Structured intake
Line-item identity, timekeeper role, rate, hours, and original narrative enter a typed transaction contract.
Enterprise legal operations infrastructure
Billing review governed by rules, evidence, and human authority.
DRE is a focused review layer for corporate legal departments managing outside-counsel invoices and Outside Counsel Guidelines. It resolves objective policy questions deterministically, isolates semantic exceptions for controlled analysis, and preserves an auditable path to authorized disposition.
One file inStructured legal-billing input
Objective controls firstExplicit, repeatable policy evaluation
Ambiguity isolatedControlled semantic exception review
Two governed outputsPassed consensus and Human Review
Platform architecture
A transaction-processing system with explicit control boundaries.
DRE does not treat a language model as the billing authority. It maintains distinct contracts for deterministic compliance, interpretive evidence, deterministic routing, financial calculation, audit state, and authorized human review.
Deterministic evaluation
Versioned rules identify objective violations, sanitize affected narratives, calculate proposed values, and determine whether interpretation is necessary.
Controlled consensus
Only unresolved semantic exceptions reach differentiated schema, domain, and anomaly evaluations.
Deterministic referee
Explicit vote thresholds select commit, reconstruction, rejection, or protected human escalation.
Unified audit receipt
Both processing paths return the same typed record of inputs, findings, pathway, votes, values, narrative, and review state.
Outside Counsel Guidelines
Policy enforcement requires provenance, not pattern matching alone.
DRE’s production direction binds each applicable control to a policy source, rule identifier and version, effective date, customer and matter context, exception logic, and the appropriate deterministic, semantic, or human treatment.
Objective requirements
Administrative charges, rate limits, thresholds, approved timekeeper classifications, and explicit prohibited activity.
Interpretive requirements
Vague narratives, block billing, substantive-purpose questions, related-work separation, and missing context.
Authorized decisions
Exceptions, disputed interpretation, incomplete evidence, policy conflicts, and final financial disposition.
Evidence architecture
Every processing decision has a reconstructable history.
The target audit record preserves the original transaction, rules applied, deterministic findings, narrative states, consensus invocation and outputs, referee result, calculated financial effect, reason for escalation, and authorized disposition.
Examine transaction orchestration →InputOriginal transaction and narrative
ControlRule source, version, evidence, and result
InterpretationAgent votes, confidence, intent, and reasoning
AuthorityRecommendation separated from disposition
Customer-owned Azure deployment
Enterprise operation inside the customer’s Microsoft control plane.
The current infrastructure definition establishes a VNet, application subnet, Key Vault, Cosmos DB account, and Azure OpenAI account with public network access disabled. It is a meaningful foundation—not a claim of completed production hardening.
Current foundation
- Reusable Bicep structure
- Region and naming parameters
- Virtual network and AppSubnet
- Azure Key Vault resource
- Azure Cosmos DB account
- Azure OpenAI account with public access disabled
Production build-out
- Application runtime and structured messaging
- Private endpoints and private DNS
- Microsoft Entra ID, managed identities, and RBAC
- Governed data containers and model deployments
- Monitor, Log Analytics, Application Insights, and alerts
- Retention, recovery, environment separation, and assurance
Validation discipline
Architecture claims become credible only when controls are tested.
DRE’s assurance progression defines the expected control, implements it, tests both the permitted and prohibited paths, captures evidence, and repeats after change.
- Static validationCompile and inspect infrastructure definitions.
- Controlled deploymentCreate an attributable test environment.
- Configuration assertionsCompare deployed properties with the contract.
- Connectivity and authorizationProve intended paths work and unintended paths fail.
- Evidence and repeatabilityRetain results, clean up, and reproduce the process.
Demonstrable DRE
Evaluate the mechanism, not a marketing simulation.
The public demonstration uses supplied synthetic legal-billing data and browser-local processing. It separates deterministic results from semantic exception handling and produces downloadable Passed Consensus and Human Review populations.
Run the supplied legal-billing file
Do not submit customer matter data to the public demonstrator. Production data belongs in the governed customer environment.