Claims
Identify checkable statements, comparisons, forecasts, recommendations and other assertions that may require support.
Reliance — Document Assurance
Evidence review for AI-assisted documents.
Reliance separates document claims from the material that supports them so reviewers can check support and limitations before relying on the wording.
Review a workflowReliance — Document Assurance is the current public identity for this bounded document claim-and-evidence review. The legacy /risk address remains available for existing links.
- Currently
- Implemented document claim-and-evidence review with support reasons and human review; legacy rule traceability remains partial.
- Output
- Saved review with claim states, evidence links, reasons and recorded review disposition.
- Boundary
- Supported is not universally true; weak or unsupported is not proven false; outside check is not verified. Enterprise-wide risk management is not implied.
Interactive explainerFictional records · not live product output
Select a claim, inspect its support and record a review decision.
Claim → source → check → disposition
Does the evidence support these words?
Four fictional claims. Four different review questions.Supported within the file
“The pilot included eight sites.”
The source supports this bounded count. It says nothing about the number of users, tasks or production deployments.
Pilot note A · paragraph 1
The pilot was conducted at eight sites.
Evidence boundarySupported means supported by this supplied record. It does not mean independently verified as true.
This example runs in your browser. No document is uploaded. Decisions reset when you reload; the export is an illustrative record, not a product audit trail.
Inputs, checks & review limits
- Ingestion before reasoning
- Check whether the supplied document can be read before analysing its claims. Extracted text does not automatically preserve page layout, tables, signatures or the authority of the source.
- Different checks, different authority
- Arithmetic and rule checks answer bounded questions. Evidence-fit assessment asks whether a supplied source supports the wording. Neither automatically verifies external facts.
- Review & record
- A reviewer’s disposition is a separate event from a support label. A receipt records the review context; it is not by itself an immutable ledger, a signature or an assurance of legal or regulatory compliance.
AI can make a document look finished before its supporting evidence has been checked.
Reliance connects claims to their supporting material and separates evidence gaps from writing quality.
Evidence fit
Determine whether the available material actually supports the wording being used.
Verification gaps
Separate what the file can establish from facts that require an external check.
Human disposition
Preserve what a reviewer accepted, revised, escalated or left unresolved.
A core distinction
Polished language is not the same as supported language.
INPUTDocument claim
The statement as written in the memo, report, summary or AI-assisted draft.
CHECKEvidence boundary
What the available material supports, what constrains the wording and what requires verification.
RESULTReview state
Supported, weak, unresolved, externally verifiable or escalated for human review.
A fluent document can still be wrong.Human review remains explicit where judgment or external verification is required.
Review flow
From a document to a reviewable claim record.
Explore the 6-step workflow
/01IngestBring the document and available supporting material into a defined review context.
/02Extract claimsIdentify statements that can be checked, supported, narrowed or challenged.
/03Map evidenceConnect each material claim to the source material that appears to support or constrain it.
/04Flag gapsIdentify wording that outruns the file or requires external verification.
/05ReviewAllow a responsible person to accept, revise, narrow, escalate or reject the claim.
/06RecordPreserve what was reviewed, what changed and what remains unresolved.
Current maturity
A bounded document-review application.
Working nowCurrent
Document ingestion, claim extraction, evidence linkage, support classification, deterministic checks and review receipts.
Product development & longer-term direction
Product developmentExpanding
Stronger approval workflows, versioning, role controls, multi-user review and integration into enterprise document processes.
Longer-termDirection
A broader evidence-and-review environment for consequential AI-assisted documents across legal, audit, compliance and professional-service workflows.
Current scope: a bounded document-review application. Enterprise approval, multi-user access and integration requirements are not implied by the website example.
How we scope, test and deploy a systemWhere it fits
Where document claims need checking.
Risk is most relevant where polished wording can create exposure if claims, sources and verification gaps are not inspected before approval or distribution.
Legal teams
Review AI-assisted memos, summaries and case material where wording needs defensible support.
Compliance & audit
Separate supported claims from statements that require additional verification or escalation.
Professional services
Inspect client-facing analysis, reports and deliverables before they become relied-upon work.
Procurement & risk
Review evidence-heavy documents, assessments and decision files with a structured record.
Review an AI-assisted document workflow.
Start with the document, its supporting material and the claims a reviewer needs to check.