]> VeritasChain Co., Ltd.

tokachi@veritaschain.org

Security Independent Submission provenance timestamp merkle tree RFC 3161 media authenticity The Content Provenance Profile (CPP) is an open specification for cryptographically verifiable media capture provenance. This document defines the core data model, hashing conventions, Merkle tree construction rules, RFC 3161 Time-Stamp Authority (TSA) anchoring protocol, and offline verification procedures for CPP. CPP enables capture devices to produce tamper-evident provenance records that bind media content to external timestamps via trusted third parties. Unlike self-attestation models, CPP requires independent timestamp verification through RFC 3161 TSA services, providing externally verifiable proof of when media was captured. This specification focuses on the interoperable core of CPP: the data structures, cryptographic operations, and verification algorithms necessary for independent third-party verification. Application-specific features such as depth analysis are defined as optional extensions.

Introduction

Problem Statement Digital media authenticity faces several fundamental challenges:

• Self-Attestation Weakness: Systems where creators sign their own claims provide no independent verification. A verifier must trust that the creator's claimed timestamp is accurate.
• Metadata Stripping: Social media platforms and messaging applications routinely strip embedded metadata, breaking provenance chains that depend on file-level embedding.
• Omission Attacks: Systems that prove individual media authenticity cannot detect when unfavorable evidence has been selectively deleted from a collection.
• Terminology Confusion: Terms like "verified" mislead users into believing content truthfulness has been established, when only provenance has been recorded.

Design Goals CPP addresses these challenges through the following design principles:

1. External Timestamp Verification: Timestamps MUST be anchored to independent RFC 3161 Time-Stamp Authorities, enabling third-party verification without trusting the capture device.
2. Omission Detection: The Completeness Invariant mechanism enables detection of deleted events within a collection.
3. Offline Verification: All data necessary for verification is included in the Evidence Pack, enabling verification without network access.
4. Provenance ≠ Truth: CPP proves when and by what device media was captured. It does NOT prove content truthfulness or scene authenticity.

Scope This document specifies:

• The CPP event data model
• Hash computation and canonicalization rules
• Merkle tree construction and proof verification
• RFC 3161 TSA anchoring requirements
• Verification procedures for implementers

This document does NOT specify:

• Application user interface requirements
• Network protocols for proof distribution
• Key management or certificate policies
• Content authenticity claims

Relationship to Other Specifications CPP defines its own Merkle tree construction that is NOT compatible with Certificate Transparency . While inspired by similar principles, CPP uses different domain separation prefixes and padding rules optimized for media provenance use cases. Implementations MUST NOT assume RFC 6962 compatibility.

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. Additionally, this document uses the following terms:

Event

A discrete, signed record representing a provenance action (capture, export, deletion).

EventHash

A SHA-256 hash of the canonicalized event data, formatted as sha256:<64_hex_chars>.

LeafHash

SHA-256(0x00 || EventHash_bytes), used as input to the Merkle tree. The 0x00 prefix provides domain separation from internal nodes.

MerkleRoot

The root hash of a binary Merkle tree containing one or more LeafHashes.

AnchorDigest

The 32-byte value submitted to the TSA. Represented as a 64-character lowercase hexadecimal string without prefix. MUST equal the MerkleRoot value (after stripping the sha256: prefix).

TreeSize

The count of original leaves in the Merkle tree before any padding. An unsigned integer. MUST be >= 1.

PaddedSize

The count of leaves after padding to a power of 2. Computed as the smallest power of 2 greater than or equal to TreeSize.

Evidence Pack

A self-contained data structure containing all information necessary for offline verification.

Tombstone

An event that records the legitimate deletion of a previous event.

Provenance Available

The recommended terminology for UI display, indicating capture provenance is recorded without implying content truth verification.

Genesis PrevHash

The constant value used as PrevHash for the first event in a chain: sha256:0000000000000000000000000000000000000000000000000000000000000000 (64 zeros).

Threat Model

Addressed Threats

Threat	Mitigation
Timestamp forgery	RFC 3161 TSA provides independent timestamp
Evidence tampering	EventHash binds content; Merkle proof binds to anchor
Selective deletion	Completeness Invariant detects missing events
TSA token swapping	messageImprint must match AnchorDigest

Explicitly Not Addressed

• Content truthfulness (scene staging, deepfakes)
• Device compromise before capture
• Key extraction from secure hardware
• TSA collusion with adversary

Data Model

Events An Event is the fundamental unit of provenance in CPP. Events are signed records that capture discrete provenance actions.

Event Types

Type	Description
INGEST	Media captured from device sensor
SEAL	Collection sealed with Completeness Invariant
EXPORT	Proof shared externally
TOMBSTONE	Legitimate deletion record

Event Structure The following fields are REQUIRED for all events:

Field	Type	Required	Description
EventID	string	REQUIRED	Unique identifier (UUID)
ChainID	string	REQUIRED	Identifier linking events in a sequence
PrevHash	string	REQUIRED	Hash of previous event in chain
Timestamp	string	REQUIRED	ISO 8601 timestamp with millisecond precision
EventType	string	REQUIRED	One of: INGEST, SEAL, EXPORT, TOMBSTONE
HashAlgo	string	REQUIRED	Always "SHA256"
SignAlgo	string	REQUIRED	"ES256" or "Ed25519"
EventHash	string	REQUIRED	SHA-256 hash of canonicalized event
Signature	string	REQUIRED	Raw Base64-encoded signature (no prefix)

Encoding Requirements All Base64-encoded fields (Signature, TSA.Token, public_key) MUST conform to:

• Section 4 (standard base64 alphabet, NOT base64url)
• No whitespace characters (no line breaks, spaces, or tabs)
• Padding characters (=) MUST be included when required

PROHIBITED:

• base64:MEUCIQDx... - Prefixes not allowed
• data:application/octet-stream;base64,... - Data URIs not allowed
• Base64url alphabet (- and _ instead of + and /)
• Line wrapping or embedded whitespace

INGEST Event INGEST events MUST include:

Field	Type	Required	Description
Asset.AssetHash	string	REQUIRED	SHA-256 hash of media bytes
Asset.AssetType	string	REQUIRED	IMAGE or VIDEO
Asset.MimeType	string	REQUIRED	MIME type of asset
Asset.AssetID	string	OPTIONAL	Unique asset identifier
Asset.AssetName	string	OPTIONAL	Original filename
Asset.AssetSize	integer	OPTIONAL	File size in bytes

SEAL Event SEAL events finalize a collection and commit the Completeness Invariant. A SEAL event MUST include:

Field	Type	Required	Description
CollectionID	string	REQUIRED	Identifier for the sealed collection
EventCount	integer	REQUIRED	Number of events in collection (excluding SEAL)
CompletenessInvariant	object	REQUIRED	Completeness verification data
MerkleRoot	string	REQUIRED	Root hash of events in collection

CompletenessInvariant Object

Field	Type	Required	Description
ExpectedCount	integer	REQUIRED	Number of events that MUST be present
HashSum	string	REQUIRED	XOR of all EventHash values (sha256: format)
FirstTimestamp	string	REQUIRED	ISO 8601 timestamp of first event
LastTimestamp	string	REQUIRED	ISO 8601 timestamp of last event

The HashSum is computed as: Where XOR operates on the 32-byte binary values of each EventHash.

SEAL Anchoring Pattern The recommended anchoring pattern for SEAL events:

1. Compute the Merkle tree over all INGEST events in the collection
2. Create the SEAL event with MerkleRoot referencing this tree
3. Include the SEAL event's EventHash as an additional leaf in a NEW Merkle tree
4. Anchor this new tree (containing the SEAL event) to a TSA

This pattern ensures the Completeness Invariant itself is bound to an external timestamp. The SEAL event's EventHash covers all CI fields, so the TSA anchor proves the CI existed at GenTime. Alternative Pattern (NOT RECOMMENDED): Including the SEAL event in the same Merkle tree it references creates a circular dependency and is prohibited.

TOMBSTONE Event TOMBSTONE events MUST additionally include:

Field	Type	Required	Description
DeletedEventId	string	REQUIRED	EventID being invalidated
Reason	string	REQUIRED	Deletion reason code
DeletedAt	string	REQUIRED	ISO 8601 deletion timestamp

Hash Chain Events form a hash chain through the PrevHash field: Verification of chain integrity:

1. For each event after the first, the verifier MUST compute EventHash of the previous event.
2. The computed hash MUST match the current event's PrevHash field.
3. If any mismatch is detected, verification MUST fail with status CHAIN_INTEGRITY_VIOLATION.

Merkle Tree Structure CPP defines its own binary Merkle tree construction optimized for media provenance. This construction uses domain separation prefixes to prevent attacks where leaf values could be confused with internal node values. Important: CPP Merkle trees are NOT compatible with RFC 6962 (Certificate Transparency). Implementations MUST use the exact algorithms specified in this section.

Domain Separation CPP uses single-byte prefixes to separate domains:

Domain	Prefix Byte	Description
Leaf	0x00	Applied to EventHash bytes
Internal	0x01	Applied to concatenated child hashes

Leaf Nodes Where:

• 0x00 is a single byte with value zero
• EventHash_bytes is the 32-byte binary representation of EventHash (after stripping the sha256: prefix)
• || denotes byte concatenation

Rationale: The 0x00 prefix ensures leaf hashes cannot collide with internal node hashes, preventing second preimage attacks on the tree structure.

Internal Nodes Where:

• 0x01 is a single byte with value one
• Left_bytes is the 32-byte hash of the left child
• Right_bytes is the 32-byte hash of the right child
• || denotes byte concatenation

Tree Construction Step 1: Compute Leaf Hashes For each event, compute LeafHash = SHA256(0x00 || EventHash_bytes). Step 2: Determine Padding PaddedSize is the smallest power of 2 >= TreeSize: = 1 paddedSize = 1 while paddedSize < treeSize: paddedSize = paddedSize * 2 return paddedSize ]]> Step 3: Pad Leaf Array If TreeSize < PaddedSize, duplicate the last leaf hash until the array length equals PaddedSize. Step 4: Build Tree 1: nextLevel = [] for i in range(0, current.length, 2): left = current[i] right = current[i + 1] parent = SHA256(0x01 || left || right) nextLevel.append(parent) levels.append(nextLevel) current = nextLevel return levels // levels[0] = leaves, levels[-1] = [root] ]]>

Merkle Proof Structure

Field	Type	Description
TreeSize	integer	Original leaf count (before padding), unsigned, MUST be >= 1
LeafHashMethod	string	MUST be exactly SHA256(0x00||EventHash) (18 ASCII characters)
LeafHash	string	Computed LeafHash for this event with sha256: prefix
LeafIndex	integer	0-based position in tree, range [0, TreeSize-1]
Proof	array	Sibling hashes from bottom to top, each with sha256: prefix
Root	string	MerkleRoot with sha256: prefix

TreeSize Constraint: An empty Merkle tree (TreeSize = 0) is not permitted. Verifiers MUST reject proofs where TreeSize < 1.

Anchor Structure

Field	Type	Description
AnchorID	string	Unique anchor identifier
AnchorType	string	MUST be "RFC3161"
AnchorDigest	string	MerkleRoot without prefix, 64 lowercase hex chars
AnchorDigestAlgorithm	string	MUST be "sha-256"
Merkle	object	Merkle proof structure
TSA	object	TSA response data

Canonicalization and Hashing

JSON Canonicalization Events MUST be canonicalized using (JSON Canonicalization Scheme) before hashing. The following fields MUST be excluded from canonicalization:

• EventHash
• Signature

All other fields MUST be included. Field names in the canonical event object use PascalCase (e.g., EventID, ChainID, PrevHash).

EventHash Computation The resulting EventHash is a 71-character string: the prefix "sha256:" followed by 64 lowercase hexadecimal characters.

LeafHash Computation The 0x00 prefix byte provides domain separation from internal nodes.

Internal Node Hash Computation The 0x01 prefix byte distinguishes internal nodes from leaves.

AnchorDigest Computation AnchorDigest is the MerkleRoot value WITHOUT the sha256: prefix, represented as 64 lowercase hexadecimal characters. PROHIBITED:

• SHA256(merkleRoot) - This would create double hashing
• SHA256(stringEncode(merkleRoot)) - This would hash the string representation
• Any transformation other than prefix removal
• Mixed case output - MUST be lowercase

Anchoring Protocol

TSA Request Construction The messageImprint in TimeStampReq MUST contain:

• hashAlgorithm: SHA-256 (OID 2.16.840.1.101.3.4.2.1)
• hashedMessage: AnchorDigest as 32-byte OCTET STRING

certReq Recommendation Producers SHOULD set certReq to TRUE to request the TSA's signing certificate be included in the response. This enables:

• Offline verification without fetching certificates separately
• Long-term verification even if TSA infrastructure changes
• Self-contained Evidence Packs

If certReq is FALSE and the TSA certificate is not included in the response, verifiers MUST attempt to obtain the certificate through other means (e.g., AIA extension, local cache) or return VALID_WARNING.

TSA Response Processing Upon receiving TimeStampResp, the producer:

1. MUST verify the response status is granted (0) or grantedWithMods (1)
2. MUST extract the TimeStampToken from the response
3. MUST store the complete DER-encoded TimeStampToken
4. MUST extract and store the messageImprint from TSTInfo
5. MUST extract and store GenTime from TSTInfo

Single-Leaf Tree Rules When TreeSize equals 1, the following invariants MUST hold:

• LeafIndex MUST equal 0
• Proof MUST be an empty array
• Root MUST equal LeafHash
• LeafHash MUST equal SHA256(0x00 || EventHash_bytes)

If any of these conditions fail, verification MUST return INVALID.

Multi-Leaf Tree Rules For TreeSize greater than 1:

• LeafIndex MUST be in range [0, TreeSize-1]
• PaddedSize = smallest power of 2 >= TreeSize
• Proof length MUST NOT exceed log2(PaddedSize)
• Sibling hashes are ordered from bottom (leaf level) to top (root level)
• Index parity determines pairing order: even=left, odd=right
• All internal nodes use SHA256(0x01 || left || right)

Verification Procedures

Verification Result Codes

Code	Meaning
VALID	All checks passed, including TSA signature verification
VALID_WARNING	Cryptographic checks passed, but TSA certificate chain could not be fully validated
INVALID	Cryptographic verification failed
CHAIN_INTEGRITY_VIOLATION	Hash chain is broken
COMPLETENESS_VIOLATION	Completeness Invariant mismatch

Note: VALID_WARNING indicates the proof is cryptographically sound but the TSA's identity could not be independently verified. Applications SHOULD display this distinction to users.

Event Verification

Merkle Proof Verification = 1") if leafIndex < 0 or leafIndex >= treeSize: return INVALID("LeafIndex out of range") paddedSize = computePaddedSize(treeSize) maxProofLength = log2(paddedSize) if proof.length > maxProofLength: return INVALID("Proof too long") // Step 2: Compute leaf hash with domain separation currentHash = computeLeafHash(eventHash) // SHA256(0x00 || bytes) // Step 3: Handle single-leaf case if treeSize == 1: if leafIndex != 0: return INVALID("LeafIndex must be 0 for single-leaf") if proof.length != 0: return INVALID("Proof must be empty for single-leaf") if lowercase(currentHash) != lowercase(expectedRoot): return INVALID("Root != LeafHash for single-leaf") return VALID // Step 4: Traverse proof from bottom to top index = leafIndex for siblingHash in proof: if index % 2 == 0: // Current is left child currentHash = computeInternalHash(currentHash, siblingHash) else: // Current is right child currentHash = computeInternalHash(siblingHash, currentHash) index = floor(index / 2) // Step 5: Compare with expected root (case-insensitive) if lowercase(currentHash) != lowercase(expectedRoot): return INVALID("Computed root != expected root") return VALID ]]>

TSA Verification TSA verification ensures the timestamp token was legitimately issued by a Time-Stamp Authority and binds the correct digest.

CMS Signature Verification Requirements Per , verifiers MUST:

1. Parse the TimeStampToken as a ContentInfo structure
2. Extract the SignedData from the content field
3. Locate the SignerInfo corresponding to the TSA
4. Verify the signature over the encapsulated TSTInfo
5. Verify the signer's certificate was valid at signing time

Verifiers SHOULD:

1. Build and validate the certificate chain to a trust anchor
2. Verify the TSA certificate contains the id-kp-timeStamping extended key usage
3. Check for certificate revocation

Chain Integrity Verification

Completeness Invariant Verification The Completeness Invariant is verified against a SEAL event. The SEAL event MUST be anchored to a TSA to provide external timestamp binding for the entire collection. ci.LastTimestamp: return COMPLETENESS_VIOLATION( "Event timestamp after collection end") return VALID ]]>

Attack Detection

Attack	Detection
Delete event	Hash sum mismatch and/or count mismatch
Add fake event	Count mismatch and/or hash sum mismatch
Reorder events	Chain integrity violation (PrevHash mismatch)
Modify event	EventHash mismatch in chain

Privacy Considerations

Location Data Location collection SHOULD be disabled by default. When enabled, implementations SHOULD:

• Clearly indicate when location is being recorded
• Allow users to delete location from individual events
• Consider privacy implications of location precision

Biometric Data Implementations MUST NOT store raw biometric data (fingerprints, face images). Human presence verification, if implemented, SHOULD:

• Process biometrics locally on-device
• Store only verification results (boolean flags)
• Never transmit biometric data to external services

Tombstone Privacy When events are deleted via TOMBSTONE:

• Original event content is removed
• TOMBSTONE preserves chain integrity
• Reason codes allow selective disclosure

Shareable vs Forensic Proofs Evidence Packs may be created with different privacy levels:

Level	Includes	Use Case
Shareable	Timestamp, device info, asset hash	Social sharing
Forensic	All metadata including location	Legal proceedings

Security Considerations

Hash Algorithm Agility This specification mandates SHA-256 for all hash computations. Future versions MAY define additional algorithms via the HashAlgo field. Verifiers MUST reject unknown hash algorithms.

Signature Algorithm Requirements Implementations MUST support ES256 (ECDSA with P-256 and SHA-256) for mobile device compatibility. Ed25519 MAY be supported for non-mobile implementations. Private keys SHOULD be stored in hardware security modules (Secure Enclave, StrongBox, TPM) where available.

TSA Trust Security of timestamp proofs depends on TSA trustworthiness. Implementations:

• MUST verify the CMS signature in the TimeStampToken per
• SHOULD validate the certificate chain to a configured trust anchor
• SHOULD use TSAs with published certificate policies
• MAY support multiple TSA services for redundancy

If certificate chain validation fails but CMS signature verification succeeds, the result SHOULD be VALID_WARNING rather than INVALID, as the timestamp binding is cryptographically sound even if the TSA's identity cannot be fully verified.

Merkle Tree Security

Domain Separation The 0x00/0x01 prefix bytes ensure:

• Leaf hashes cannot equal internal node hashes for any input
• An attacker cannot construct a valid proof by substituting internal nodes for leaves
• The tree structure is unambiguous given a root hash

This construction differs from Certificate Transparency which uses a similar but incompatible scheme.

Padding Security Duplicating the last leaf for padding:

• Is deterministic (no randomness)
• Produces the same tree for the same inputs
• Does not leak information about padding count (TreeSize is explicitly stored)

Clock Accuracy Device timestamps (Timestamp field) are self-attested and may be inaccurate. The authoritative timestamp is GenTime from the TSA response. Implementations SHOULD warn users when device time differs significantly from TSA GenTime (e.g., more than 5 minutes).

Deletion Detection Limitations The Completeness Invariant detects deletions within a sealed collection. It does NOT detect:

• Events never created (adversary captured but never recorded)
• Events in other collections
• Deletions before sealing

XOR Hash Sum Limitations The Completeness Invariant uses XOR for omission detection, NOT for cryptographic commitment. Important limitations:

• Collision by design: XOR is commutative and self-inverse. An attacker who can forge TWO events with EventHashes that XOR to zero can delete both without detection.
• Not a commitment scheme: Unlike Merkle roots, the XOR hash sum does not cryptographically bind to a specific set of events.
• Complementary mechanism: The CI is designed to work WITH the Merkle tree anchor, not replace it. The TSA-anchored Merkle root provides the cryptographic commitment; the CI provides additional omission detection for collections.

Threat model: The CI protects against accidental deletion or deletion by parties who cannot forge events (e.g., device owners deleting their own legitimately-captured evidence). It does NOT protect against adversaries who control event creation.

Canonicalization Attacks JSON canonicalization per prevents ordering and whitespace attacks. However, implementations MUST ensure:

• Field names exactly match the specification (PascalCase for events)
• No additional fields are introduced before hashing
• Unicode normalization is handled consistently

IANA Considerations This document has no IANA actions.

Implementation Experience

VeraSnap (Non-Normative) VeraSnap is a consumer iOS application implementing CPP. It demonstrates:

• Secure Enclave key storage with ES256 signatures
• RFC 3161 TSA integration with multiple providers
• Merkle tree construction per this specification
• Offline proof verification

Implementation validated that:

• Single-leaf Merkle proofs verify correctly
• TSA messageImprint extraction works across TSA providers
• Evidence Packs enable verification without network access

Deployment experience informed the explicit specification of:

• AnchorDigest computation (avoiding double-hashing)
• Single-leaf tree invariants
• LeafHashMethod field for algorithm agility

References Normative References Informative References Coalition for Content Provenance and Authenticity

JSON Examples

Canonical Event (Normative) The canonical event structure uses PascalCase field names. This is the structure that MUST be used for EventHash computation.

SEAL Event (Normative)

Anchor Structure (Normative)

Evidence Pack (Non-Normative) The Evidence Pack is a distribution format. Field names use snake_case for compatibility with common JSON conventions in web APIs. This format is non-normative; implementations MAY use alternative formats. Note: When verifying an Evidence Pack, implementations MUST reconstruct the canonical event structure (PascalCase) from the evidence pack fields before computing EventHash.

Test Vectors All test vectors in this section use the domain-separated hash construction defined in this specification.

Test Vector 1: Single-Leaf Tree Input: Computation: Expected Anchor: Verification:

Test Vector 2: Two-Leaf Tree Input: Computation: Verification of Index 0: current is LEFT child 3. siblingHash = sha256:4f16119d36ccd0da91102f57692d73934fd0ad2494280df88449accedbbfb7ea 4. currentHash = SHA256(0x01 || currentHash_bytes || siblingHash_bytes) = sha256:03938e2c8f758e6cae443d499b41c899c373eb0c0198bae61796a069f2b05904 5. Compare with expected Root: MATCH Result: VALID ]]>

Test Vector 3: TSA messageImprint Verification Input: Verification:

Acknowledgements The authors thank:

• The VeritasChain Standards Organization (VSO) for developing the CPP specification series
• The implementers of VeraSnap for validation feedback that improved this specification
• Early reviewers who identified the domain separation and Completeness Invariant modeling issues