Source context
System, file, sensor, ledger or workflow source used by the AI-supported process.
Evidence Fabric: the proof layer behind AI-assisted work.
Evidence Fabric is the shared evidence layer across CloudSeals applications. It records source context, model/agent details, purpose, policy/authority context, outcomes, approvals/overrides, downstream actions and final results.
What an evidence record connects
- Source eventWhat happened and where the data came from.
- Agent recommendationThe proposal, model/agent context and purpose.
- Policy decisionThe rule, threshold and authority applied.
- Human approvalWho approved, rejected or overrode it.
- Execution resultWhat was carried out and what followed.
- Evidence recordThe connected context retained for later review.
Each stage links to the same workflow record. Access, retention and deployment controls are agreed during security review; this diagram does not imply a certification or a specific storage guarantee.
Evidence record fields
Policy context
Applicable rule, operating procedure, approval threshold or regulatory reference.
Human decision
Reviewer, approval status, exception reason and final action separated from AI recommendation.
Model and agent details
Which model or agent produced the recommendation, so a reviewer knows what they are looking at.
Purpose
What the AI-supported step was for — the reconciliation, the safety rule, the carbon figure.
Authority context
Who was entitled to approve, and the limits that applied to them.
Outcome, approvals and overrides
What the AI proposed, what the person decided, and any override with its reason.
Downstream actions and final results
What was carried out after approval and what the final result was.
Retention, access, export and review
Retention
A configured decision per deployment, not a fixed default, because reporting and audit obligations differ by organisation and jurisdiction. See Security.
Access
Controlled through identity and access management; use of the record is itself recorded.
Export
Records are readable in the application and exported where the workflow needs it — for example, CarbonSight exports evidence for assurance and reporting.
Review
Written for someone who was not there: an internal reviewer, an auditor or a regulator can follow what happened and on what basis.
Who it is for, and how it is obtained
Evidence Fabric is used by the reviewers who make decisions in LedgerSight, CompliSight and CarbonSight, and read by the auditors and regulators who check them. It is not purchased separately: it comes with each application, and the deployment model is agreed per engagement. It is the evidence layer of what CloudSeals calls an accountable AI platform.
Evidence Fabric connects source information, policy context, human review and workflow outcomes in a reviewable record.
Where this is heading
CloudSeals research is extending TrustOps and Evidence Fabric into governed digital twins — how policy, simulation, human authority and evidence could govern increasingly autonomous cyber-physical operations.
Research direction — not currently available as a product.
Why a shared layer
If each application keeps its own notes, an auditor has to reconcile three stories. A shared evidence layer means a finance exception, a safety alert and a carbon figure are all recorded the same way — same fields, same shape, one place to look.
One record shape
Source, purpose, policy context, decision and outcome are recorded the same way across applications.
Written as work happens
The record is produced by the workflow, not assembled afterwards from memory.
Readable by a reviewer
The point is that someone who was not there can follow what happened.
What Evidence Fabric is not
Worth stating plainly, because evidence claims attract scrutiny.
Not a separate product
It is the shared layer underneath LedgerSight, CompliSight and CarbonSight.
Not a compliance guarantee
It records what happened. Whether that satisfies a given regulation is a judgement made by your auditors and regulators.
Not a replacement for your systems of record
It records the decision context around the systems you already run.
Questions we get asked
What is Evidence Fabric?
Evidence Fabric is the shared evidence layer across CloudSeals applications. It records source context, model/agent details, purpose, policy/authority context, outcomes, approvals/overrides, downstream actions and final results.
What exactly is recorded?
Source context, model and agent details, the purpose of the action, the policy and authority context that applied, the outcome, approvals and overrides, downstream actions and final results.
Is Evidence Fabric something I buy separately?
No. It is the shared layer beneath the CloudSeals applications rather than a separate product.
Does an evidence record prove compliance?
It gives a reviewer what they need to follow a decision. Whether that meets a particular obligation is decided by your auditors and regulators, not by the software.
Who is the record written for?
Someone who was not in the room — an internal reviewer, an auditor, a regulator — reading it after the fact.
How does this relate to TrustOps?
TrustOps is the operating discipline for accountable AI and requires that a record exists. Evidence Fabric is where that record lives.
Read an example decision record.
This simplified LedgerSight example explains the record fields. It is not a customer record, an actual product export or proof of a completed deployment.
- Source
- Sample bank statement row 17 and ledger row 42.
- Recommendation and purpose
- Investigate an unmatched receipt with a missing payment reference. Illustrative recommendation; no live model was run.
- Policy and authority
- Finance-reviewer approval is required before a proposed resolution is applied.
- Human decision
- Review requested. Reviewer not yet assigned. No approval or override recorded.
- Action and outcome
- No action approved or executed. The exception remains open pending human review.