A single ML inference event with full operational metadata. More...

#include <inference_log.hpp>

Public Member Functions
std::vector< uint8_t >	serialize () const
	Serialize the record to a deterministic byte sequence.

Static Public Member Functions
static expected< InferenceRecord >	deserialize (const uint8_t *data, size_t size)
	Reconstruct an InferenceRecord from its serialized byte representation.

Public Attributes
int64_t	timestamp_ns {0}
	Inference timestamp (nanoseconds since epoch)

std::string	model_id
	Model identifier (e.g., "gpt-4", "bert-base")

std::string	model_version
	Model version hash or checkpoint ID.

InferenceType	inference_type {InferenceType::CLASSIFICATION}
	Type of inference.

std::vector< float >	input_embedding
	Input embedding (optional, may be empty)

std::string	input_hash
	SHA-256 hash of raw input (for privacy)

std::string	output_hash
	SHA-256 hash of raw output.

float	output_score {0.0f}
	Primary output score/probability.

int64_t	latency_ns {0}
	Inference latency in nanoseconds.

int32_t	batch_size {1}
	Batch size.

int32_t	input_tokens {0}
	Input token count (LLM, 0 if N/A)

int32_t	output_tokens {0}
	Output token count (LLM, 0 if N/A)

std::string	user_id_hash
	Hashed user ID (for privacy)

std::string	session_id
	Session identifier.

std::string	metadata_json
	Additional JSON metadata.

std::string	training_dataset_id
	Training data identifier.

int64_t	training_dataset_size {0}
	Number of samples in training dataset.

std::string	training_data_characteristics
	Description of training data properties.

int64_t	model_training_end_ns {0}
	Timestamp when model training completed (EU AI Act Art.12)

int64_t	model_training_data_cutoff_ns {0}
	Latest data timestamp used in training.

std::string	model_retraining_schedule
	Cron or description of retraining schedule (EU AI Act Art.13)

Detailed Description

A single ML inference event with full operational metadata.

Captures everything needed to audit and reproduce an inference: model identity, timing, resource usage, and privacy-preserving hashes of inputs and outputs. Raw data is never stored — only SHA-256 hashes — to comply with GDPR data minimization.

Definition at line 64 of file inference_log.hpp.

Member Function Documentation

◆ deserialize()

static expected< InferenceRecord > signet::forge::InferenceRecord::deserialize	(	const uint8_t *	data,
		size_t	size
	)

inlinestatic

Reconstruct an InferenceRecord from its serialized byte representation.

Parameters

data	Pointer to the serialized bytes.
size	Number of bytes available at `data`.

Returns: The deserialized record, or an error if the data is truncated.

Definition at line 167 of file inference_log.hpp.

◆ serialize()

std::vector< uint8_t > signet::forge::InferenceRecord::serialize ( ) const

inline

Serialize the record to a deterministic byte sequence.

Format: each field written sequentially as little-endian values. Strings: 4-byte LE length prefix + raw bytes. Vectors: 4-byte LE count + float data (4 bytes each, LE).

Definition at line 97 of file inference_log.hpp.

Member Data Documentation

◆ batch_size

int32_t signet::forge::InferenceRecord::batch_size {1}

Batch size.

Definition at line 74 of file inference_log.hpp.

◆ inference_type

InferenceType signet::forge::InferenceRecord::inference_type {InferenceType::CLASSIFICATION}

Type of inference.

Definition at line 68 of file inference_log.hpp.

◆ input_embedding

std::vector<float> signet::forge::InferenceRecord::input_embedding

Input embedding (optional, may be empty)

Definition at line 69 of file inference_log.hpp.

◆ input_hash

std::string signet::forge::InferenceRecord::input_hash

SHA-256 hash of raw input (for privacy)

Definition at line 70 of file inference_log.hpp.

◆ input_tokens

int32_t signet::forge::InferenceRecord::input_tokens {0}

Input token count (LLM, 0 if N/A)

Definition at line 75 of file inference_log.hpp.

◆ latency_ns

int64_t signet::forge::InferenceRecord::latency_ns {0}

Inference latency in nanoseconds.

Definition at line 73 of file inference_log.hpp.

◆ metadata_json

std::string signet::forge::InferenceRecord::metadata_json

Additional JSON metadata.

Definition at line 79 of file inference_log.hpp.

◆ model_id

std::string signet::forge::InferenceRecord::model_id

Model identifier (e.g., "gpt-4", "bert-base")

Definition at line 66 of file inference_log.hpp.

◆ model_retraining_schedule

std::string signet::forge::InferenceRecord::model_retraining_schedule

Cron or description of retraining schedule (EU AI Act Art.13)

Definition at line 91 of file inference_log.hpp.

◆ model_training_data_cutoff_ns

int64_t signet::forge::InferenceRecord::model_training_data_cutoff_ns {0}

Latest data timestamp used in training.

Definition at line 90 of file inference_log.hpp.

◆ model_training_end_ns

int64_t signet::forge::InferenceRecord::model_training_end_ns {0}

Timestamp when model training completed (EU AI Act Art.12)

Definition at line 89 of file inference_log.hpp.

◆ model_version

std::string signet::forge::InferenceRecord::model_version

Model version hash or checkpoint ID.

Definition at line 67 of file inference_log.hpp.

◆ output_hash

std::string signet::forge::InferenceRecord::output_hash

SHA-256 hash of raw output.

Definition at line 71 of file inference_log.hpp.

◆ output_score

float signet::forge::InferenceRecord::output_score {0.0f}

Primary output score/probability.

Definition at line 72 of file inference_log.hpp.

◆ output_tokens

int32_t signet::forge::InferenceRecord::output_tokens {0}

Output token count (LLM, 0 if N/A)

Definition at line 76 of file inference_log.hpp.

◆ session_id

std::string signet::forge::InferenceRecord::session_id

Session identifier.

Definition at line 78 of file inference_log.hpp.

◆ timestamp_ns

int64_t signet::forge::InferenceRecord::timestamp_ns {0}

Inference timestamp (nanoseconds since epoch)

Definition at line 65 of file inference_log.hpp.

◆ training_data_characteristics

std::string signet::forge::InferenceRecord::training_data_characteristics

Description of training data properties.

Definition at line 86 of file inference_log.hpp.

◆ training_dataset_id

std::string signet::forge::InferenceRecord::training_dataset_id

Training data identifier.

Definition at line 84 of file inference_log.hpp.

◆ training_dataset_size

int64_t signet::forge::InferenceRecord::training_dataset_size {0}

Number of samples in training dataset.

Definition at line 85 of file inference_log.hpp.

◆ user_id_hash

std::string signet::forge::InferenceRecord::user_id_hash

Hashed user ID (for privacy)

Definition at line 77 of file inference_log.hpp.

The documentation for this struct was generated from the following file:

include/signet/ai/inference_log.hpp

Public Member Functions

Static Public Member Functions

Public Attributes