 1136214245 seconds and 700000000 nanoseconds is rendered as 2006-01-02T15:04:05.7Z. The timestamp with 1136214245 seconds and zero nanoseconds is rendered as 2006-01-02T15:04:05Z.


The duration proto expresses a raw number of seconds and nanoseconds. This will be rendered as longer time units of days, hours, and minutes, plus any remaining seconds, in that order. Leading and trailing zero-quantity units will be omitted, but all units in between nonzero units will be shown, e.g. 3 days, 0 hours, 0 minutes, 5 seconds.

Even longer time units such as months or years are imprecise. Weeks are precise, but not commonly used - 91 days is more immediately legible than 13 weeks. Although days can be problematic, e.g. noon to noon on subsequent days can be 23 or 25 hours depending on daylight savings transitions, there is significant advantage in using strict 24-hour days over using only hours (e.g. 91 days vs 2184 hours).

When nanoseconds are nonzero, they will be shown as fractional seconds, with only the minimum number of digits, e.g 0.5 seconds.

A duration of exactly zero is shown as 0 seconds.

Units will be given as singular (no trailing s) when the quantity is exactly one, and will be shown in plural otherwise.

Negative durations will be indicated with a leading minus sign (-).


  • 1 day
  • 30 days
  • -1 day, 12 hours
  • 3 hours, 0 minutes, 53.025 seconds


  • Bytes of length shorter or equal to 35 are rendered in hexadecimal, all capital letters, without the 0x prefix.
  • Bytes of length greater than 35 are hashed using SHA256. The rendered text is SHA-256=, followed by the 32-byte hash, in hexadecimal, all capital letters, without the 0x prefix.
  • The hexadecimal string is finally separated into groups of 4 digits, with a space ' ' as separator. If the bytes length is odd, the 2 remaining hexadecimal characters are at the end.

The number 35 was chosen because it is the longest length where the hashed-and-prefixed representation is longer than the original data directly formatted, using the 3 rules above. More specifically:

  • a 35-byte array will have 70 hex characters, plus 17 space characters, resulting in 87 characters.
  • byte arrays starting from length 36 will be be hashed to 32 bytes, which is 64 hex characters plus 15 spaces, and with the SHA-256= prefix, it takes 87 characters. Also, secp256k1 public keys have length 33, so their Textual representation is not their hashed value, which we would like to avoid.

Note: Data longer than 35 bytes are not rendered in a way that can be inverted. See ADR-050's section about invertability for a discussion.


Inputs are displayed as byte arrays.

  • [0]: 00
  • [0,1,2]: 0001 02
  • [0,1,2,..,34]: 0001 0203 0405 0607 0809 0A0B 0C0D 0E0F 1011 1213 1415 1617 1819 1A1B 1C1D 1E1F 2021 22
  • [0,1,2,..,35]: SHA-256=5D7E 2D9B 1DCB C85E 7C89 0036 A2CF 2F9F E7B6 6554 F2DF 08CE C6AA 9C0A 25C9 9C21

address bytes

We currently use string types in protobuf for addresses so this may not be needed, but if any address bytes are used in sign mode textual they should be rendered with bech32 formatting


Strings are rendered as-is.

Default Values

  • Default Protobuf values for each field are skipped.


message TestData {
string signer = 1;
string metadata = 2;
myTestData := TestData{
Signer: "cosmos1abc"

We get the following encoding for the TestData message:

TestData object
> Signer: cosmos1abc


Boolean values are rendered as True or False.

[ABANDONED] Custom msg_title instead of Msg type_url

This paragraph is in the Annex for informational purposes only, and will be removed in a next update of the ADR.

Click to see abandoned idea.
  • all protobuf messages to be used with SIGN_MODE_TEXTUAL CAN have a short title associated with them that can be used in format strings whenever the type URL is explicitly referenced via the cosmos.msg.v1.textual.msg_title Protobuf message option.
  • if this option is not specified for a Msg, then the Protobuf fully qualified name will be used.
message MsgSend {
option (cosmos.msg.v1.textual.msg_title) = "bank send coins";
  • they MUST be unique per message, per chain


  • -> governance v1 vote

Best Pratices

We recommend to use this option only for Msgs whose Protobuf fully qualified name can be hard to understand. As such, the two examples above (MsgSend and MsgVote) are not good examples to be used with msg_title. We still allow msg_title for chains who might have Msgs with complex or non-obvious names.

In those cases, we recommend to drop the version (e.g. v1) in the string if there's only one version of the module on chain. This way, the bijective mapping can figure out which message each string corresponds to. If multiple Protobuf versions of the same module exist on the same chain, we recommend keeping the first msg_title with version, and the second msg_title with version (e.g. v2):

  • mychain.mymodule.v1.MsgDo -> mymodule do something
  • mychain.mymodule.v2.MsgDo -> mymodule v2 do something