Back to Demos
Research Preview

Answers That Admit What Is Unknown

When evidence is missing, most systems substitute a default and carry on. The number stays green, the gap stays invisible, and the failure surfaces at audit.

Watch it run — sign-up to result

Stage 1 of 5
aribot · Innovation Preview3 of 3 areas
AreasToday vs nextWhere to see itUnder NDA

Each area is described by what changes for you. How it is done is covered under NDA.

Verified outcomes
observable outcome

What tools do today

A fix is reported done when the instruction was sent.

What you get instead

A finding closes only after the resource has been re-read and agrees.

Concrete example

Bucket re-read after the change: public access blocked → finding closed

Already visible in the shipped product: /demos/ai-remediation

Authority that matches use
observable outcome

What tools do today

Access is recorded at the moment it is granted, and never compared to behaviour.

What you get instead

What an identity holds is shown beside what it has ever actually used.

Concrete example

ci-deploy-bot: 12 scopes granted · 4 ever used → 8 reducible

Already visible in the shipped product: /demos/agent-assurance

Answers that admit what is unknown
observable outcome

What tools do today

A missing measurement is replaced with a default, so untested reads as passing.

What you get instead

Unmeasured stays a visible state and is excluded from the score.

Concrete example

281 controls not evaluated — excluded, not counted as passing

Already visible in the shipped product: /demos/cxo-workbench

Each of these is a rule about what the product will and will not claim. You can check all three today on the demos above — which is the point: a claim you can watch is worth more than one we explain.

Useryou do this step

Ask for a posture number

GOES IN

Scope + framework

COMES OUT

working…

ILLUSTRATIVE SEED DATA — NOT CUSTOMER DATA

62% of 18% tested

Run log
t+0.0s [User] Ask for a posture number
working on: Scope + framework

The problem

A control nobody tested reports the same as a control that passed. Both render as a number, and only one of them was measured.
Percentages are shown without their denominator, so a high score over a tiny sample reads as estate-wide assurance.
The error is largest exactly where the stakes are highest: the board pack and the audit, where nobody can see the missing denominator behind the figure.

What Aribot does

  • Unmeasured is a visible state in its own right, not a zero and not a pass.
  • Coverage travels with every score, so a reader always knows how much of the estate the figure describes.
  • Untested items are excluded from the denominator and reported separately rather than folded in.
  • Where a check could not run, the reason is kept and shown, so a gap can be closed rather than argued about.

In the product

  1. 1
    Ask for a posture number
    In: Scope + framework. Out: Score with its basis.
  2. 2
    See what was not measured
    In: Test results to date. Out: Untested list.
  3. 3
    Understand why
    In: A gap. Out: Reason it could not run.
  4. 4
    Close the gap
    In: Connect / test. Out: Higher coverage.
  5. 5
    Report it honestly
    In: Period. Out: Pack stating both.
Book a walkthrough All demos

Why teams use this

Use cases and case studies

Use case

Report untested controls as untested

Aribot keeps unmeasured as its own visible state, so every score carries its coverage and an untested control never counts as a pass.

Read the use case One-pager Carousel

Case study · Shadow AI · Customer zero

Our own code called five AI providers and 25 models

134
distinct places in our code that call an AI model
5
AI providers in use, across 25 different models

Case study · AI cost · Customer zero

150 paid AI calls a night, down to a handful

~150 → 4–7
paid model calls in the nightly cost run, before the fix and on the first two nights after
$4.85 → $0.26
cost of that run on the last night before, and the higher of the first two nights after

All use cases and case studies