HAL assessment

HAL Control Profile

HAL: Human-Accountable-for-the-Loop

For workflows where individual review does not scale. Twenty-four questions across eight domains, plus a non-scored capability evidence state. About five minutes. You'll get an eight-domain evidence profile, a capability state, critical-gate findings, authority-specific control expectations, and deployment priorities.

Nothing is sent anywhere. The assessment runs entirely in your browser.

Not sure if HAL applies? Start with the Loop model →

Calculator context: …. We've pre-selected that authority level below so the result can show the controls it requires.

Recalculate →

Authority context not calculated yet.

The Workflow Calculator identifies the appropriate Loop model and the controls expected for what the system can do.

Check the workflow →

0 / 24 answered

First, what does this system do?

This sets the authority context. Greater authority changes which controls must be stronger; it does not create a numerical pass mark.

Advice System Produces outputs for a human to use. Takes no action of its own. Recommendation System Influences decisions by ranking, scoring, or classifying. A human still acts. Execution System Takes actions in systems of record, within bounded authority. Autonomous System Takes consequential, externally-facing actions with real-world effect.

Human capability

Capability sustainability is a cross-cutting condition, not a ninth scored domain. Record the evidence state of the human judgement this control case relies on. Complete the capability map where Review, Monitor or Own depends on that judgement.

Assumed Capability is inferred from title, qualification or historic experience. Defined The judgement required for meaningful oversight is explicit. Designed Necessary developmental experience has been preserved, compressed, simulated or moved. Tested People have demonstrated the capability against realistic uncertainty and AI error. Sustained Performance, continued exposure and the future pipeline are monitored and acted upon over time.

01 Ownership

Is there a single, named individual accountable for this system?

Absent.No. It is owned by a team, committee, or "the platform". Partial.Informally: people know roughly who, but it is not recorded. Implemented.Yes: a named owner is recorded, but no deputy. Tested.Yes: a named owner and deputy are recorded and current.

Does the accountable owner have the power to suspend the system immediately?

Absent.No. Switching it off requires a change board or vendor. Partial.In theory, but it has never been tested and the path is unclear. Implemented.Yes, but only via a process that takes hours. Tested.Yes: the owner can pause it immediately and the kill-switch is tested.

Did the owner formally accept the authority delegated to the system?

Absent.No sign-off exists. Partial.Verbal or informal agreement only. Implemented.Signed off once at launch, not revisited. Tested.Signed off, recorded in a register, and re-confirmed on schedule.

02 Authority

Is the system's authority expressed as an explicit list of permitted actions?

Absent.No. It is described in free text or a prompt only. Partial.Loosely: a general description of what it "should" do. Implemented.Mostly: a documented list with some gaps. Tested.Yes: a precise allow-list of concrete permitted actions.

Is authority enforced in code, or only requested of the model?

Absent.Only in the prompt: the model is asked to behave. Partial.Mostly prompt, with a few technical guards. Implemented.Enforced at the integration layer for the riskiest actions. Tested.Enforced in code at the integration layer for all actions.

Is authority bounded by quantitative limits (value, volume, recipients)?

Absent.No quantitative bounds at all. Partial.Some informal bounds, not enforced. Implemented.Key bounds enforced, others missing. Tested.Comprehensive, enforced bounds on value, volume, and counterparties.

03 Limits

Are there explicitly prohibited actions the system can never take?

Absent.No prohibited-action list exists. Partial.A few prohibitions described informally. Implemented.A documented list, partially enforced. Tested.A clear list enforced as hard guard rails.

What happens automatically when a hard limit is reached?

Absent.Nothing. Limits are advisory at best. Partial.A warning is logged but the action proceeds. Implemented.The action is blocked but no one is alerted. Tested.The action stops, the system pauses, and a human is alerted.

Have the limits been tested adversarially (deliberately trying to breach them)?

Absent.Never tested. Partial.Tested informally during development. Implemented.Tested at launch, not since. Tested.Tested regularly, including red-teaming, with results recorded.

04 Escalation

Does the system escalate to a human on uncertainty or novel cases?

Absent.No. It always proceeds, even when unsure. Partial.Sometimes, based on undocumented heuristics. Implemented.Yes for errors, but not for uncertainty or novelty. Tested.Yes: explicit triggers for uncertainty, novelty, limits, and high stakes.

Is each escalation routed to a named role with a response-time expectation?

Absent.No defined recipient or SLA. Partial.A shared inbox with no SLA. Implemented.A named role, but no agreed response time. Tested.A named role with an agreed, monitored SLA.

Does the system pause the affected action while awaiting a human?

Absent.No. It proceeds on its best guess. Partial.It proceeds but flags the case for later. Implemented.It pauses high-risk actions only. Tested.It pauses the affected action until a human responds.

05 Evidence

Can any single decision be reconstructed after the fact?

Absent.No. Only the final outcome is visible. Partial.Partially: outputs are kept, inputs are not. Implemented.Mostly: inputs and outputs, but not versions, policy results, or actions. Tested.Fully: material inputs, outputs, versions, policy results, actions, authority, and available rationale are recorded.

Is the decision record immutable and time-stamped?

Absent.No record, or one that can be freely edited. Partial.A mutable log with no integrity controls. Implemented.Time-stamped logs, but editable by admins. Tested.Immutable, time-stamped, tamper-evident records.

Are human reviews, approvals, overrides and interventions attributable, with proportionate evidence of capability where judgement is a material control?

Absent.No. Human actions are not logged. Partial.Some touchpoints are logged, with no evidence of the judgement involved. Implemented.Reviews, approvals and overrides are attributable. Tested.Attributable human actions, plus proportionate evidence of capability where judgement is a material control.

06 Monitoring

Do production signals cover service, input, behaviour, action, human operation, capability and outcome?

Absent.No production monitoring. Partial.Basic uptime and volume only. Implemented.Error, escalation and drift metrics, but not human operation or capability. Tested.Signals cover service, behaviour, action, human operation, capability and outcome in proportion to risk.

Do alert thresholds notify the owner of anomalies?

Absent.No alerting. Partial.Alerts exist but go to an unmonitored channel. Implemented.Alerts reach the team, with unclear thresholds. Tested.Tuned thresholds page the named owner.

Can monitoring automatically pause the system on a severe anomaly?

Absent.No. Monitoring only reports. Partial.A human must notice and act manually. Implemented.Auto-pause exists for a narrow set of conditions. Tested.Auto-pause is in place for severe anomalies, with review.

07 Review

Is there a scheduled date for formal re-examination of the system?

Absent.No review is scheduled. Partial.Reviews happen ad hoc when something goes wrong. Implemented.An annual review is scheduled. Tested.A review date and owner are set and tracked.

Does review re-test whether the original purpose, delegated authority and assumed human capability remain justified?

Absent.There is no review. Partial.Review checks performance only, not justification or assumed capability. Implemented.Justification is considered; human capability is assumed from role or qualification. Tested.Review re-tests purpose, delegated authority and assumed human capability.

Can a review re-scope, re-approve, or retire the system?

Absent.No mechanism to change or retire it. Partial.Changes require a separate, undefined process. Implemented.Review can recommend changes but not enforce them. Tested.Review is empowered to re-scope, re-approve, or retire.

08 Liability

Is it documented which of the system's actions can create legal or financial obligations?

Absent.No. This has not been analysed. Partial.Understood informally by some staff. Implemented.Documented for the highest-risk actions. Tested.Comprehensively documented and mapped to controls.

Do vendor and customer contracts address autonomous-system actions?

Absent.Contracts are silent on AI-driven actions. Partial.Generic clauses that do not address autonomy. Implemented.Some contracts updated, others not. Tested.Liability deliberately allocated across the relevant contracts.

Is insurance and disclosure for autonomous actions confirmed?

Absent.Neither confirmed. Partial.Assumed covered, never verified. Implemented.One of insurance or disclosure addressed. Tested.Insurance confirmed and disclosure obligations defined.

Answer all 24 questions and record a capability evidence state.

Your result

Critical-gate status

…

Critical gaps

These controls are gates. Other implemented controls do not compensate for an absent one.

Deployment recommendation

Capability evidence state

Not averaged with the eight-domain profile. An unsupported assumption weakens Ownership, Evidence, Monitoring and Review where those controls rely on human judgement.

Map what else the work did →

Domain breakdown

Evidence priorities

The weakest domains and the next control evidence to establish. This is not a points path.

This is a self-assessment aid, not legal advice, an audit, or a certification. Statuses reflect the answers given; they are claims until evidenced. Note that a completed assessment is itself a governance record. Decide deliberately where it is stored and how deployment decisions refer to it.