Key Transparency Protocol

brendanmcmillion@gmail.com

linkerfelix@gmail.com

SEC KEYTRANS Working Group key transparency While there are several established protocols for end-to-end encryption, relatively little attention has been given to securely distributing the end-user public keys for such encryption. As a result, these protocols are often still vulnerable to eavesdropping by active attackers. Key Transparency is a protocol for distributing sensitive cryptographic information, such as public keys, in a way that reliably either prevents interference or detects that it occurred in a timely manner. About This Document The latest revision of this draft can be found at . Status information for this document may be found at . Discussion of this document takes place on the Key Transparency Working Group mailing list (), which is archived at . Subscribe at . Source for this draft and an issue tracker can be found at .

Introduction End-to-end encrypted communication services rely on the secure exchange of public keys to ensure that messages remain confidential. It is typically assumed that service providers correctly manage the public keys associated with each user's account. However, this is not always true. A service provider that is compromised or malicious can change the public keys associated with a user's account without their knowledge, thereby allowing the provider to eavesdrop on and impersonate that user. This document describes a protocol that enables a group of users to ensure that they all have the same view of the public keys associated with each other's accounts. Ensuring a consistent view allows users to detect when unauthorized public keys have been associated with their account, indicating a potential compromise. More detailed information about the protocol participants and the ways the protocol can be deployed can be found in .

Conventions and Definitions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here.

Presentation Language This document uses the TLS presentation language to describe the structure of protocol messages, but does not require the use of a specific transport protocol. As such, implementations do not necessarily need to transmit messages according to the TLS format and can choose whichever encoding method best suits their application. However, cryptographic computations MUST be done with the TLS presentation language format to ensure the protocol's security properties are maintained. An optional value is encoded with a presence-signaling octet, followed by the value itself if present. When decoding, a presence octet with a value other than 0 or 1 MUST be rejected as malformed. ; ]]>

Tree Construction A Transparency Log is a verifiable data structure that maps a label-version pair to some unstructured data such as a cryptographic public key. Labels correspond to user identifiers, and a new version of a label is created each time the label's associated value changes. KT uses a prefix tree to store a mapping from each label-version pair to a commitment to the label's value at that version. Every time the prefix tree changes, its new root hash and the current timestamp are stored in a log tree. The benefit of the prefix tree is that it is easily searchable and the benefit of the log tree is that it can easily be verified to be append-only. The data structure powering KT combines a log tree and a prefix tree, and is called the combined tree. This section describes the operation of prefix trees, log trees, and the combined tree structure, at a high level. More precise algorithms for computing the intermediate and root values of the trees are given in .

Terminology Trees consist of nodes, which have a byte string as their value. A node is either a leaf if it has no children, or a parent if it has either a left child or a right child. A node is the root of a tree if it has no parents, and an intermediate if it has both children and parents. Nodes are siblings if they share the same parent. The descendants of a node are that node, its children, and the descendants of its children. A subtree of a tree is the tree given by the descendants of a particular node, called the head of the subtree. The direct path of a root node is the empty list, and of any other node is the concatenation of that node's parent along with the parent's direct path. The copath of a node is the node's sibling concatenated with the list of siblings of all the nodes in its direct path, excluding the root. The size of a tree or subtree is defined as the number of leaf nodes it contains.

Log Tree Log trees store information in the chronological order that it was added, and are constructed as left-balanced binary trees. A binary tree is balanced if its size is a power of two and for any parent node in the tree, its left and right subtrees have the same size. A binary tree is left-balanced if for every parent, either the parent is balanced, or the left subtree of that parent is the largest balanced subtree that could be constructed from the leaves present in the parent's own subtree. Given a list of n items, there is a unique left-balanced binary tree structure with these elements as leaves. Note also that every parent always has both a left and right child.

A log tree containing five leaves. Log trees initially consist of a single leaf node. New leaves are added to the right-most edge of the tree along with a single parent node to construct the left-balanced binary tree with n+1 leaves.

Example of inserting a new leaf with index 5 into the previously depicted log tree. Observe that only the nodes on the path from the new root to the new leaf change. Leaves can have arbitrary data as their value, and are frequently referred to as "log entries" later in the document. The value of a parent node is always the hash of the combined values of its left and right children. Log trees are powerful in that they can provide both inclusion proofs, which demonstrate that a leaf is included in a log, and consistency proofs, which demonstrate that a new version of a log is an extension of a previous version. Inclusion and consistency proofs in KT differ from similar protocols in that proofs only ever contain the values of nodes that are the head of a balanced subtree. Whenever the value of the head of a non-balanced subtree is needed by a verifier, the prover breaks down the non-balanced subtree into the smallest-possible number of balanced subtrees and provides the value of the head of each. This allows verifiers to cache a larger number of intermediate values than would otherwise be possible, reducing the size of subsequent responses. As a result, an inclusion proof for a leaf is given by providing the copath values of the leaf with any non-balanced subtrees broken down as mentioned. The proof is verified by hashing the leaf value together with the copath values, re-computing the head values of non-balanced subtrees where needed, and checking that the result equals the root value of the log.

Illustration of an inclusion proof. To verify that leaf 2 is included in the tree, the prover provides the verifier with the values of leaf 2's copath. These nodes are marked by (X). When requesting a consistency proof, verifiers are expected to have retained the head values of the largest-possible balanced subtrees (these will later be defined as the "full subtrees") of the previous version of the log. A consistency proof then consists of the minimum set of node values that are necessary to compute the root value of the new version of the log from the values that the verifier retained.

Illustration of a consistency proof between a log with 5 and with 7 leaves respectively. The verifier is expected to already have the values (X), so the prover provides the verifier with the values of the nodes marked [X]. By combining these, the verifier is able to compute the new root value of the log.

Prefix Tree Prefix trees store a mapping between search keys and their corresponding values, with the ability to efficiently prove that a search key's value was looked up correctly. Each leaf node in a prefix tree represents a specific mapping from search key to value, while each parent node represents some prefix which all search keys in the subtree headed by that node have in common. The subtree headed by a parent's left child contains all search keys that share its prefix followed by an additional 0 bit, while the subtree headed by a parent's right child contains all search keys that share its prefix followed by an additional 1 bit. The root node, in particular, represents the empty string as a prefix. The root's left child contains all search keys that begin with a 0 bit, while the right child contains all search keys that begin with a 1 bit. A prefix tree can be searched by starting at the root node and moving to the left child if the first bit of a search key is 0, or the right child if the first bit is 1. This is then repeated for the second bit, third bit, and so on until the search either terminates at a leaf node (which may or may not be for the desired search key), or a parent node that lacks the desired child.

A prefix tree containing five entries. New key-value pairs are added to the tree by searching it according to the same process. If the search terminates at a parent without a left or right child, a new leaf is simply added as the parent's missing child. If the search terminates at a leaf for the wrong search key, one or more intermediate nodes are added until the new leaf and the existing leaf would no longer reside in the same place. That is, until we reach the first bit that differs between the new search key and the existing search key.

The previous prefix tree after adding the key-value pair: 01101 -> F. The value of a leaf node is the encoded key-value pair, while the value of a parent node is the hash of the combined values of its left and right children (or a stand-in value when one of the children doesn't exist). An inclusion proof is given by providing the leaf value, along with the value of each copath node along the search path. A non-inclusion proof is given by providing an abridged inclusion proof that follows the path for the intended search key, but ends either at a stand-in node or a leaf for a different search key. In either case, the proof is verified by hashing together the leaf with the copath values and checking that the result equals the root value of the tree.

Combined Tree Log trees are desirable because they can provide efficient consistency proofs to show verifiers that nothing has been removed from a log that was present in a previous version. However, log trees can't be efficiently searched without downloading the entire log. Prefix trees are efficient to search and can provide inclusion proofs to show verifiers that the returned search results are correct. However, it's not possible to efficiently prove that a new version of a prefix tree contains the same data as a previous version with only new values added. In the combined tree structure, based on , each label-version pair stored by a Transparency Log corresponds to a search key in a prefix tree. This prefix tree maps the label-version pair's search key to a commitment to the label's value at that version. To allow users to track changes to the prefix tree, a log tree contains a record of each version of the prefix tree along with the timestamp of when it was published. With some caveats, this combined structure supports both efficient consistency proofs and can be efficiently searched. Note that, while a Transparency Log implementation would likely maintain a single logical prefix tree, every modification to the prefix tree changes its root value, and the updated root value is then stored in the log tree. As part of the protocol, the Transparency Log is often required to perform lookups in different versions of the prefix tree. Different versions of the prefix tree are identified by the log entry where their root value was stored.

An example evolution of the log tree in the combined tree structure. Every new log entry added contains the timestamp T_n of when it was created and the new prefix tree root hash PT_n. / \ .------+----. / \ | / \ / \ /_____\ (T_n, PT_n) /_____\ (T_n, PT_n) (T_n+1, PT_n+1) ]]> It's also worth noting that an arbitrary number of label-version pairs can be added or removed from the prefix tree between subsequent log entries. That is, the Transparency Log is not constrained to making one change to the prefix tree per log entry. This allows changes to the prefix tree to be efficiently batched, and batching prefix tree modifications allows Transparency Logs to achieve significantly higher throughput than they otherwise would be able to.

Updating Views of the Tree As users interact with the Transparency Log over time, they will see many different root hashes as the contents of the log changes. It's necessary for users to guarantee that the root hashes they observe are consistent with respect to two important properties:

If root hash B is shown after root hash A, then root hash B contains all the same log entries as A with any new log entries added to the rightmost edge of A.
All log entries in the range starting from the rightmost log entry of A and ending at the rightmost log entry of B, have monotonically increasing timestamps.

The first property is necessary to ensure that the Transparency Log never removes a log entry after showing it to a user, as this would allow the Transparency Log to remove evidence of its own misbehavior. The second property ensures that all users have a consistent view of when each portion of the tree was created. As will be discussed in later sections, users rely on log entry timestamps to decide whether to continue monitoring certain labels and which portions of the tree to skip when searching. Disagreement on when portions of the tree were created can cause users to disagree on the value of a label-version pair, introducing the same security issues as a fork. Proving the first property, that the log tree is append-only, can be done by providing a consistency proof from the log tree. Proving the second property, that newly added log entries have monotonically increasing timestamps, requires establishing some additional structure on the log's contents.

Implicit Binary Search Tree Intuitively, the leaves of the log tree can be considered a flat array representation of a binary tree. This structure is similar to the log tree, but distinguished by the fact that not all parent nodes have two children.

A binary tree constructed from 14 entries in a log The implicit binary search tree containing n entries can be defined inductively. The index of the root log entry in the implicit binary search tree is the greatest power of two, minus one, that is less than the size of the log. That is i_root = 2^floor(log2(n)) - 1. The left subtree is the implicit binary search tree of size i_root, i.e. the implicit binary search tree for all elements with a smaller index than the root. The right subtree is the implicit binary search tree of size n-i_root-1, but offset with i_root+1. Initially, these will be all indices larger than the root. Users ensure that log entry timestamps are monotonic by enforcing that the structure of this search tree holds. That is, users check that any timestamp they observe in the root's left subtree is less than or equal to the root's timestamp, and that any timestamp they observe in the root's right subtree is greater than or equal to the root's timestamp, and so on recursively. Following this tree structure ensures that users can detect misbehavior quickly while minimizing the number of log entries that need to be checked. As an example, consider a log with 50 entries. Instead of having the root be the typical "middle" entry of 50/2 = 25, the root would be entry 31. As new log entries are added, users that interact with the Transparency Log will consistently verify the timestamps of other log entries against that of entry 31 until the log roughly doubles in size and a new root is established at entry 63. The protocol frequently refers to the frontier of the log, which is defined as the root log entry followed by the root's right child, the right child's right child, and so on until the rightmost log entry is reached. Using the same example of a log with 50 entries, the frontier would be entries: 31, 47, 49. Example code for efficiently navigating the implicit binary search tree is provided in .

Algorithm Users retain the following information about the last tree head they've observed:

The size of the log tree (that is, the number of leaves it contained).
The head values of the log tree's full subtrees. The full subtrees are the balanced subtrees which are as large as possible, meaning that they do not have another balanced subtree as their parent.
The log entries along the frontier.

When users make queries to the Transparency Log, they advertise the size of the last tree head they observed. If the Transparency Log responds with an updated tree head, it first provides a consistency proof to show that the new tree head is an extension of the previous one. It then also provides the following:

In the new implicit binary search tree, compute the direct path of the log entry with index size-1, where size is the tree size advertised by the user. Provide the timestamp of each log entry in the direct path whose index is greater than or equal to size.
The last of these log entries will lie on the new tree's frontier. From this log entry, compute the remainder of the frontier. That is, compute the log entry's right child, the right child's right child, and so on. Provide the timestamps for these log entries as well.

Users verify that the first timestamp is greater than or equal to the timestamp of the rightmost log entry they retained, and that each subsequent timestamp is greater than or equal to the one prior. This only requires users to verify a logarithmic number of the newly added log entries' timestamps and guarantees that two users with overlapping views of the tree will detect any violations. While retaining only the rightmost log entry's timestamp would be sufficient for this purpose, users retain all log entries along the frontier. The additional data is retained to make later parts of the protocol more efficient. The Transparency Log defines two durations: how far ahead and how far behind the current time the rightmost log entry's timestamp may be. Users verify this against their local clock at the time they receive the query response. For users that have never interacted with the Transparency Log before and don't have a previous tree head to advertise, the Transparency Log simply provides the log entries along the frontier. The user verifies that the timestamp of each is greater than or equal to the one prior, and that the rightmost timestamp is within the defined bounds of the user's local clock.

Binary Ladder A binary ladder is a series of lookups, producing a series of inclusion and non-inclusion proofs, from a single log entry's prefix tree. The purpose of a binary ladder varies depending on the exact context in which it is provided, but it is generally to establish some bound on the greatest version of a label that existed as of a particular log entry. All binary ladders are variants of the following series of lookups that exactly determines the greatest version of a label that exists:

First, version x of the label is looked up, where x is a consecutively higher power of two minus one (0, 1, 3, 7, ...). This is repeated until the first non-inclusion proof is produced.
Once the first non-inclusion proof is produced, a binary search is conducted between the greatest version that was proved to be included and the version that was proved to not be included. Each step of the binary search produces either an inclusion or non-inclusion proof which guides the search left or right until it terminates.

As an example, if the greatest version of a label that existed in a particular log entry was version 6, that would be established by the following: inclusion proofs for versions 0, 1, 3, a non-inclusion proof for version 7, then followed by inclusion proofs for versions 5 and 6. This series of lookups uniquely identifies 6 as the greatest version that exists, in the sense that the Transparency Log would be unable to prove a different greatest version to any user. While the description above may imply that the series of lookups is interactive, this is not the case in practice. Users may receive one or more binary ladders, corresponding to the same or different log entries, in a single query response. The Transparency Log's query response always contains sufficient information to allow users to predict the outcome of each lookup (inclusion or non-inclusion of a particular version) in the binary ladder. Example code for computing the versions of a label that go in a binary ladder is provided in .

Greatest-Version Search Users often wish to search for the "most recent" version, or the greatest version, of a label. Since label owners regularly verify that the greatest version is correctly represented in the log, this enables a relatively simple approach to searching. Users reuse the implicit binary search tree from to execute their search. This ensures that all users will check the same or similar log entries when searching for a label, allowing the Transparency Log to be monitored efficiently. This section additionally defines the concept of a distinguished log entry, which is any log entry that label owners are required to check for correctness. Given this, users can start their search at the rightmost distinguished log entry and only consider new versions which have been created since then.

Reasonable Monitoring Window Transparency Logs define a duration, referred to as the Reasonable Monitoring Window (RMW), which is the frequency with which the Transparency Log generally expects label owners to perform monitoring. Distinguished log entries are chosen according to the recursive algorithm below, such that there is roughly one per every interval of the RMW:

Take as input: a log entry, the timestamp of a log entry to its left, and the timestamp of a log entry to its right.
If the right timestamp minus the left timestamp is less than the Reasonable Monitoring Window, terminate the algorithm. Otherwise, declare that the given log entry is distinguished and then:
If the given log entry has a left child in the implicit binary search tree, recurse to its subtree by executing this algorithm with: the given log entry's left child, the given left timestamp, and the timestamp of the given log entry.
If the given log entry has a right child, recurse to its subtree by executing this algorithm with: the given log entry's right child, the timestamp of the given log entry, and the given right timestamp.

The algorithm is initialized with these parameters: the root node in the implicit binary search tree, the timestamp 0, and the timestamp of the rightmost log entry. Note that step 2 is specifically "less than" and not "less than or equal to"; this ensures correct behavior when the RMW is zero. This process for choosing distinguished log entries ensures that they are regularly spaced. Having irregularly spaced distinguished log entries risks either overwhelming label owners with a large number of them, or delaying consensus between users by having arbitrarily few. Distinguished log entries must reliably occur at roughly the same interval as the Reasonable Monitoring Window regardless of variations in how quickly new log entries are added. This process also ensures that distinguished log entries are stable. Once a log entry is chosen to be distinguished, it will never stop being distinguished. This ensures that, if a user looks up a label and checks consistency with some distinguished log entry, this log entry can't later avoid inspection by the label owner by losing its distinguished status.

Binary Ladder To perform a search, users need to be able to inspect individual log entries and determine the greatest version of the label that was present in the prefix tree at that time. Specifically, they need to be able to determine if the greatest version of the label was greater than, equal to, or less than their target version. This is accomplished by having the Transparency Log provide a binary ladder from the log entry. Binary ladders provided for the purpose of searching the tree are called search binary ladders and follow the series of lookups described in , but with two optimizations: First, the series of lookups ends after the first inclusion proof for a version greater than the target version, or the first non-inclusion proof for a version less than or equal to the target version. Providing additional lookups is unnecessary, since the user only needs to know whether the greatest version of the label that exists is greater than, equal to, or less than the target version, rather than its exact value. However, note that the binary ladder continues after receiving an inclusion proof for a version equal to the target version, as this is often needed to determine whether or not any versions greater than the target version exist. Second, depending on the context in which the binary ladder is provided, the Transparency Log may omit inclusion proofs for any versions where another inclusion proof for the same version was already provided in the same query response for a log entry to the left. Similarly, the Transparency Log may omit non-inclusion proofs for any versions of the label where another non-inclusion proof for the same version was already provided in the same query response for a log entry to the right. Whether or not these lookups are omitted is specified in context.

Algorithm The algorithm for performing a greatest-version search is described below as a recursive algorithm. It starts at the rightmost distinguished log entry, or the root of the implicit binary search tree if there are no distinguished log entries, and then recurses down the remainder of the frontier, each time starting back at step 1:

Obtain a search binary ladder from the current log entry where the target version is the claimed greatest version of the label, omitting redundant lookups.
Verify that the binary ladder terminates in a way that is consistent with the claimed greatest version of the label. That is, verify that every lookup for a version greater than the target version results in a non-inclusion proof. If this is the rightmost log entry, additionally verify that every lookup for a version less than or equal to the target version results in an inclusion proof.
If this is not the rightmost log entry, recurse to the log entry's right child.

The terminal log entry of the search is defined as the leftmost log entry inspected that contains the greatest version of the label. If the Transparency Log is deployed in Contact Monitoring mode and the terminal log entry of the search is to the right of the rightmost distinguished log entry, the user MUST monitor the label as described in .

Fixed-Version Search When searching the combined tree structure described in for a specific version of a label, users essentially perform a binary search for the first log entry where the prefix tree contained the target version of the label. This search may terminate early if the user discovers a log entry where the target version of the label is the greatest that exists, as this is assumed to have been verified by the label owner (discussed in ).

Maximum Lifetime A Transparency Log operator MAY define a maximum lifetime for log entries. If defined, it MUST be greater than zero and greater than the RMW. Whether a log entry is expired is determined by subtracting the timestamp of the log entry in question from the timestamp of the rightmost log entry and checking if the result is greater than or equal to the defined duration. A user executing a search may arrive at an expired log entry by either of two ways: The user may have inspected a log entry which is not expired and decided to recurse to the log entry's left child, which is expired. Alternatively, the root log entry might be expired, in which case the user would've started their search at an expired root log entry. Regardless of how the user arrived at the expired log entry, the user's next step is always to recurse to the log entry's right child (if one exists) without receiving a binary ladder. This allows the Transparency Log to prune large sections of the log tree and any versions of the prefix tree that are older than the defined maximum lifetime. Pruning is explained in more detail in .

Algorithm The algorithm for performing a fixed-version search is described below as a recursive algorithm. It starts with the root log entry, as defined by the implicit binary search tree, and then recurses to left or right children, each time starting back at step 1.

If the log entry is expired, recurse to the log entry's right child. If the log entry does not have a right child, proceed to step 6.
Obtain a search binary ladder from the current log entry for the target version, omitting redundant lookups as described in . Determine whether the binary ladder indicates a greatest version of the label that is greater than, equal to, or less than the target version.
If the binary ladder indicates a greatest version less than the target version (that is, if it contains a non-inclusion proof for a version less than or equal to the target version), then recurse to the log entry's right child. If the log entry does not have a right child, proceed to step 6.
If the binary ladder indicates a greatest version greater than the target version (that is, if it contains an inclusion proof for a version greater than the target version), then recurse to the log entry's left child. If the log entry does not have a left child, proceed to step 6.
If the binary ladder indicates a greatest version equal to the target version (that is, it contains inclusion proofs for all expected versions less than or equal to the target and non-inclusion proofs for all expected versions greater than the target), then:
1. If there are no expired log entries in the current log entry's direct path, then terminate the search successfully.
2. Otherwise, identify whether the log entry itself is distinguished, or whether there are any unexpired distinguished log entries in its direct path and to its left. If yes, terminate the search successfully. If no, terminate the search with an error indicating that the target version of the label is expired.
If this step is reached, the search has terminated without finding an unexpired log entry where the target version is the greatest that exists. In this case, out of all the log entries inspected, identify the leftmost one where the binary ladder indicated a greatest version greater than the target version.
1. If there is no such log entry, terminate the search with an error indicating that the target version of the label does not exist.
2. If any expired log entries were encountered in the search, and there are no unexpired distinguished log entries to the left of the identified log entry, terminate the search with an error indicating that the target version of the label is expired. Clients MUST NOT accept a proof where the identified log entry is itself the leftmost unexpired and distinguished log entry. Since the target version is not the greatest that exists in the log entry, it may not have been checked for correctness by the label owner.
3. Otherwise, look up the target version of the label in the log entry's prefix tree. If the result is a non-inclusion proof, terminate the search with an error indicating that the requested version of the label does not exist. If the result is an inclusion proof, terminate the search successfully.

The terminal log entry of the search is defined as the log entry that triggered step 5, or the log entry identified in step 6. If the Transparency Log is deployed in Contact Monitoring mode and the terminal log entry of the search is to the right of the rightmost distinguished log entry (defined in ), the user MUST monitor the label as described in .

Monitoring the Tree As new entries are added to the log tree, the search path that's traversed to find a specific version of a label may change. New intermediate nodes may be established between the search root and the terminal log entry, or a new search root may be created. The goal of monitoring a label is to efficiently ensure that, when these new parent nodes are created, they're created correctly such that searches for the same versions of a label continue producing the same results. Label owners MUST monitor their labels regularly, ensuring that past versions of the label are still correctly represented in the log and that any new versions of the label are permissible, alerting the user if not. If the Transparency Log is deployed in Contact Monitoring mode, then the users that looked up a label (either through a fixed-version or greatest-version search) are also sometimes required to monitor the label. Specifically, if a user looks up a label and the terminal log entry of their search is to the right of the rightmost distinguished log entry, the user MUST regularly monitor the label-version pair until its monitoring path intersects a distinguished log entry. That is, until a new distinguished log entry is established to its right and the two log entries are verified to be consistent. The purpose of this monitoring is to ensure that the label-version pair is not removed or obscured by the Transparency Log before the label owner has had an opportunity to detect it. If the Transparency Log is deployed with a Third-Party Auditor or Third-Party Manager, this monitoring is unnecessary assuming that either the Service Operator or the Third Party are honest. However, the user MAY still perform it to detect collusion between the Service Operator and the Third Party. If a user looks up a label and the terminal log entry of their search is either a distinguished log entry or to the left of any distinguished log entry, monitoring is never necessary. In this case, the only state that would be retained from the query would be the tree head, as discussed in . "Regular" monitoring SHOULD be performed roughly as frequently as the RMW and MUST, if at all possible, happen more frequently than the log entry maximum lifetime.

Binary Ladder Similar to the algorithm for searching the tree, the algorithms for monitoring the tree requires a way to prove that the greatest version of a label stored in a particular log entry's prefix tree is greater than or equal to a target version. The target version in this case is the version of the label that the user is monitoring. Unlike in a search though, users already know that the target version of the label exists and only need proof that there hasn't been an unexpected downgrade. Binary ladders provided for the purpose of monitoring are called monitoring binary ladders and follow the series of lookups described in , omitting any lookup for a version greater than the target version. As a result, all lookups in the binary ladder will result in an inclusion proof if the Transparency Log is behaving honestly.

Contact Algorithm To monitor a given label, users maintain a small amount of state: a map from a position in the log to a version counter. The version counter is the greatest version of the label proven to exist at that log position. Users initially populate this map by setting the position of the terminal log entry of their search to map to the version of the label they searched for. A map may track several different versions simultaneously if a user has been shown different versions of the same label. To update this map, users receive the most recent tree head from the Transparency Log and follow these steps for each entry in the map, from rightmost to leftmost log entry:

Determine if the log entry is distinguished. If so, leave the position-version pair in the map and move on to the next map entry.
Compute the ordered list of log entries to inspect:
1. Initialize the list by setting it to be the log entry's direct path in the implicit binary search tree based on the current tree size.
2. Remove all entries that are to the left of the log entry.
3. If any of the remaining log entries are distinguished, terminate the list just after the first distinguished log entry.
4. If the Contact Monitoring operation is being executed in the context of an Owner Monitoring operation, and the log entry in the monitoring map is equal to or to the right of the owner's advertised start position, skip inspecting the final distinguished log entry. This log entry will be inspected by the algorithm in .
For each log entry in the computed list, from left to right:
1. Check if a binary ladder from this log entry was already provided in the same query response. If so:
  1. If the previously provided binary ladder had a greater target version than the current map entry, then this version of the label no longer needs to be monitored. Remove the position-version pair with the lesser version from the map and move on to the next map entry.
  2. If it had a target version less than or equal to that of the current map entry, terminate and return an error to the user.
2. Obtain a monitoring binary ladder from this log entry where the target version is the version currently in the map. Verify that all expected lookups are present and all show inclusion.
3. If the above check fails, terminate and return an error to the user. Otherwise, remove the current position-version pair from the map and replace it with a new one for the position of the log entry that the binary ladder came from.

Once the map entries are updated according to this process, the final step of monitoring is to remove all mappings where the position corresponds to a distinguished log entry. All remaining entries will be non-distinguished log entries lying on the log's frontier. This algorithm works by progressively moving up the implicit binary search tree as new intermediate or root nodes are established, and verifying that they're constructed correctly. Once a distinguished log entry is reached and successfully verified, monitoring is no longer necessary and the corresponding entry is removed from the map. Users will often be able to execute the monitoring process, at least partially, with the output of a fixed-version or greatest-version search for the label. This may reduce the need for monitoring-specific requests. It is also worth noting that the work required to monitor several versions of the same label scales sublinearly because the direct paths of the different versions will often intersect. Intersections reduce the total number of entries in the map and therefore the amount of work that will be needed to monitor the label from then on.

Owner Algorithm Label owners initialize their state by providing the Transparency Log with a starting position coresponding to the log entry where they wish their ownership of the label to begin. This starting position MUST correspond to an unexpired distinguished log entry. The user then executes the following algorithm:

Compute the list of log entries to inspect: this list starts with the log entry at the requested starting position, followed by the log entries that are on the starting position's direct path and to its left, ending just before the first expired log entry.
Obtain the greatest version of the label that existed as of each of these log entries. If the label did not exist, no value is provided. Verify that each version is less than or equal to the one prior.
Obtain VRF proofs for version zero of the label and all other versions of the label that would appear in a search binary ladder where the target version was any of the versions given in step 2.
Obtain the commitment to the label's value at each version where a VRF proof was provided in step 3 and the version is understood to exist based on the information provided in step 2.
Obtain a search binary ladder from each log entry in the list computed in step 1 where the target version is the corresponding version given in step 2, or zero if no version was given, without omitting redundant lookups. Verify that each binary ladder terminates in a way that is consistent with the claimed greatest version of the label.

If any new versions of the label were created after the requested starting position, the label owner will need to process each of these individually as described in . Once the label owner has fully initialized their state, through the algorithm above and by processing any remaining new versions, they can begin regular monitoring. The label owner advertises to the Transparency Log the greatest version of the label that they're aware of and the rightmost distinguished log entry that they've verified is correct. For a number of subsequent distinguished log entries, the Transparency Log provides a binary ladder proving that no new unexpected versions of the label exist. This is described below as a recursive algorithm, starting with the root log entry:

If the current log entry is not distinguished, stop.
If the current log entry's index is less than or equal to that of the log entry advertised by the user:
1. If the current log entry has a right child, recurse to the right child.
2. Regardless of the outcome of step 1, stop.
If the current log entry has a left child, recurse to the left child. Afterwards, proceed to step 4.
If a stop condition has been reached, stop. From a user's perspective, the only stop condition is having consumed all of the Transparency Log's response. The Transparency Log may stop at this point if the greatest version of the label present at this log entry is greater than the version advertised by the user, or if a maximum output size has been reached.
Obtain a search binary ladder from the current log entry where the target version is the greatest version of the label expected to exist at this point according to the label owner's local state, without omitting redundant lookups. Verify that the binary ladder terminates in a way that is consistent with the expected version being the greatest that exists.
If the current log entry has a right child, recurse to the right child.

To avoid excessive load, the Transparency Log SHOULD limit the number of distinguished log entries that it provides binary ladders for in a single response. Users repeatedly query the Transparency Log until they detect that the above algorithm has either hit an unresolvable error or successfully reached the rightmost distinguished log entry. Since this algorithm checks only distinguished log entries for correctness, label owners use it in conjunction with the algorithm in to ensure the correctness of updates that occur between distinguished log entries.

Updating a Label As discussed in , a label owner is the authoritative source for a label's contents and must either initiate all changes to the label's value themself or at least be informed of changes afterwards. This section describes the mechanism by which label owners ensure that new versions of a label are inserted correctly into the Transparency Log. Label owners MUST follow this process for every new version of a label that is created after their ownership begins.

Algorithm Whenever a log entry is added to the Transparency Log that contains some new versions of a label, the Transparency Log informs the label owner of the following:

The index of the log entry where the new versions were inserted.
The commitment opening that was chosen for each new version of the label.
If the Transparency Log is deployed with a Third-Party Manager, the signature produced by the Service Operator over each new version.
VRF proofs for the following versions of the label:
- Compute the set of all versions that would be contained in a search binary ladder for the new greatest version of the label.
- If more than one new version of the label was created, additionally include each of these individual versions.
- Of the versions matching the two criteria above, omit any versions that would be contained in a search binary ladder for the previous greatest version of the label, as the label owner is expected to already know these VRF outputs. If there was no previous greatest version of the label, omit only the version zero.

To ensure that the new versions of the label were inserted correctly, the label owner considers the Transparency Log as it existed at two points in time: The first is the previous tree, which is defined as the log tree up to but excluding the log entry where the new versions were added. The second is the current tree, which is defined as the log tree as it is currently presented to the user, containing the new log entry and potentially other log entries to its right. Given this, the user executes the following algorithm:

Starting from the root log entry of the previous tree, proceed down the frontier of the previous tree and identify the first log entry that is not distinguished in the current tree. If there is no such log entry, skip to step 3.
Starting from the identified log entry, proceed down the remainder of the previous tree's frontier from left to right:
1. If a binary ladder would have already been received from this log entry in step 2.2 when processing a previous label update, skip this log entry.
2. Obtain a search binary ladder from this log entry where the target version is the previous greatest version of the label that existed, or 0 if no version of the label existed. Lookups that would be omitted in a greatest-version search for the label are also omitted here. Note that this means that lookups that would occur in log entries that were skipped by step 2.1 will still be omitted as if the log entries had been inspected.
3. Verify that the binary ladder terminates in a way that is consistent with the previous greatest version of the label being the greatest that exists.
If the log entry where the new versions were added is distinguished in the current tree, and if there are any new versions of the label that would not be looked up in a search binary ladder for the new greatest version, obtain a PrefixProof from the log entry for those versions in ascending order by version. Verify that all lookups result in an inclusion proof. If all checks succeed, the label owner retains the position and new greatest version of the label for later verification with the second algorithm of .
If the log entry where the new versions were added is not distinguished in the current tree, obtain a PrefixProof from it with lookups corresponding to a search binary ladder where the target version is the new greatest version of the label, omitting redundant lookups. Verify that all lookups terminate in a way that is consistent with the new greatest version of the label being the greatest that exists. Additionally, if there are any new versions of the label that were created in this log entry but that weren't included in the search binary ladder, obtain a second PrefixProof from the log entry with those lookups in ascending order by version. Verify that all lookups result in an inclusion proof. If all checks succeed, the label owner retains the position and new greatest version of the label for later verification with . The label owner additionally inserts a new entry into their contact monitoring map, discussed in , that maps the log entry to the new greatest version of the label.

In essence, this algorithm has two phases. The first phase is doing a greatest-version search in the previous tree. This is necessary to prevent forms of misbehavior where a Transparency Log might create a malicious version of a label, later change the corresponding value to something non-malicious, and try to only inform the label owner of the non-malicious value. The second phase is verifying how the new versions of the label were created in the claimed log entry. This depends on whether the log entry is distinguished or not to minimize redundant lookups with the second algorithm of .

Walking Distinguished Heads There are many cases in KT where it can be useful to allow a user to walk the most recently issued distinguished log entries. One particular case where this is useful is in deployments that rely on credentials, explained further in . It can also be helpful for new label owners, running the first algorithm of , in deciding what they want their "starting" log entry to be. And perhaps most importantly, walking the recent distinguished log entries gives users common reference points in the log tree and allows them to verify that they aren't being shown a fork.

Algorithm Walking the recent distinguished log entries is essentially done as a right-to-left depth-first search in the log tree. Users and the Transparency Log share an application-defined concept of what a "recent" distinguished log entry is, and enumerate all of the log entries that meet this definition. Users can optionally provide the Transparency Log with a stopping position corresponding to a point where the enumeration can stop early if hit. This is described below as a recursive algorithm, starting with the root log entry:

If the current log entry is not distinguished, stop.
If the current log entry has a right child, recurse to the right child.
If a stopping position was provided and the current log entry's position is less than or equal to the requested stopping position, stop.
If the current log entry doesn't meet the application's definition of "recent", stop.
If the current log entry has a left child, recurse to the left child.

This algorithm works with any definition of "recent" for distinguished log entries that is monotonic. However, the two simplest possible definitions are: a.) a distinguished log entry is "recent" if it is one of the n rightmost, or b.) a distinguished log entry is "recent" if subtracting its timestamp from the timestamp of the rightmost log entry yield a value less than some threshold. It is RECOMMENDED that applications choose one of these definitions.

Detecting Forks As discussed in , two of the primary ways that a user can detect the existence of a fork is through either an anonymous channel, or some form of direct peer-to-peer communication. Both of these approaches rely on exchanging root values of the log tree in a way where the Transparency Log can't reliably partition users. To compute the log tree root values to exchange, a user follows the algorithm in , using data provided through the same channel that the user typically interacts with the Transparency Log. The user then computes the log tree root value at each point where one of the recent distinguished log entries is the rightmost log entry. The exact number of distinguished log entries to use, and therefore the number of root values to compute, depends on the application. That said, using a larger number provides more resilience against clock skew and applications generally SHOULD compute at least two root values. Users then obtain the same information either by requesting it over their anonymous channel, or through direct peer-to-peer communication. The information exchanged can either be just the list of root values, if bandwidth is a particular concern, or a signed query response, if the ability to provide non-repudiable evidence of misbehavior is desired. The list of root values determined from the authenticated channel, and the list of root values received over the partition-resistent channel, are verified by checking that each list is either a prefix or a suffix of the other, having at least one root value in common. Example code for this is provided below:

Cryptographic Computations

Cipher Suites Each Transparency Log uses a single fixed cipher suite, chosen when it is initially created, that specifies the following primitives and parameters for cryptographic computations:

A hash algorithm
A signature algorithm
A Verifiable Random Function (VRF) algorithm
Nc: The size in bytes of commitment openings
Kc: A fixed string of bytes used in the computation of commitments

The hash algorithm is used to calculate intermediate and root values of hash trees. The signature algorithm is used for signatures from both the Service Operator and the Third Party, if one is present. The VRF is used for preserving the privacy of labels. Throughout the document, the following shorthands are used to denote different parameters of the current cipher suite:

Hash.Nh denotes the hash function's output length in bytes.
VRF.Nh denotes the VRF algorithm's output length in bytes.
VRF.Np denotes the VRF algorithm's proof size in bytes.

Cipher suites are represented with the CipherSuite type and are defined in .

Tree Head Signature The head of a Transparency Log, which represents its most recent state, is encoded as: ; } TreeHead; ]]> where tree_size is the number of log entries. If the Transparency Log is deployed in Third-Party Management mode, then the public key used to verify the signature belongs to the Third-Party Manager; otherwise the public key used belongs to the Service Operator. The signature itself is computed over a TreeHeadTBS structure, which incorporates the log's current state as well as long-term log configuration: ; opaque vrf_public_key<0..2^16-1>; select (Configuration.mode) { case contactMonitoring: case thirdPartyManagement: opaque leaf_public_key<0..2^16-1>; case thirdPartyAuditing: uint64 max_auditor_lag; uint64 auditor_start_pos; opaque auditor_public_key<0..2^16-1>; }; uint64 max_ahead; uint64 max_behind; uint64 reasonable_monitoring_window; optional maximum_lifetime; } Configuration; struct { Configuration config; uint64 tree_size; opaque root[Hash.Nh]; } TreeHeadTBS; ]]> The ciphersuite field contains the cipher suite for the Transparency Log, chosen from the registry in . The mode field specifies whether the Transparency Log is deployed in Contact Monitoring mode, or with a Third-Party Manager or Auditor. The signature_public_key field contains the public key to use for verifying signatures on the TreeHeadTBS structure. The vrf_public_key field contains the VRF public key to use for evaluating VRF proofs provided in the BinaryLadderStep.proof field described in . If the deployment mode specifies a Third-Party Manager, a public key is provided in leaf_public_key. This public key is used to verify the Service Operator's signature on modifications to the Transparency Log, as described in . If the deployment mode specifies a Third-Party Auditor, the maximum amount of time in milliseconds that the auditor may lag behind the most recent version of the Transparency Log is provided in max_auditor_lag. The position of the first log entry that the auditor started processing is provided in auditor_start_pos. A public key for verifying the auditor's signature on views of the Transparency Log is provided in auditor_public_key. The max_ahead and max_behind fields contain the maximum amount of time in milliseconds that a tree head may be ahead of or behind the user's local clock without being rejected. The reasonable_monitoring_window contains the Reasonable Monitoring Window, defined in , in milliseconds. If the Transparency Log has chosen to define a maximum lifetime for log entries, per , this duration in milliseconds is stored in the maximum_lifetime field. Finally, the root field contains the root value of the log tree with tree_size leaves.

Auditor Tree Head Signature In deployment scenarios where a Third-Party Auditor is present, the auditor's view of the Transparency Log is presented to users with an AuditorTreeHead structure: ; } AuditorTreeHead; ]]> Users verify an AuditorTreeHead with the following steps:

If the user advertised a previously observed tree head, verify that the tree_size of the AuditorTreeHead structure in the previous tree head (which may be from a different auditor) is greater than or equal to auditor_start_pos for the current auditor.
Verify that the timestamp of the rightmost log entry is greater than or equal to timestamp, and that the difference between the two is less than or equal to Configuration.max_auditor_lag.
Verify that tree_size is less than or equal to that of the TreeHead provided by the Transparency Log.
Verify signature as a signature over the AuditorTreeHeadTBS structure:

The config field contains the long-term configuration for the Transparency Log. The timestamp and tree_size fields match that of AuditorTreeHead. The root field contains the root value of the log tree when it had tree_size leaves.

Full Tree Head Verification Tree heads are presented to users on the wire as follows: The head_type field may be set to same if the user advertised a previously observed tree size in their request and the Transparency Log wishes to continue using this same tree head. Otherwise, head_type is set to updated and a new, more recent tree head is provided. Users verify a FullTreeHead with the following steps:

If head_type is same, verify that the user advertised a previously observed tree size and that the timestamp of the rightmost log entry of this tree is still within the bounds set by max_ahead and max_behind.
If head_type is updated:
1. If the user advertised a previously observed tree size, verify that TreeHead.tree_size is greater than the advertised tree size.
2. Verify TreeHead.signature as a signature over the TreeHeadTBS structure.
3. If there is a Third-Party Auditor, verify auditor_tree_head as described in .

Update Format The leaves of the prefix tree contain commitments which open to the value of a label-version pair, potentially with some additional information depending on the deployment mode of the Transparency Log. The contents of these commitments is serialized as an UpdateValue structure: ; }; } UpdateSuffix; struct { opaque value<0..2^32-1>; UpdateSuffix suffix; } UpdateValue; ]]> The value field contains the value associated with the label-version pair. In the event that Third-Party Management is used, the suffix field contains a signature from the Service Operator, using the public key from Configuration.leaf_public_key, over the following structure: ; uint32 version; opaque value<0..2^32-1>; } UpdateTBS; ]]> The value field contains the same contents as UpdateValue.value. Users MUST successfully verify this signature before consuming UpdateValue.value.

Commitment Commitments are computed with HMAC using the hash function specified by the cipher suite. To produce a new commitment, the application generates a random Nc-byte value called opening and computes: where Kc is a string of bytes defined by the cipher suite and CommitmentValue is specified as: ; uint32 version; UpdateValue update; } CommitmentValue; ]]> The Transparency Log MAY generate opening in a non-random way, such as deriving it from a secret key, as long as the result is indistinguishable from random to other participants. The Transparency Log SHOULD ensure that individual opening values can later be deleted in a way where they can not feasibly be recovered. This preserves the Transparency Log's ability to delete certain information in compliance with privacy laws, discussed further in .

Verifiable Random Function Each label-version pair corresponds to a unique search key in the prefix tree. This search key is the output of executing the VRF, with the private key corresponding to Configuration.vrf_public_key, on the combined label and version: ; uint32 version; } VrfInput; ]]>

Log Tree The value of a leaf node in the log tree is computed as the hash, with the cipher suite hash function, of the following structure: The timestamp field contains the timestamp that the leaf was created in milliseconds since the Unix epoch. The prefix_tree field contains the updated root value of the prefix tree after making any desired modifications. The value of a parent node in the log tree is computed by hashing together the values of its left and right children: where Hash denotes the cipher suite hash function.

Prefix Tree The value of a leaf node in the prefix tree is computed as the hash, with the cipher suite hash function, of the following structure: vrf_output contains the VRF output for the label-version pair and commitment contains the commitment to the corresponding UpdateValue structure. The value of a parent node in the prefix tree is computed by hashing together the values of its left and right children: If one of the children does not exist, an all-zero byte string of length Hash.Nh is used instead.

Tree Proofs

Log Tree In the interest of efficiency, KT combines multiple inclusion proofs and consistency proofs into a single batch proof. Recalling from the discussion in ,

Whenever the Transparency Log serves an inclusion proof for a leaf of the log tree, it provides the minimum set of head values from balanced subtrees that allows the user to compute the root value when combined with the leaf's value.
Whenever the Transparency Log serves a consistency proof, the user is expected to have retained the head values of the full subtrees of the previous version of the log. The Transparency Log provides the minimum set of head values from balanced subtrees that allows the user to compute the new root value when combined with the retained values.

These two proof types are composed together as such: considering the leaf values which will be proved included, and any node values the user is understood to have retained, the Transparency Log provides the minimum set of head values from balanced subtrees that allows the user to compute the root value when combined with the leaf and retained values. This proof is encoded as follows: ; } InclusionProof; ]]> The contents of the elements array is in left-to-right order: if a node is present in the root's left subtree, then its value is listed before the values of any nodes in the root's right subtree, and so on recursively. Batching together inclusion and consistency proofs creates an edge case that requires special care: when a user has requested a consistency proof, and also requested inclusion proofs for leaves located in one or more of the subtrees that the user has retained the head of. When this happens, the portion of the batch proof that shows inclusion for the leaves in these subtrees will itself be sufficient to recompute the retained head values. This makes the retained values redundant for the purpose of computing the new root value, which could result in the retained values being disregarded in a naive implementation. Users MUST verify that the computed value for the head of any such subtree matches the retained value to avoid accepting invalid proofs.

Prefix Tree A proof from a prefix tree authenticates that a search was done correctly for a given search key. Such a proof is encoded as: ; HashValue elements<0..2^16-1>; } PrefixProof; ]]> The results field contains the search result for each individual value, provided in the order requested. For PrefixProof structures that correspond to a binary ladder, this means the entries of results correspond directly with the lookups of the binary ladder. The result_type field of each PrefixSearchResult indicates what the terminal node of the search for that value was:

inclusion for a leaf node matching the requested search key.
nonInclusionLeaf for a leaf node not matching the requested search key. In this case, the terminal node is provided since it can not be inferred.
nonInclusionParent for a parent node that lacks the desired child.

The depth field indicates the depth of the terminal node of the search and is provided to assist proof verification. The root node of the prefix tree corresponds to a depth of 0, the root's children correspond to a depth of 1, and so on recursively. The elements array consists of the fewest node values that can be hashed together with the provided leaves to produce the root. The contents of the elements array is kept in left-to-right order: if a node is present in the root's left subtree, its value is listed before any values from nodes that are in the root's right subtree, and so on recursively. In the event that a node does not exist, an all-zero byte string of length Hash.Nh is listed instead. The proof is verified by hashing together the provided values, in the left/right arrangement dictated by the bits of the search keys, and checking that the result equals the root value of the prefix tree.

Combined Tree As users execute the algorithms defined in , , , , and , they inspect a series of log entries. For some of these, only the timestamp of the log entry is needed. For others, both the timestamp and a PrefixProof from the log entry's prefix tree are needed. This subsection defines a general structure, called a CombinedTreeProof, that contains the minimum set of timestamps and PrefixProof structures that a user needs for their execution of these algorithms. For the purposes of this protocol, the user always executes the algorithm to update their view of the tree as described in , followed immediately by one or more of the other algorithms. Proofs are encoded as follows: ; PrefixProof prefix_proofs<0..2^8-1>; HashValue prefix_roots<0..2^8-1>; InclusionProof inclusion; } CombinedTreeProof; ]]> The timestamps field contains the timestamps of specific log entries, and the prefix_proofs field contains search proofs from the prefix trees of specific log entries. There is no explicit indication as to which log entry the elements correspond to, as they are provided in the order that the algorithm the user is executing would request them. The elements of the prefix_roots field are, in left-to-right order, the prefix tree root hashes for any log entries whose timestamp was provided in timestamps but a search proof was not provided in prefix_proofs. If a log entry's timestamp is referenced multiple times by algorithms in the same CombinedTreeProof, it is only added to the timestamps array the first time. Additionally, when a user advertises a previously observed tree size in their request, log entry timestamps that the user is expected to have retained are always omitted from timestamps. This may result in there being elements of prefix_proofs that correspond to log entries whose timestamps are not included in timestamps. Users MUST verify that any such proof in prefix_proof is consistent with their retained prefix tree root hash for the log entry, due to the fact that the log entry will not be included in inclusion. If different algorithms in the same CombinedTreeProof require a search proof from the same log entry, the prefix_proofs array will contain multiple PrefixProof structures for the same log entry. Users MUST verify that all PrefixProof structures corresponding to the same log entry compute the same prefix tree root hash. Users processing a CombinedTreeProof MUST verify that the timestamps, prefix_proofs, and prefix_roots fields contain exactly the expected number of entries -- no more and no less. Additionally, users MUST verify that the timestamps explicitly included in timestamps, along with any retained timestamps, represent a monotonic series. That is, users verify that any given timestamp is greater than or equal to all observed timestamps to its left. Finally, the inclusion field contains the minimum set of node values from the log tree that would allow a user to compute:

The root value of the log tree, and
If an AuditorTreeHead was provided by the Transparency Log, the root value of the log tree when it had AuditorTreeHead.tree_size leaves,

from the following:

The values of all leaf nodes whose timestamp was provided in timestamps, and
If the user advertised a previously observed tree size in their request, any intermediate node values the user is expected to have retained.

Updating View For a user to update their view of the tree, the following is provided:

If the user has not previously observed a tree head, the timestamp of each log entry along the frontier.
If the user has previously observed a tree head, the timestamp of each log entry from the list computed in .

Greatest-Version Search For a user to search the combined tree for the greatest version of a label, the following is provided:

From each log entry along the frontier, starting from the log entry identified in , a PrefixProof corresponding to a search binary ladder.

Note that the frontier log entry timestamps are either already provided as part of updating the user's view of the tree, or are expected to have been retained by the user, and no additional timestamps are necessary to identify the starting log entry.

Fixed-Version Search For a user to search the combined tree for a specific version of a label, the following is provided:

For each log entry touched by the algorithm in :
- The log entry's timestamp.
- If the log entry is not expired, then a PrefixProof corresponding to a search binary ladder in the log entry's prefix tree.
If step 6.3 is reached, a second PrefixProof from the identified log entry specifically looking up the target version.

Contact Monitoring For a user to monitor a label in the combined tree, the following is provided:

For each entry in the user's monitoring map, from rightmost to leftmost log entry:
- The timestamp of each log entry on the path from the root to the parent of the log entry in the user's monitoring map, stopping if a non-distinguished log entry is established.
- From each log entry in the list computed in , a PrefixProof corresponding to a monitoring binary ladder.

Owner Initialization For a label owner to initialize their state to begin monitoring a label, the following is provided:

The timestamp of each log entry on the path from the root to the user's requested starting position.
For each log entry in the list computed in the first algorithm of , a PrefixProof corresponding to a search binary ladder.

The timestamps are provided to allow both the client and server to verify the precondition that the user's requested starting position is unexpired and distinguished.

Owner Monitoring For a label owner to perform regular monitoring, the following is provided:

The timestamp for each log entry that causes the second algorithm of to recurse either left or right.
For each log entry that reaches step 5 in the second algorithm of , a PrefixProof corresponding to a search binary ladder.

Updating a Label For a label owner to verify that some new versions of a label have been correctly inserted, the following is provided:

The timestamp of each log entry on the path from the root to the parent of the rightmost log entry of the previous tree, stopping if a non-distinguished log entry is established.
The timestamp of each log entry on the path from the root to the parent of the log entry where the new versions were inserted, stopping if a non-distinguished log entry is established.
Any PrefixProof structures that may be required by steps 2.2, 3, or 4 of the algorithm in .

Walking Distinguished Heads For a user to walk the most recent distinguished log entries, the following is provided:

The timestamp of each log entry that reaches step 2 of the algorithm in .

User Operations The basic user operations are organized as a request-response protocol between a user and the Transparency Log. Users MUST retain the most recent TreeHead and AuditorTreeHead they've successfully verified as part of any query response, and populate the last field of any subsequent query request with the tree_size of the last TreeHead. This ensures that all operations performed by the user return consistent results. Modifications to a user's state MUST only be persisted once the query response has been fully verified. Queries that fail full verification MUST NOT modify the user's protocol state in any way.

Search Users initiate a Search operation by submitting a SearchRequest to the Transparency Log containing the label that they wish to search for. Users can optionally specify a version of the label that they'd like to receive, if not the greatest one. last; opaque label<0..2^8-1>; optional version; } SearchRequest; ]]> In turn, the Transparency Log responds with a SearchResponse structure: commitment; } BinaryLadderStep; struct { FullTreeHead full_tree_head; select (SearchRequest.version) { case absent: uint32 version; }; opaque opening[Nc]; UpdateValue value; BinaryLadderStep binary_ladder<0..2^8-1>; CombinedTreeProof search; } SearchResponse; ]]> If no target version was specified in SearchRequest.version for a fixed-version search, the greatest version of the label is provided in SearchResponse.version. Each BinaryLadderStep structure contains information related to one version of the label in the binary ladder for the target version, listed in the same order that the versions are output by the algorithm in . The proof field contains the VRF proof. The commitment field contains the commitment to the label's value at that version. The commitment field is omitted only for versions of the label that don't exist and for the target version of the label, as the commitment to the target version is computed from opening and value. The search field contains the output of updating the user's view of the tree to match TreeHead.tree_size followed by either a fixed-version or greatest-version search for the requested label. Users verify a SearchResponse by following these steps:

Verify value as described in .
Verify that the expected number of entries is present in binary_ladder and that no commitment value is provided for the target version.
Compute the VRF output for each version of the label from the proofs in binary_ladder. Compute the commitment to the value of the target version with opening and value.
If SearchRequest.version is absent, verify the proof in search as described in . Otherwise, verify the proof in search as described in .
Compute a candidate root value for the tree from the proof in search.inclusion and any previously retained full subtrees of the log tree.
With the candidate root value for the tree, verify FullTreeHead as described in .

Contact Monitor If the Transparency Log is deployed in Contact Monitoring mode, users that looked up a label initiate a Contact Monitoring operation by submitting a ContactMonitorRequest to the Transparency Log containing information about the label they wish to monitor. last; opaque label<0..2^8-1>; MonitorMapEntry entries<0..2^8-1>; } ContactMonitorRequest; ]]> The Transparency Log verifies that each MonitorMapEntry in entries is sorted in ascending order by position, that no position or version is duplicate, and that position always lies on the direct path of the first log entry to contain the associated version of the label. In turn, the Transparency Log responds with a ContactMonitorResponse structure: The monitor field contains the output of updating the user's view of the tree to match TreeHead.tree_size followed by executing the algorithm in . Users verify a ContactMonitorResponse by following these steps:

Verify the proof in monitor as described in .
Compute a candidate root value for the tree from the proof in monitor.inclusion and any previously retained full subtrees of the log tree.
With the candidate root value for the tree, verify FullTreeHead as described in .

While access control decisions generally belong solely to the application, users of a Transparency Log deployed in Contact Monitoring mode MUST be able to monitor versions of a label they previously looked up, even if they would no longer be allowed to make the same query. One simple way for a user to prove that they were previously allowed to look up a particular version of a label would be for them to provide the corresponding commitment opening.

Owner Initialization Users that wish to initialize their owner state for a label submit an OwnerInitRequest to the Transparency Log containing the label they wish to take ownership of and their desired starting log entry. last; opaque label<0..2^8-1>; uint64 start; } OwnerInitRequest; ]]> The Transparency Log verifies that start identifies a log entry that is unexpired and distinguished. In turn, the Transparency Log responds with an OwnerInitResponse structure: ; BinaryLadderStep binary_ladder<0..2^16-1>; CombinedTreeProof init; } OwnerInitResponse; ]]> The greatest_versions field contains the greatest version of the label that exists in each log entry computed in step 1 of the first algorithm in , ending at the first log entry where the label doesn't exist. The binary_ladder field contains one entry for each version computed in step 3 of the first algorithm in , in ascending order by version. If a version of the label exists, the commitment to the label's value at that version is provided in the commitment field of the BinaryLadderStep. The init field contains the output of updating the user's view of the tree to match TreeHead.tree_size followed by executing the first algorithm in . Users verify an OwnerInitResponse by following these steps:

Verify that the number of entries in greatest_versions is less than or equal to the number of log entries in the computed list, and that greatest_versions is in descending order.
Verify that the expected number of entries is present in binary_ladder and that a commitment is provided for each version in greatest_versions. (Note that the existence of a version does not require the existence of all lesser versions.) Compute the VRF output for each version of the label from the proofs in binary_ladder.
Verify the proof in init as described in the first algorithm of .
Compute a candidate root value for the tree from the proof in init.inclusion and any previously retained full subtrees of the log tree.
With the candidate root value for the tree, verify FullTreeHead as described in .

Owner Monitoring Users that wish to monitor a label they own submit an OwnerMonitorRequest to the Transparency Log containing the label they wish to monitor and their current monitoring state. last; opaque label<0..2^8-1>; MonitorMapEntry entries<0..2^8-1>; uint64 start; optional greatest_version; } OwnerMonitorRequest; ]]> The entries list contains the label owner's monitoring state for versions of the label that have not yet been proven to be included in a distinguished log entry. The start field contains the rightmost distinguished log entry that the label owner has verified as correct. The greatest_version field contains the greatest version of the label that the label owner is aware of, if any. The Transparency Log verifies an OwnerMonitorRequest by following these steps:

Verify that each MonitorMapEntry in entries is sorted in ascending order by position, that no position or version is duplicate, and that position always lies on the direct path of the first log entry to contain the associated version of the label.
Verify that start is less than the size of the tree.
Verify that greatest_version is either absent, or less than or equal to the greatest version of the label. Verify that, if some version of the label existed at start, that greatest_version is present and greater than or equal to that version.

In turn, the Transparency Log responds with an OwnerMonitorResponse structure: The monitor field contains the output of updating the user's view of the tree to match TreeHead.tree_size, followed by executing the algorithm in , followed by executing the second algorithm in . Users verify an OwnerMonitorResponse by following these steps:

Verify the proof in monitor as described in and the second algorithm of .
Compute a candidate root value for the tree from the proof in monitor.inclusion and any previously retained full subtrees of the log tree.
With the candidate root value for the tree, verify FullTreeHead as described in .

Updating a Label Users that wish to create one or more new versions of a label they own submit an UpdateRequest to the Transparency Log containing the label and the new values to store. ; } LabelValue; struct { optional last; opaque label<0..2^8-1>; optional greatest_version; LabelValue values<0..2^8-1>; } UpdateRequest; ]]> The greatest_version field contains the greatest version of the label that the user is aware of, if any. The values field contains the values for the new versions of the label that the user would like to create. The values field MAY be empty to indicate that the user only wishes to be informed of already-existing versions of the label greater than greatest_version, rather than create any new ones. If greatest_version matches the current greatest version of the label that exists and values is non-empty, the Transparency Log inserts the new versions of the label into the prefix tree and sequences a new log entry. Version counters are assigned sequentially and in the same order as given in values. If the greatest_version field does not match the current greatest version of the label that exists, values is disregarded. The Transparency Log returns an UpdateResponse structure: ; UpdateInfo info<0..2^8-1>; BinaryLadderStep binary_ladder<0..2^8-1>; CombinedTreeProof update; } UpdateResponse; ]]> The position field contains the index of the log entry where the new versions of the label were inserted. This may or may not be the rightmost log entry. If greatest_version is less than the current greatest version of the label, UpdateResponse.values contains the values of all of the immediately subsequent versions of the label that were created in the log entry position, in ascending order by version. If UpdateResponse.values is non-empty, this indicates that UpdateRequest.values was disregarded by the Transparency Log. Inversely, if UpdateResponse.values is empty, this indicates that all of the new versions of the label in UpdateRequest.values were successfully created and no versions of the label other than those in UpdateRequest.values were created in the log entry position. The info field contains an UpdateInfo for each new version of the label created in the log entry position. The entries of info correspond to the entries of UpdateResponse.values if it is non-empty, and to the entries of UpdateRequest.values otherwise. The opening field of an UpdateInfo structure contains the commitment opening that was chosen for the version of the label and, if in Third-Party Management mode, the suffix field contains the Service Operator's signature over the new value. The binary_ladder field contains VRF proofs for the versions specified in in ascending order by version. If a version of a label exists and is less than or equal to the user's advertised greatest_version, the commitment to the label's value at that version is provided in the commitment field of the BinaryLadderStep. The update field contains the output of updating the user's view of the tree to match TreeHead.tree_size, followed by executing the algorithm in . Users verify an UpdateResponse by following these steps:

Verify that position is to the right of where the previous greatest version of the label was inserted. If this is the first version inserted since the user became the label owner, verify that position is to the right of the starting position chosen in .
If UpdateResponse.values is non-empty, verify that the length of info is equal to that of UpdateResponse.values. Otherwise, verify that the length of info is equal to that of UpdateRequest.values. Regardless, verify that info is non-empty.
If the Transparency Log is deployed with a Third-Party Manager, verify that the signatures provided in each element of info are valid.
Verify that the expected number of entries is present in binary_ladder, that all of the proofs properly evaluate into a VRF output, and that no commitment is provided for any version greater than greatest_version.
Verify the proof in update as described in .
Compute a candidate root value for the tree from the proof in update.inclusion and any previously retained full subtrees of the log tree.
With the candidate root value for the tree, verify FulTreeHead as described in .

The Transparency Log MAY provide multiple UpdateResponse structures in response to a single UpdateRequest, each corresponding to a subsequent position where more new versions of the label were inserted. These UpdateResponse structures are processed by clients serially as if an UpdateRequest with the following parameters had been sent:

last set to the tree size of the previously processed UpdateResponse
label set to the same value as the previous UpdateRequest
greatest_version incremented to cover the versions created by the previously processed UpdateResponse
values remains unchanged until the first UpdateResponse with an empty values field is received, and is empty from then on.

Walking Distinguished Heads Users that wish to walk the recently issued distinguished log entries submit a DistinguishedRequest to the Transparency Log, optionally containing their requested stopping position. last; optional stop; } DistinguishedRequest; ]]> In turn, the Transparency Log responds with a DistinguishedResponse structure: Users verify a DistinguishedResponse by following these steps:

Verify the proof in distinguished as described in .
Compute a candidate root value for the tree from the proof in distinguished.inclusion and any previously retained full subtrees of the log tree.
With the candidate root value for the tree, verify FullTreeHead as described in .

Credentials Credentials are proofs that are designed to be sent directly between users and verified without direct interaction with the Transparency Log. They are frequently useful in applications where anonymity is important, as they generally prevent users from needing to make direct requests to the Transparency Log regarding their contacts. Credentials are encoded as follows: ; uint32 version; opaque opening[Nc]; UpdateValue value; BinaryLadderStep binary_ladder<0..2^8-1>; uint64 position; CredentialType credential_type; select (Credential.credential_type) { case standard: PrefixProof distinguished; case provisional: TreeHead tree_head; CombinedTreeProof search; }; } Credential; ]]> In essence, credentials in KT are implemented as a greatest-version search for a user's label. As such, the fields version, opening, value, and binary_ladder are the same as they would be in a SearchResponse for a greatest-version search, described in . The position field contains the position of a recently issued distinguished log entry, and this log entry will be the basis for verifying the credential. Applications define their own policy for what constitutes a "recently issued" distinguished log entry. Users can learn of and retain all of the recently issued distinguished log entries by making a DistinguishedRequest, or through the process of monitoring labels they own. Once a distinguished log entry no longer meets the application's definition of "recent", any credentials relying on the log entry are considered expired. The credential_type field specifies whether the credential is of the standard type, meaning that the target label-version pair is included in a distinguished log entry, or is of the provisional type, meaning that the label-version pair is not yet included in a distinguished log entry. Regardless of credential type, users start verification of a credential by following these steps:

Verify that the user is aware of a recently issued distinguished log entry at position. To preserve user anonymity, implementations MUST NOT use a verification failure at this point as a prompt to request an updated tree head from the Transparency Log.
Verify value as described in .
Verify that the expected number of entries is present in binary_ladder and that no commitment value is provided for the target version.
Compute the VRF output for each version of the label from the proofs in binary_ladder. Compute the commitment to the value of the target version with opening and value.

Additional verification based on credential type is described in the following two subsections.

Standard If the credential type is standard, the distinguished field is present and contains lookups corresponding to a search binary ladder for the target version of the label in a recently issued distinguished log entry. Users follow these steps to verify a standard credential:

Verify that the binary ladder in distinguished terminates in a way that is consistent with version being the greatest version of the label that exists.
Verify that the prefix tree root value produced by evaluating distinguished matches that of the recently issued distinguished log entry at position.

Provisional If the credential type is provisional, the tree_head and search fields are present. The tree_head field contains a signature from the Transparency Log over a view of the tree that, at minimum, includes the log entry at position. The search field contains the output of updating the user's view of the tree from a tree size of position+1 to TreeHead.tree_size, followed by a greatest-version search for label. Users follow these steps to verify a provisional credential:

Verify that TreeHead.tree_size is greater than position.
Verify the proof in search as described in . Verify that the terminal log entry of the search is to the right of position.
Compute a candidate root value for the tree from the proof in search.inclusion and the retained full subtrees of the log tree.
With the candidate root value for the tree, verify TreeHead.signature as described in .

The verification process explicitly rejects a provisional credential if a standard credential could have been produced from the same log entry position. However, when generating a credential, users MAY choose to output a provisional credential from an older distinguished log entry rather than a standard credential from a newer distinguished log entry. This can allow users to accommodate reasonable delays in the distribution of newly issued distinguished log entries. Applications MAY define a policy restricting provisional credentials to a subset of recently issued distinguished log entries. Verifying a provisional credential MUST NOT have any effect on the state used for the user's direct interactions with the Transparency Log, or on the verification of other credentials (even for the same label). In particular, the view of the tree presented in a provisional credential MUST NOT cause a user to change its view of the tree for any other purpose. Successfully verified provisional credentials are retained by the receiving user until the sender provides a CredentialUpdate structure: The position field identifies the first distinguished log entry to the right of the terminal log entry of the search. The proof in monitor then contains the output of executing the algorithm in . The proof in monitor.inclusion is computed with a tree size of position+1 and assumes that the user has retained the full subtrees of the log tree with size Credential.tree_head.tree_size. Users follow these steps to verify a CredentialUpdate:

Verify that the user is aware of a recently issued distinguished log entry at position. Again, a verification failure at this point MUST NOT prompt a request for an updated tree head from the Transparency Log. Implementations MAY retain the CredentialUpdate and re-attempt verification at a later time.
Verify that, according to the user's retained state, position is the first distinguished log entry to the right of the terminal log entry of the search done in Credential.search.
Verify the proof in monitor as described in .
Compute a candidate root value for the tree from the proof in monitor.inclusion. Verify that this matches the user's retained state for the distinguished log entry at position.

If a user verifies a provisional credential and doesn't receive a CredentialUpdate from the sender before the credential expires, the user MUST fetch and verify a CredentialUpdate from the Transparency Log themself. The user SHOULD attempt to fetch the CredentialUpdate over an anonymous channel, to preserve the anonymity of the sending user as much as possible. If the user is still unable to obtain or verify a CredentialUpdate structure for a provisional credential, this SHOULD be reported to the user as evidence of potential misbehavior by the Transparency Log.

Detecting Forks In a user's typical interactions with a Transparency Log, every response is required to prove that it comes from the same view of the log tree as every prior response. However, this model doesn't work for credential verification. Since credentials are provided in a non-interactive and peer-to-peer manner, when a user is creating a credential they have no way to know what other credentials (and therefore which tree heads) the receiving user(s) may want to check consistency with. In fact, trying to convey this information to the user creating a credential risks violating the anonymity of the receiving user's other contacts. KT addresses this by using the recently issued distinguished log entries as common reference points for the state of the log tree. Users compute and retain the full subtrees of the log tree at each point where a recently issued distinguished log entry is the log's rightmost log entry. This retained state is then used either implicitly or explicitly in credential verification. In particular, when a provisional credential is received, the proof in Credential.search is constructed as if the user advertised a previously observed tree size of Credential.position+1. As a result, computing the candidate root value for the log tree requires incorporating the user's retained full subtrees for the log tree up to this point. This prevents the user from accepting a view of the log that forked at any point prior to the log entry at Credential.position. Once a CredentialUpdate is received for the credential, it will necessarily include a proof of inclusion for the first distinguished log entry that's to the right of the search's terminal log entry. Users take this proof, compute the full subtrees of the log tree just up to the point of the subsequent distinguished log entry, and verify that it matches their retained state. This ensures that the terminal log entry of the search, containing the value of the label that the user consumed, is also contained in a fully-consistent view of the log tree.

Third Parties Third-Party Management and Third-Party Auditing are two deployment modes that require the Transparency Log to delegate part of its operation to a third party. Users are able to run more efficiently as long as they can assume that the Transparency Log and the Third Party won't collude to trick them into accepting malicious results.

Management With the Third-Party Management deployment mode, a third party is responsible for the majority of the work of storing and operating the Transparency Log. The Service Operator serves only to enforce access control, authorize the addition of new versions of labels, and prevent the creation of forks by the Third-Party Manager. Critically, the Service Operator is trusted to ensure that only one value for each version of a label is authorized. All user queries specified in are initially sent by users directly to the Service Operator and are forwarded to the Third-Party Manager if they pass access control. While other operations are forwarded by the Service Operator unchanged, UpdateRequest structures are forwarded to the Third-Party Manager with the Service Operator's signature over each new value: last; opaque label<0..2^8-1>; optional greatest_version; UpdateValue values<0..2^8-1>; uint32 signed_version; } ManagerUpdateRequest; ]]> This structure is the same as UpdateRequest, except that the values field is an array of UpdateValue structures with the Service Operator's signature over each value, computed as described in . This structure also contains a signed_version field. If values is empty, the signed_version field is always zero. If values is non-empty, the signed_version field contains the version that was used in the computation of the Service Operator's signature over the first element of values. If values is non-empty and signed_version is greater than the next greatest version of the label to be created, the Manager MUST insert a sufficient number of "dummy" entries into the Transparency Log's prefix tree such that the next greatest version of the label to be created is signed_version, and sequence at least one new log entry containing the updated prefix tree root, before creating the versions requested in the ManagerUpdateRequest. These dummy entries in the prefix tree map the VRF output for the label-version pair to an all-zero commitment. As a result, the Manager will also need to provide an UpdateResponse to the user that covers the dummy versions of the label before providing an UpdateResponse that covers the requested versions. If signed_version is less than the next greatest version of the label to be created, this generally indicates a bug in the Service Operator and the request MUST be rejected by the Manager.

Auditing With the Third-party Auditing deployment mode, the Service Operator obtains signatures from a Third-Party Auditor attesting to the fact that the Service Operator is constructing the tree correctly. These signatures are provided to users along with the responses to their queries. For each new log entry that the Service Operator adds to the log, it produces a corresponding AuditorUpdate structure and sends this to the Third-Party Auditor. The auditor MUST receive and successfully verify an AuditorUpdate structure for a log entry before providing the Service Operator with an AuditorTreeHead structure whose tree_size field would include that log entry. ; PrefixLeaf removed<0..2^16-1>; PrefixProof proof; } AuditorUpdate; ]]> The timestamp field contains the timestamp of the corresponding log entry. The added field contains the list of PrefixLeaf structures that were added to the prefix tree in the corresponding log entry. The removed field contains the list of PrefixLeaf structures that were removed from the prefix tree. The proof field contains a batch lookup proof in the previous log entry's prefix tree for all search keys referenced by added or removed. The proof.results field contains the result of the search for each element of added in the order provided, followed by the result of the search for each element of removed in the order provided. An auditor processes a single AuditorUpdate by following these steps:

Verify that timestamp is greater than or equal to the timestamp of the previous log entry.
Verify that the elements of added are sorted in ascending order by vrf_output and that the same vrf_output is not present twice. Verify the same in removed. Note that a VRF output in added is also allowed to be in removed.
For each element of added where the same VRF output is not in removed, verify that the PrefixSearchResult provided in proof has a result_type of nonInclusionParent or nonInclusionLeaf
Verify that the PrefixSearchResult provided in proof for each element of removed has a result_type of inclusion.
For each element of removed, verify that, with the addition of the new log entry, the prefix tree leaf was published in at least one distinguished log entry before removal.
With proof and the PrefixLeaf structures in removed, compute the root value of the previous log entry's prefix tree. Verify that this matches the auditor's state.
With proof and the PrefixLeaf structures in added and removed, compute the new root value of the prefix tree. Compute the new root value of the log tree after adding a leaf with the specified timestamp and prefix tree root value.
Optionally, provide an AuditorTreeHead to the Service Operator where AuditorTreeHead.timestamp is set to timestamp and AuditorTreeHead.tree_size is set to the new size of the log tree after the addition of the new leaf. The signature is computed with the log tree root value computed in the previous step.

Security Considerations The security properties provided by this protocol are discussed in detail in . Generally speaking, the Key Transparency protocol ensures that all users of a Transparency Log have a consistent view of the data stored in the log. Service Operators may still be able to make malicious modifications to stored data, such as by attaching new public keys to a user's account and encouraging other users to encrypt to these public keys when messaging the user. However, since the existence of these new public keys is equally visible to the user whose account they affect, the user can promptly act to have them removed from their account or inform contacts out-of-band that their communication may be compromised. Key Transparency relies on users coming online regularly to monitor for unexpected or malicious modifications to their account. Users that go offline for longer than the log entry maximum lifetime may not detect if the Transparency Log made malicious modifications to their labels. Similarly, Key Transparency relies on the ability of users to retain long-term state regarding their account and past views of the Transparency Log. Users which are unable to maintain long-term state, or may lose their state, have a correspondingly limited ability to detect misbehavior by the Service Operator. In particular, users which are completely stateless will generally gain nothing by participating in this protocol over simply verifying a signature from the Service Operator and, if there is one, the Third-Party Auditor or Manager. Ultimately, ensuring that all users have a consistent view of the Transparency Log requires that the Service Operator is not able to create and maintain long-term network partitions between users. As such, users need access to at least one communication channel (even a very low-bandwidth one) that is resistant to partitions. The protocol directly provides for a Third-Party Auditor or Manager, which is trusted to prevent such partitions. Other options include allowing users to gossip with each other, or allowing users to contact the Transparency Log over an anonymous channel. Key Transparency provides users with a limited assurance that query responses are authentic: a network attacker will not be able to forge false responses to queries, but may provide responses which are up to max_behind milliseconds stale. Key Transparency provides no privacy from network observers and does not have the ability to authenticate specific users to the Transparency Log. To mitigate these limitations, users SHOULD contact the Transparency Log over a protocol that provides transport-layer encryption and an appropriate level of authentication for both parties.

IANA Considerations This document requests the creation of the following new IANA registry:

KT Cipher Suites ()

This registry should be under a heading of "Key Transparency", and assignments are made via the Specification Required policy . See for additional information about the KT Designated Experts (DEs). RFC EDITOR: Please replace XXXX throughout with the RFC number assigned to this document

KT Cipher Suites A cipher suite is a specific combination of cryptographic primitives and parameters to be used in an instantiation of the protocol. Cipher suite names follow the naming convention: The columns in the registry are as follows:

Value: The numeric value of the cipher suite.
Name: The name of the cipher suite.
Recommended: Whether support for this cipher suite is RECOMMENDED. Valid values are "Y", "N", and "D", as described below. The default value of the "Recommended" column is "N". Setting the Recommended item to "Y" or "D", or changing an item whose current value is "Y" or "D", requires Standards Action .
- Y: Indicates that the item is RECOMMENDED. This only means that the associated mechanism is fit for the purpose for which it was defined. Careful reading of the documentation for the mechanism is necessary to understand the applicability of that mechanism. A cipher suite may, for example, be recommended that is only suitable for use in applications where the Transparency Log's contents are public. Mechanisms with limited applicability may be recommended, but in such cases applicability statements that describe any limitations of the mechanism or necessary constraints will be provided.
- N: Indicates that the item's associated mechanism has not been evaluated and is not RECOMMENDED (as opposed to being NOT RECOMMENDED). This does not mean that the mechanism is flawed.
- D: Indicates that the item is discouraged and SHOULD NOT be used. This marking could be used to identify mechanisms that might result in problems if they are used, such as a weak cryptographic algorithm or a mechanism that might cause interoperability problems in deployment.
Reference: The document where this cipher suite is defined.

Initial contents:

Value	Name	R	Ref
0x0000	RESERVED	-	RFC XXXX
0x0001	KT_128_SHA256_P256	Y	RFC XXXX
0x0002	KT_128_SHA256_Ed25519	Y	RFC XXXX
0xF000 - 0xFFFF	Reserved for Private Use	-	RFC XXXX

Both cipher suites currently specified share the following primitives and parameters:

The hash algorithm is SHA-256, as defined in .
Nc: 16
Kc: The byte sequence equal to the hex-encoded string d821f8790d97709796b4d7903357c3f5

The KT_128_SHA256_P256 cipher suite is as follows:

The signature algorithm is ECDSA over the NIST curve P-256. Messages are hashed with SHA-256 before being signed. Public keys are encoded as an uncompressed point as defined in SEC 1, Version 2.0, Section 2.3.3. Signatures are encoded as the concatenation of two 256-bit big endian integers r and s.
The VRF algorithm is ECVRF-P256-SHA256-TAI as defined in . Public keys are encoded as a compressed point as defined in SEC 1, Version 2.0, Section 2.3.3.

The KT_128_SHA256_Ed25519 cipher suite is as follows:

The signature algorithm is Ed25519 as defined in . Public key and signature encodings are as defined in .
The VRF algorithm is ECVRF-EDWARDS25519-SHA512-TAI as defined in with the output truncated to 32 bytes.

KT Designated Expert Pool Specification Required registry requests are registered after a three-week review period on the KT Designated Expert (DE) mailing list kt-reg-review@ietf.org on the advice of one or more of the KT DEs. However, to allow for the allocation of values prior to publication, the KT DEs may approve registration once they are satisfied that such a specification will be published. Registration requests sent to the KT DEs' mailing list for review SHOULD use an appropriate subject (e.g., "Request to register value in KT registry"). Within the review period, the KT DEs will either approve or deny the registration request, communicating this decision to the KT DEs' mailing list and IANA. Denials SHOULD include an explanation and, if applicable, suggestions as to how to make the request successful. Registration requests that are undetermined for a period longer than 21 days can be brought to the IESG's attention for resolution using the iesg@ietf.org mailing list. Criteria that SHOULD be applied by the KT DEs includes determining whether the proposed registration duplicates existing functionality, whether it is likely to be of general applicability or useful only for a single application, and whether the registration description is clear. IANA MUST only accept registry updates from the KT DEs and SHOULD direct all requests for registration to the KT DEs' mailing list. It is suggested that multiple KT DEs who are able to represent the perspectives of different applications using this specification be appointed, in order to enable a broadly informed review of registration decisions. In cases where a registration decision could be perceived as creating a conflict of interest for a particular KT DE, that KT DE SHOULD defer to the judgment of the other KT DEs.

References Normative References Key Transparency Architecture This document defines the terminology and interaction patterns involved in the deployment of Key Transparency in a general secure group messaging infrastructure, and specifies the security properties that the protocol provides. It also gives more general, non- prescriptive guidance on how to securely apply Key Transparency to a number of common applications. Key words for use in RFCs to Indicate Requirement Levels In many standards track documents several words are used to signify the requirements in the specification. These words are often capitalized. This document defines these words as they should be interpreted in IETF documents. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements. Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words RFC 2119 specifies common key words that may be used in protocol specifications. This document aims to reduce the ambiguity by clarifying that only UPPERCASE usage of the key words have the defined special meanings. The Transport Layer Security (TLS) Protocol Version 1.3 This document specifies version 1.3 of the Transport Layer Security (TLS) protocol. TLS allows client/server applications to communicate over the Internet in a way that is designed to prevent eavesdropping, tampering, and message forgery. This document updates RFCs 5705 and 6066, and obsoletes RFCs 5077, 5246, and 6961. This document also specifies new requirements for TLS 1.2 implementations. HMAC: Keyed-Hashing for Message Authentication This document describes HMAC, a mechanism for message authentication using cryptographic hash functions. HMAC can be used with any iterative cryptographic hash function, e.g., MD5, SHA-1, in combination with a secret shared key. The cryptographic strength of HMAC depends on the properties of the underlying hash function. This memo provides information for the Internet community. This memo does not specify an Internet standard of any kind Guidelines for Writing an IANA Considerations Section in RFCs Many protocols make use of points of extensibility that use constants to identify various protocol parameters. To ensure that the values in these fields do not have conflicting uses and to promote interoperability, their allocations are often coordinated by a central record keeper. For IETF protocols, that role is filled by the Internet Assigned Numbers Authority (IANA). To make assignments in a given registry prudently, guidance describing the conditions under which new values should be assigned, as well as when and how modifications to existing values can be made, is needed. This document defines a framework for the documentation of these guidelines by specification authors, in order to assure that the provided guidance for the IANA Considerations is clear and addresses the various issues that are likely in the operation of a registry. This is the third edition of this document; it obsoletes RFC 5226. Verifiable Random Functions (VRFs) A Verifiable Random Function (VRF) is the public key version of a keyed cryptographic hash. Only the holder of the secret key can compute the hash, but anyone with the public key can verify the correctness of the hash. VRFs are useful for preventing enumeration of hash-based data structures. This document specifies VRF constructions based on RSA and elliptic curves that are secure in the cryptographic random oracle model. This document is a product of the Crypto Forum Research Group (CFRG) in the IRTF. Edwards-Curve Digital Signature Algorithm (EdDSA) This document describes elliptic curve signature scheme Edwards-curve Digital Signature Algorithm (EdDSA). The algorithm is instantiated with recommended parameters for the edwards25519 and edwards448 curves. An example implementation and test vectors are provided. Informative References CONIKS: Bringing Key Transparency to End Users SEEMless: Secure End-to-End Encrypted Messaging with less trust OPTIKS: An Optimized Key Transparency System Merkle^2: A Low-Latency Transparency Log System Secure hash standard National Institute of Standards and Technology (U.S.)

Implicit Binary Search Tree The following Python code demonstrates efficient algorithms for navigating the implicit binary search tree: > k) > 0: k += 1 return k-1 # The level of a node in the tree. Leaves are level 0, their parents # are level 1, etc. If a node's children are at different levels, # then its level is the max level of its children plus one. def level(x): if x & 0x01 == 0: return 0 k = 0 while ((x >> k) & 0x01) == 1: k += 1 return k # The root index of a search if the log has `n` entries. def root(n): return (1 << log2(n)) - 1 # The left child of an intermediate node. def left(x): k = level(x) if k == 0: raise Exception('leaf node has no children') return x ^ (0x01 << (k - 1)) # The right child of an intermediate node. def right(x, n): k = level(x) if k == 0: raise Exception('leaf node has no children') x = x ^ (0x03 << (k - 1)) while x >= n: x = left(x) return x ]]>

Binary Ladder The following Python code demonstrates efficient algorithms for computing the versions of a label to include in a binary ladder: n: break # Binary search between the established lower and upper bounds. lower_bound = out[-2] upper_bound = out[-1] while lower_bound+1 < upper_bound: value = (lower_bound + upper_bound) // 2 out.append(value) if value <= n: lower_bound = value else: upper_bound = value return out # Returns the set of versions that would be looked up in a search binary ladder # where the target version is t and the greatest version of the label that # exists in a given version of the prefix tree is n. def search_binary_ladder( t, n, left_inclusion = [], right_non_inclusion = [] ): def would_end(v): # (Proof of inclusion for a version greater than t) OR # (Proof of non-inclusion for a version less than or equal to t) return (v <= n and v > t) or (v > n and v <= t) def would_be_duplicate(v): return (v in left_inclusion) or (v in right_non_inclusion) out = base_binary_ladder(n) end = next((i+1 for i,v in enumerate(out) if would_end(v)), len(out)) filtered_out = [v for v in out[:end] if not would_be_duplicate(v)] return filtered_out # Returns the set of versions that would be looked up in a monitoring binary # ladder where the monitored version of the label is t. def monitoring_binary_ladder(t, left_inclusion = []): out = base_binary_ladder(t) filtered_out = [v for v in out if v <= t and v not in left_inclusion] return filtered_out ]]>

CombinedTreeProof Implementation The CombinedTreeProof structure is designed to provide all of the information a user needs to evaluate the algorithms defined in this document with as little overhead as possible. The main strategy used to reduce overhead is deduplication, such that the same timestamp or prefix tree root value is never provided twice. Another strategy used is providing data in the specific order that it's required by the algorithm being executed, rather than explicitly indicating which timestamp or proof belongs to which log entry. As such, the generally simplest way to process a CombinedTreeProof is for an implementation to execute the algorithms specified in this document, and have the algorithm implementations call functions like get_timestamp(x) or get_search_ladder(x, target_ver) where x is the log entry's position. These functions can then pop the first timestamp or PrefixProof from the corresponding field as a queue, and associate this value with the log entry x for the purpose of deduplication and, eventually, evaluating the associated inclusion proof. These functions can also evenly enforce the invariants that all timestamps are monotonic, and that if there are multiple PrefixProof structures from the same log entry, that they all compute the same prefix tree root value.

Log Entry Maximum Lifetime and the Reasonable Monitoring Window The log entry maximum lifetime is an optional configuration parameter that a Transparency Log can define to allow pruning. Transparency Logs are only required to retain a logarithmic number of log entries that have passed their maximum lifetime, and aren't required to retain any versions of the prefix tree that have passed their maximum lifetime. This prevents the amount of storage that the Transparency Log needs from growing infinitely over time, which naturally allows the Transparency Log to remain operational much longer that it otherwise would be able to. The Reasonable Monitoring Window (RMW) is a similar but mandatory parameter defined by the Transparency Log, and represents the general frequency with which users are expected to monitor labels they own. The protocol creates roughly one distinguished log entry per every interval of the RMW, and these distinguished log entries act as common reference points for all users in their monitoring of the Transparency Log. Similar to how the log entry maximum lifetime bounds the amount of state that the Transparency Log needs to maintain, the Reasonable Monitoring Window bounds how log users need to retain certain monitoring-related state. Importantly, the log entry maximum lifetime, if defined, will always be greater than the RMW. This ensures that users are able to take advantage of the common reference points provided by distinguished log entries prior to the Transparency Log permanently deleting them. The log entry maximum lifetime and the RMW can interact in somewhat subtle ways, and this section hopes to clarify those interactions. The rightmost distinguished log entry is always on the frontier. The algorithm for computing distinguished log entries is structured such that if a given log entry is distinguished, then its parent will always be distinguished as well. Every log entry in the implicit binary search tree is either on the frontier itself, or in the left subtree of a log entry that is on the frontier. Given these two points, if a log entry is distinguished and not on the frontier, then it has in its direct path a frontier log entry that is also distinguished and to its right. The rightmost distinguished log entry is never expired. This is obviously true in the case where the rightmost distinguished log entry is the rightmost log entry, since the rightmost log entry is never expired by definition. If the rightmost distinguished log entry is not the rightmost log entry, then it has a right child. Say that T_i is the timestamp of the rightmost distinguished log entry and that T_rightmost is the timestamp of the rightmost log entry. The right child of the rightmost distinguished log entry obviously can not itself be distinguished. Through the algorithm in , this corresponds to the statement that T_rightmost - T_i < RMW. Since we know by definition that the RMW is less than the log entry maximum lifetime, or that RMW < ML, this directly implies that T_rightmost - T_i < RMW < ML, or that the rightmost distinguished log entry is not expired. This property is particularly important for the algorithms in KT since it means that, once a distinguished log entry is created in a Transparency Log, there will always be at least one unexpired distinguished log entry that users can inspect and use as a common reference point for monitoring. The Contact Monitoring algorithm will always intersect the first distinguished log entry established to its right. The Contact Monitoring algorithm takes as input a log entry that is, or was previously, to the right of the rightmost distinguished log entry. It computes the direct path of this log entry and checks that the (newly created) log entries on the right side of this path are constructed correctly, stopping once it reaches the first distinguished log entry. The claim here is that, assuming the monitored log entry doesn't itself become distinguished, this algorithm always intersects the first distinguished log entry established to the right of the monitored log entry. Equivalently, the claim is that the algorithm terminates within a bounded amount of time, being roughly the RMW. Assume we know the position of the first distinguished log entry established to the right of the monitored log entry. Then one of the following cases apply:

Case 1: The monitored log entry is in the left subtree of the distinguished log entry. In this case, the claim of this section is true by definition.
Case 2: The distinguished log entry is in the right subtree of the monitored log entry. Recalling that if a log entry is distinguished then it's parent is also always distinguished, then in this case the monitored log entry is now itself distinguished. In this case, the Contact Monitoring algorithm removes the version of the label from monitoring without further inspection.
Case 3: Neither of the above cases apply. In this case, there is a "middle" log entry where the monitored log entry is in its left subtree and the distinguished log entry is in its right subtree. Because the middle log entry is a parent of the distinguished log entry, it must be distinguished as well. This is a contradiction, since the distinguished log entry can not be the "first" distinguished log entry if the middle log entry is distinguished and to its left.

Assuming no clock skew and no Transparency Log outages, a user that performs Contact Monitoring every interval of the RMW will never hit an expired log entry. Each Contact Monitoring request follows a path from a monitored log entry to a frontier log entry. The frontier log entry, if it is not distinguished, becomes the monitored log entry in the user's subsequent Contact Monitoring request. As a result of this structure, each Contact Monitoring request only inspects log entries that were created since the prior Contact Monitoring request. If the rightmost log entry's timestamp during the first request was T_rightmost, then the rightmost log entry's timestamp during the subsequent request will be approximately T_rightmost + RMW, and all inspected log entries will have a timestamp in this range. Since the RMW is less than the log entry maximum lifetime, no log entry in this range can be expired. However, clock skew is permitted by the protocol, configurable by the max_ahead and max_behind parameters, and some degree of service outages are generally unavoidable. The maximum lifetime parameter needs to be set with sufficient allowance for both of these eventualities.