Adoption path
Autonomy is earned in stages, and never before rollback has been proven.
ArtOfTheHack starts with no ability to act. Each stage grants a narrow, revocable increase in authority only after the previous stage has been measured in the grantee organization environment.
Sequence
Five stages, each with an exit criterion
No stage is skipped. A new action class introduced later re-enters the sequence at shadow evaluation.
01
Read-only integration
Connect without any ability to act.
ArtOfTheHack connects to evidence sources through APIs, webhooks, event streams, secure file transfer, batch ingestion, message queues, database and data-lake connectors, identity federation, software development kits, or grantee organization-controlled private connectors. No action permission is granted at this stage.
Grantee organization control
- No write credentials issued to ArtOfTheHack
- Evidence sources selected and scoped by the grantee organization
- Connection revocable at any point without operational impact
Exit criterion: Evidence-quality baseline established for every connected source.
02
Shadow evaluation
Measure decision quality against the existing process.
ArtOfTheHack evaluates real decisions in parallel with the current process and executes nothing. Historical replay establishes the baseline; live shadow operation tests the workflow under current conditions.
Grantee organization control
- Existing process remains the operating process
- All divergence documented for review
- Success criteria agreed before evaluation begins
Exit criterion: Measured agreement, calibration, and contradiction discovery meet the agreed threshold.
03
Approval-gated production
Prepare actions that a named authority approves.
ArtOfTheHack assembles the action, the evidence, the alternatives, the uncertainty, and the rollback plan, then routes it to the authority permitted to approve that action class. Nothing executes without approval.
Grantee organization control
- Authority model defined by the grantee organization, including two-person approval
- Separation of duty enforced by the policy engine
- Every approval and denial recorded with identity and rationale
Exit criterion: Approval throughput and action outcomes are stable and documented.
04
Bounded automation
Automate only reversible, low-impact, policy-approved actions.
Automation is granted per action class, never globally. An action becomes eligible only after its rollback path has been demonstrated in the grantee organization environment and its blast radius has been bounded in policy.
Grantee organization control
- Blast-radius limits and rate limits set per action class
- Reversibility check enforced before execution
- Emergency kill switch returns enforcement to native systems
Exit criterion: Automated action classes hold their outcome and reversal performance over time.
05
Expansion
Extend from one workflow to a decision fabric.
The path runs from one decision workflow to one business unit, then across multiple systems, then to an organization decision fabric, then to AI-agent governance, then to additional industry modules.
Grantee organization control
- Each expansion re-enters the same validation sequence
- New action classes start in shadow, not in automation
- Policy packs reused rather than rebuilt per business unit
Exit criterion: Ongoing. Governance discipline does not relax as scope grows.
Shadow mode
What gets measured before anything is allowed to act
Shadow evaluation is where claims become measurements. These are recorded in the grantee organization environment, against the organization's own historical and live cases.
Agreement with analysts
How often the governed determination matches the qualified human outcome.
Decision accuracy
Whether the determination held once the outcome became known.
Evidence completeness
Whether the available evidence was sufficient to support a determination at all.
Contradiction discovery
Conflicts between sources that the existing process did not surface.
Time to decision
Elapsed time from evidence availability to a defensible determination.
Unnecessary escalations
Cases escalated under the current process that did not require escalation.
False positives
Determinations that would have triggered action without justification.
Operational consequences
Modeled disruption of the action that would have been taken.
Confidence calibration
Whether stated confidence tracked observed correctness.
Override behavior
Where analysts disagreed, and whether the disagreement was systematic.
Expansion
From one workflow to a decision fabric
- 01One decision workflow
- 02One business unit
- 03Multiple security or business systems
- 04Organization decision fabric
- 05AI-agent governance
- 06Additional industry modules
Boundaries
What never changes as scope grows
Reversibility precedes autonomy
An action class becomes eligible for automation only after its rollback path has been demonstrated in the grantee organization environment.
Human authority is named
Consequential outcomes remain owned by identified people, with approval, denial, and override recorded against identity.
Native systems remain the fallback
An emergency kill switch returns enforcement to the organization's existing systems without dependency on ArtOfTheHack remaining available.
KRYOS-XS Hypercube is designed to augment qualified institutional decision-makers and existing systems. It does not independently replace legal, medical, financial, security, engineering, or regulatory professionals.