Runtime Dataset™

Overview

Runtime Dataset™ is the unified runtime dataset of Atlas Runtime Governance™.

Atlas Agent continuously writes Runtime Observations into the Runtime Dataset, and all downstream investigations are completed based on this exact same Dataset.

The Runtime Dataset is neither a log file nor an Evidence Pack.

It is the raw data foundation for Runtime Evidence.

Runtime Dataset Architecture

Runtime Surface
        │
        ▼
Runtime Observation
        │
        ▼
Runtime Dataset
        │
        ├──────── Candidate Timeline
        │
        ├──────── Manual Time Slice
        │
        ├──────── Candidate Evidence Pack
        │
        └──────── Primary Evidence Pack

Atlas consistently maintains a single Runtime Dataset.

Different investigation modes share the exact same data source.

Runtime Dataset Lifecycle

The Runtime Dataset lifecycle consists of:

Observe
    ↓
Persist
    ↓
Rolling Buffer
    ↓
Time-range Export

Atlas never constructs separate Datasets for different Investigation Workflows.

The entire system maintains only a single Runtime Dataset lifecycle.

Observe

Atlas Agent continuously observes Runtime Surfaces.

Observations can originate from:

Sensor Runtime
Driver Runtime
Linux Runtime
Power
Bus
ROS Runtime
Custom Runtime Surfaces

All Observations flow into the Runtime Dataset.

Persist

Persist continuously writes Observations into the Runtime Dataset.

Persist is not responsible for analyzing events.

Its responsibility is to ensure Runtime Evidence can be referenced by downstream Investigations.

Rolling Buffer

By default, the Runtime Dataset uses a Rolling Buffer.

Once data exceeds the retention window, the oldest data is automatically overwritten according to policy.

Retention Policies are configurable:

rolling_buffer_hours
max_storage_size
overwrite_policy
dataset_lock_on_ref

Default recommendations:

Policy	Default
rolling_buffer_hours	48 hours
overwrite_policy	circular
dataset_lock_on_ref	enabled

OEMs can adjust policies based on deployment needs.

Dataset Lock

When an REF enters formal investigation, Atlas locks the corresponding time range.

During the lock period:

It is excluded from the Rolling Buffer
Overwriting is prohibited
Runtime Evidence integrity is preserved

The lock remains active until the Evidence Pack export is completed.

Dataset Lock strictly affects only the designated time slice.

Time-range Export

The Runtime Dataset supports exporting by time range.

Exports can be used for:

Manual Investigation
Candidate Investigation
Evidence Pack Generation
Historical Archives

Exporting does not alter the underlying Runtime Dataset.

Runtime Dataset Organization

The Runtime Dataset is composed of Observations from multiple Runtime Surfaces.

For example:

Runtime Dataset
│
├── Camera Observation
├── LiDAR Observation
├── IMU Observation
├── Power Observation
├── Ethernet Observation
├── USB Observation
├── Driver Observation
├── Linux Runtime Observation
└── ROS Topic Observation

All Observations are organized along a unified timeline.

Runtime Timeline

All Observations in the Runtime Dataset are aligned to a single, unified Runtime Timeline.

Time
────────────────────────────────────>

Camera

LiDAR

Ethernet

Driver

Linux Runtime

ROS Topic

A unified timeline facilitates:

Cross-stream Correlation
Evidence Window Generation
Runtime Investigation

Runtime Dataset Consumption

The Runtime Dataset can be consumed by different modules.

Runtime Dataset
        │
        ├──────── Manual Time Slice
        │
        ├──────── Candidate Timeline
        │
        ├──────── Candidate Evidence Pack
        │
        ├──────── Primary Evidence Pack
        │
        └──────── Runtime Export

All modules share the exact same Dataset.

The Runtime Dataset is never duplicated.

Manual Time Slice

When Tier 1 knows the approximate time of an incident:

Atlas operates based on the specified time range:

Approximate REF Time
        │
        ▼
Time Slice
        │
        ▼
Evidence Pack

There is no need to re-collect data.

Candidate Timeline

Some incidents cannot be assigned an accurate occurrence time.

For example:

"The robot stopped unexpectedly sometime yesterday afternoon."

Atlas can scan the Runtime Dataset to identify Candidate Timelines worth investigating further.

Candidate Timelines:

Provide diagnostic leads
Offer candidate time windows
Do NOT confirm Root Cause
Do NOT confirm causality

Candidate Evidence Pack

Every Candidate Timeline can generate a corresponding Candidate Evidence Pack.

For example:

Candidate Timeline
        │
        ├── EP-C01
        ├── EP-C02
        ├── EP-C03
        └── EP-C04

All Candidate Evidence Packs are retained.

Tier 2 engineers can then:

Review
Compare
Merge
Select Primary Evidence Pack

Atlas does not automatically delete Candidate Evidence Packs.

Primary Evidence Pack

During the investigation process, Tier 2 can select:

Primary Evidence Pack

As the formal Investigation Evidence.

Primary Evidence Packs:

Reference the Runtime Dataset
Utilize a Five-Segment Window
Retain the complete Runtime Timeline

Primary Evidence Packs do not impact the retention of Candidate Evidence Packs.

Runtime Dataset Principles

The Runtime Dataset adheres to the following principles:

Only one Runtime Dataset exists across the entire system
All investigations share the exact same Dataset
Dataset lifecycle is unified
Runtime Observations are never re-collected
Candidate Timelines do not alter the Dataset
Evidence Packs reference the Dataset
Exporting does not duplicate the Dataset

Runtime Dataset vs Evidence Pack

Runtime Dataset:

Continuously exists
Continuously grows
Managed by Rolling Buffer
Retains complete Runtime Observations

Evidence Pack:

Targets a single REF
Generated from the Runtime Dataset
Uses a Five-Segment Window
Used for Runtime Investigation

Evidence Packs originate from the Runtime Dataset.

The Runtime Dataset is not equal to an Evidence Pack.

Runtime Dataset Boundary

The Runtime Dataset strictly stores runtime data.

It does NOT contain:

Root Causes
Investigation Results
Lessons Learned
Historical RGAs
Investigation Contexts

These investigation assets belong to Runtime Investigation™.

Summary

Runtime Dataset™ is the sole runtime data foundation of Atlas Runtime Governance™.

Atlas Agent continuously writes Runtime Observations into the Runtime Dataset and manages data retention, locking, and exporting through a unified lifecycle.

All Investigation Workflows—including Manual Time Slices, Candidate Timelines, Candidate Evidence Packs, and Primary Evidence Packs—share the exact same Runtime Dataset without creating independent data lifecycles.

Next Reading

Evidence Pack™
Historical RGA™
Investigation Context™

Overview​

Runtime Dataset Architecture

Runtime Dataset Lifecycle

Observe​

Persist​

Rolling Buffer​

Dataset Lock​

Time-range Export​

Runtime Dataset Organization

Runtime Timeline

Runtime Dataset Consumption

Manual Time Slice

Candidate Timeline

Candidate Evidence Pack

Primary Evidence Pack

Runtime Dataset Principles

Runtime Dataset vs Evidence Pack

Runtime Dataset Boundary

Summary

Next Reading

Overview

Observe

Persist

Rolling Buffer

Dataset Lock

Time-range Export