Codance Ltd · Project Aurora

From capable AI
to justified confidence.

Agentic AI can decide, delegate and act. Codance is developing Project Aurora, a working alpha platform that turns assurance methods into controlled, evidence-producing campaigns so qualified people can reconstruct what the system materially did.

Codance is developing Project Aurora: a working alpha that operates assurance methods as controlled, evidence-producing investigations of agentic systems.

Working alpha · bounded demonstration completedEvidential demonstrationInviting aligned investment & collaboration
Codance geometric C containing a structured data stack
Working towards trusted, useful probabilistic intelligence.Built in the UK · technology-vendor neutral · designed for independent challenge

The assurance gap

A plausible answer is not evidence of safe operation.

An agent may provide an acceptable visible response while the application has already crossed an authority boundary, invoked a consequential tool or produced a material side effect. System records may show that individual events occurred, but not establish why they mattered to the assurance question, how they affected subsequent work or whether the prescribed method responded appropriately. Without a governed relationship between the method, the evidence and the resulting consequences, qualified reviewers cannot reconstruct a sufficient basis for confidence.

What Aurora does

Aurora makes evidence drive the investigation.

Aurora operates a commissioned assurance method as a governed process. Each admitted material event can initiate versioned deterministic and probabilistic work—not merely become another stored record. That work can validate exact conditions, apply bounded contextual judgement, relate new evidence to earlier products and initiate further controlled interaction with the system. Every input, processing step, product and resulting action retains its identity and lineage.

Govern the method

Define the assurance question, boundaries, permitted evidence, collection points and review conditions before interpreting results.

Control what enters

Register campaign identity separately and admit each material event against the commissioned conditions. Evidence that cannot satisfy them fails closed.

Investigate and respond

Process evidence through versioned deterministic and probabilistic skills. Results can initiate bounded further work or controlled interaction with the system.

Reconstruct and challenge

Preserve every input, step, product, interaction and consequence so qualified reviewers can reconstruct how the finding was produced.

Aurora assurance investigation showing a separately governed campaign gate, event-initiated deterministic and probabilistic work, a bounded return interaction with the system under test, competent human review and reverse evidential reconstruction

The demonstrated difference

Same records. Same policy. Opposite recommendations.

Our accounts-payable demonstrator presents the same overdue invoice, business records and policy to two governed configurations of the same sealed SUT revision. One recommends HOLD. The other recommends PAY after contextual correspondence is promoted into formal receipt evidence and carried through a two-stage agentic process. Aurora reconstructed that change in meaning, located the first material divergence and produced a governed human-readable assessment.

The same business records and policy produce HOLD and PAY recommendations after different receipt-evidence guides shape the first-stage evidence compilation
ERP-authoritative Recommend HOLD

80 of 100 units are formally received in ERP.

but
Operational correspondence Recommend PAY

An upstream guide transforms contextual email into formal receipt evidence.

The answer shows what the system concluded. The evidence shows how it got there. A competent reviewer must be able to trace the recommendation through the source records, policy, installed guide, intermediate compilation and exact handoff between model calls.

Controlled demonstration conclusionCONTROLLED_CONFIGURATION_EFFECT_SUPPORTED
Read the controlled report →

Why Codance

A specialist role in a market that will not have one universal answer.

Codance is building Aurora to become one of the specialist platforms organisations choose because it turns competent assurance intent into controlled, reconstructible evidence.

Focused IP ownership

Codance is the coherent development and IP-holding vehicle for Project Aurora. Product knowledge compounds in one accountable organisation.

Method and evidence capital

Each serious campaign can add reusable method patterns, evidential experience, observed limitations and delivery knowledge beyond the code itself.

Designed to collaborate

Competent experts retain authority over methods and judgements; Codance supplies the operational platform through which they can be executed and evidenced.

Explore the programme

Explore in more depth.

These programme pages provide further technical, evidential and development detail.

Interested in challenging, applying or supporting the work?

Discuss collaboration →