Skip to content

CONSULTING-AS-CODE™ / VUDUVATIONS

Own the logic
that runs your business.

You carry the risk.Keep the evidence behind the answer.

When someone challenges a recommendation, inspect its sources, calculations, qualifications and unresolved questions—even after the model, provider or team changes.

Source-groundedAuditableSupported checksClient-retained

AI workflows built around your business task, with agreed records and review controls.

See how it fits your work.

Models, providers and people change.Keep the basis of your work.

Keep the basis of your work when models, providers or people change. Transcript and supporting files connect to an evidence receipt, knowledge graph and audit report. For the next person or agent: keep the source attached, separate fact from interpretation, carry limits forward, check how claims combine and trace the next recommendation. Supported, validated paths.
View full-size illustration (opens in a new tab)

EXISTING APPLICATIONS / CUSTOM WORKFLOWS

Choose a workflow
for your business.

Our protocols are repeatable workflows for specific business tasks. Start with an existing application, or scope a custom workflow with us.

EXISTING APPLICATION

Key Person Audit

Find critical knowledge dependencies and handover gaps. Keep a proposed handover distinct from demonstrated readiness.

Explore the handover example
EXISTING APPLICATION

Sales Intel

Understand the buyer’s stated needs and a supported next move, with the source and qualifications retained.

Explore the deal example
EXISTING APPLICATION

AI Discovery

Structure AI opportunities, priorities and open questions from the information your business supplies.

Explore AI Discovery
EXISTING APPLICATION

M&A Diligence

Review diligence findings with their source support, open questions and saved exports.

Explore M&A Diligence
CUSTOM DOMAIN PROTOCOL

Need a different workflow?

We can build a custom workflow around your task, evidence and review rules, with scope and acceptance checks agreed first.

Discuss a custom workflow

HOW AN ENGAGEMENT WORKS

Start with the work.
Keep its supporting record.

Agree the business question, the inputs, who reviews the result and the checks needed before use.

  1. 01

    Bring your transcript.

    Supply an exported meeting transcript and the supporting files needed for the task.

  2. 02

    Use a defined workflow.

    Choose an existing application or develop a custom protocol around your questions, rules and review process.

  3. 03

    Keep the result and its basis.

    Review the findings, supporting evidence and open questions. Retain the agreed records for the next person or agent.

Walk through an engagement

THE RECORDS BEHIND THE RESULT

Know what you keep.

Your business deliverable addresses the agreed question. These supporting records help your team inspect its basis.

Evidence Receipt

Inspect the figures used, their recorded source quotations, calculation inputs and check outcomes within the receipt’s scope.

Knowledge Graph

Explore the available relationships among people, processes, risks and findings in the exported graph.

Audit Report

Review recorded processing steps, checks, results and the cost information available for the analysis.

Files and field coverage vary by protocol and engagement. Keep the original transcript and any supporting files needed to check the record.

See an Audit Report sampleSeparate fictional Acme example

Keep the agreed evidence, records and handover materials for your business. Software and reuse rights are defined for the engagement.

THE WORK / THE RETAINED BASIS

Follow one finding.
See what supports it.

Trace the source, repeat the arithmetic and see what remains unconfirmed in this fictional staffing example.

TALLGRASS / FICTIONAL EXAMPLE

One finding, with its source and limits.

Inspect the example · expand for detail
SOURCE → CALCULATION → QUALIFICATIONS

The arithmetic checks. Savings are still unconfirmed.

Fictional example

Swipe across the image to see both records.

01Source02Calculation
Conceptual glass records with JSON. Source: interview, 11 people, approximate annual cost of $54,000 each. Calculation: operands 11 and 54000; formula 11 * 54000; approximate annual estimate $594,000 per year. Actual payroll and savings remain open questions.
CONCEPTUAL RECORDS · SIMPLIFIED JSONView image full size

Still unconfirmedActual payroll costs and how much could be saved.

Combined total withheldA separate $210,000 has no stated period and may overlap with staffing cost.

Inspect the source, calculation and limits

Source context

“We have eleven people in order entry. Fully loaded, each of them is about $54,000 a year.”

Keep the original words and context. The speaker describes eleven people in order entry and an approximate fully loaded annual cost per person, including employment costs.

These are figures recorded in a fictional interview. Their source links let a later agent or authorized reviewer inspect what was supplied; they do not independently verify payroll.

Calculation & limits

Inputs
11 people × about $54,000 per person per year, as stated in the interview.
Calculation
11 × 54,000 = 594,000. The result remains approximate because the per-person input is approximate.
Check scope
Repeating this product checks the arithmetic using the supplied operands. It does not verify payroll, establish savings, or approve a finding.
Return to the source quotation
Explore three views of this finding

Read the finding, inspect its simplified JSON or review the check notes. These MD, JSON and TXT views illustrate the same example; they are not the customer’s Evidence Receipt, Knowledge Graph and Audit Report package.

ONE FINDING / THREE EXAMPLE VIEWSIllustrative preview
FINDING / QUALIFIED CONCLUSIONMD
THE AGENT’S RECOMMENDATION

Investigate order entry before estimating savings.

Interview figures imply about $594,000 per year in staffing cost. That supports investigating the workflow; it does not establish how much could be saved.

SOURCE EXCERPT“We have eleven people in order entry. Fully loaded, each of them is about $54,000 a year.”
Keep the qualification

The per-person amount is approximate. Actual payroll, achievable savings and possible overlap with a separate cost line remain unconfirmed.

Keep the basis after the run.MD / JSON / TXT example views · not a receipt, graph export or full Audit Report.

Explore technical exports

Graph JSON & Cypher

KPA’s graph snapshot is JSON; its full audit and source companion are separate. Neo4j/Cypher tooling depends on the product and access.

Explore KPA exports

The audit and source companion

Keep the source context, recorded findings, qualifications and open questions alongside the graph. The available files depend on the product and access.

Explore the retained KPA record
Before the next person or agent uses the finding

Keep the source references, inputs, interpretation, qualifications and recorded refusals together. A later reviewer can inspect that basis without asking a changed model to recreate the original answer.

Having the record does not authorize every new use. Sentinel checks integrity, meaning and current-use authority in workflows configured to enforce those controls.

Try the separate claim-use example
About this example

The source quotation and staffing calculation simplify one saved synthetic record. The $210,000 amount, $450,000 budget and $75,000 threshold illustrate the separate question of what may be combined; this panel adds no receiving-test result. Source correspondence and arithmetic checks do not independently verify payroll, establish savings or prove a recommendation correct. Export reuse requires a compatible receiver and the required source companion; the demonstrated check used a separate tool, not an independent human review.

28 OCTOBER 2026 / ONE SYNTHETIC CASE

A separate receiving tool checked source links, repeated the calculation, and rejected altered inputs.

Inspect the receiving test
The receiving test: results and limits

The hosted full-JSON export matched the substantive payload previously accepted by a separate, confined receiving tool; the receiver was not rerun on the downloaded file. Using a prepared synthetic graph and a separately supplied source companion, that tool checked source correspondence and recomputed 11 × $54,000 ≈ $594,000 per year without a model.

Nine declared checks passed. Seven deliberately altered versions—source text, speaker, amount, unit, operand value, formula text, and unsupported promotion—were rejected. Unsupported entries remained quarantined; legacy unverified entries stayed unverified.

This was one synthetic case with assistant-operated checks, not independent human validation. Reuse requires a compatible receiver, the source companion, and the specified check rules. The calculation does not verify payroll or establish savings.

Read the MCOS technical brief

AUDITABILITY / REPRODUCIBILITY

What your next tool needs.

Keep the saved inputs, original source files and check rules together so another tool can inspect the work.

Calculations can be recomputed; judgments can differ. The next tool must support the exported records and specified checks.

Inspect the check requirements and limits

What a receiving check needs

Retained inputs
The engagement record, original source references, calculation operands, formulas and stated qualifications.
Compatible receiver
A receiver that can inspect the export and apply the specified check rules using the required source companion.
Declared scope
Repeat the supported source and calculation checks. A successful check does not verify every business fact or establish savings.

What stays when the model changes

Across platforms: the retained record is separate from the proprietary runtime. A compatible receiver applies its own checks using the required source companion.

Across models: retain the evidence and check rules when comparing model interpretations. Calculations can be recomputed; judgments can differ. Provider choice and model integration remain path-specific.

Read the audit and portability evidence
How this fits alongside agent certification

Agent/system assurance and claim-level controls are complementary; reliability and accountability overlap.

THE CONSULTING PLATFORM

MCOS

What supports this business conclusion?

Structures inputs, typed evidence, model-assisted analysis, and Layer 4 deliverables.

Inspectable work. Retained knowledge.
OUR EVIDENCE-USE APPLICATION

Sentinel

May this reading support this use now?

Uses Claimlatch’s Evidence Passports to separate integrity, meaning, and current-use authority.

Hold → fresh supported check → separate release
AN EXTERNAL CERTIFICATION STANDARD

AIUC-1

How does the scoped agent/system meet the standard?

Third-party testing and audits of behavior and controls for security, safety, reliability, and related risks.

Independent audit report and certificate.

Why an exact quotation needs another check

A reading can quote a clause exactly and still omit an exception or change a trigger. Sentinel checks meaning separately before approval. Dated amendments can supersede a reading; adverse meaning findings can hold reuse. A hold requires a fresh supported check and a separate release. Signed history stays intact.

This comparison implies no AIUC-1 certification, endorsement, or insurance coverage for MCOS or Sentinel. Insurance is separately underwritten.

Offline boundary: signed exports support integrity verification under the supplied trust. Only enforcing consumers see and act on current-use holds.

FROM THE RECORD TO ITS NEXT USE

Try a claim.
Check its next use.

A budget quotation can be intact and still fail to support the proposed use. Explore the three separate checks in this fictional example.

Try the claim checks.

One source quotation. One proposed use.

ONE CLAIM · THREE SEPARATE CHECKS

Traceability is the starting point. Permission is a separate check.

MCOS keeps the documented basis inspectable after the run. Sentinel addresses a further question: may this particular interpretation support the agent’s intended use now?

SEPARATE FICTIONAL BUDGET EXAMPLE Explore the checks; this is not a live verification.

SOURCE QUOTATION
“Budget up to $300,000, subject to board approval.”
AGENT’S PROPOSED CLAIM
“An approved $300,000 budget is available.”

Intended use: another agent treats the amount as approved funding.

Does the agent’s quote match? Is the record intact?

Intact in this scenario

The words match. That is one check.

For this fictional scenario, assume the quotation matches the source and the recorded evidence and signatures are intact. This establishes integrity in the example; it does not establish that the budget is approved.

Integrity preserves the connection to the source. Meaning and current-use authority remain separate questions.

What if a later review finds a problem?

A reading that was previously permitted can be held when a later meaning review finds a problem. In workflows that enforce these controls, reuse pauses while the original signed record stays intact.

  1. 1Hold reuse

    Pause use of the affected reading. Preserve what was signed and recorded.

  2. 2Fresh supported check

    Review new meaning evidence. A fresh check does not release the hold by itself.

  3. 3Separate release

    An explicit release decision is required before the held reading can be used again.

  4. 4Check at use again

    Released readings face integrity, meaning, and current-use checks again at each use.

View the full reuse flow
REUSABLE EVIDENCE PASSPORTSSENTINEL
01 / Integrity02 / Meaning03 / Current-use authority
LATER ADVERSE MEANINGAttested readingEvidence PassportCheck current useIntegrity · meaning · authorityPermitted useWithin the checked scopeHold reuseRecorded history preservedFresh supported checkNew meaning evidenceSeparate releaseExplicit release decisionLATER ADVERSE MEANINGReadingPassportCheck use3 assurancesUseIn scopeHoldReuse pausedRecheckSupportedReleaseSeparate act
A fresh check does not release a hold by itself. Released readings face use-time checks again.

Illustrative outcomes, separate from the MCOS receiving test above. These checks do not prove a recommendation correct or verify a real budget. Model-based meaning checks support review; they do not replace independent expert validation.

DESIGN / BUILD / GOVERN

Bring the work
your business needs done.

Start with one decision, process or knowledge handover. We’ll agree the inputs, domain rules, deliverables, review authority and acceptance checks for a scoped engagement.

Explore the full workflowTranscript input, MCOS stages, model choices and the records you retain

THE WORK / THE RECORD / THE NEXT USE

Different work.
One shared foundation.

Start with the transcript. MCOS structures the evidence and analysis. Keep the basis after the run, with controls for the next use.

Scroll to explore. Select any area for a closer look.
01 / START WITH THE TRANSCRIPTCold start · Existing supported protocol

Fictional transcript excerpt · Supplied transcript input; meeting logos do not imply a direct integration.

02 / STRUCTURE THE WORKShared L1–L3 foundation
03 / APPLY THE DOMAIN PROTOCOLExisting applications + custom protocols

Shared MCOS foundation. Domain-specific configuration and acceptance checks.

04 / KEEP THE BASISConcrete artifacts, separate from the runtime
Evidence ReceiptFigures used, source quotations, calculation inputs and recorded checks within the receipt’s scope.
Knowledge GraphThe available nodes, relationships and structured fields in an exported snapshot.
Audit ReportRecorded processing steps, checks, results and available cost information.

Formats and field coverage vary by protocol. Cypher tooling and additional exports depend on product access and engagement scope.

05 / INSPECT & REPEATFor the auditor, underwriter or next reviewer
ONE WORKED EXAMPLE

Follow the source, calculation and limits in the staffing example.

Inspect the worked example

Specified checks · Compatible receiver · Required source companion. Recomputed arithmetic does not establish source truth, savings or a correct recommendation.

Conceptual architecture · Controls configured per workflow

TranscriptMCOSprotocolsrecordAudit & repeat
Explore custom-domain examplesHow a shared foundation can be configured for different business requirements

APPLY THE SHARED FOUNDATION

Your domain.
Your supported workflow.

Explore how the shared foundation can be configured for different business requirements.

MCOS + SENTINEL / ACROSS DOMAINSExplore an example
View the cross-domain illustration and system roles
Conceptual architecture: a three-layer MCOS core and a separate Sentinel claim-review gateway connect through agent and model nodes to logistics, environmental monitoring and knowledge-handover scenes. Both systems span the domains; the illustration depicts a configurable approach.
MCOS + SENTINEL / SHARED ACROSS DOMAINSSeparate systems. Domain-specific configuration. Conceptual illustration.View illustration full size

MCOS structures the work.

Source handling, typed evidence and scoped agent analysis. Domain protocols shape the questions and outputs.

Sentinel checks the next use.

Integrity, meaning and current-use authority stay separate. Claim controls are configured for the intended purpose.

Models serve the workflow.

Choose compatible agents and models for the task. Keep the retained record separate from the runtime.

Explore the shared L1–L3 foundation
L4 / DOMAIN PROTOCOLProposed application

Delivery exception review

Understand where delivery exceptions accumulate.

Supplied material
Dispatch logs, shipment records, invoices and operating procedures.
Configure L2 + L3
Shipment references, stated timestamps, quantities, costs and service requirements.

Examine documented delays and dependencies. Keep proposed explanations distinct from recorded events.

Qualified deliverable
A qualified exception review with source links, assumptions and open questions.
What would we agree and validate?

Agree event coverage, matching periods, duplicate-cost checks and who may approve a proposed operational change.

Logistics and environmental examples are proposed applications. The KPA handover is an illustrative example of an existing application. Each engagement needs its own supported scope and acceptance checks. MCOS and Sentinel have separate evidence and controls; integration, enforcement and model compatibility depend on the implemented workflow.
Explore the source-to-record architectureMCOS stages, retained evidence and separate Sentinel controls
AGENTIC WORK / RETAINED EVIDENCE

Trace the work. Govern the reading.

Agents perform the work. MCOS retains its documented basis. Sentinel governs claim use.

EXPLORE MCOS + SENTINEL

MCOSStructure agent work. Retain its evidence.

SentinelGovern claim use across agent workflows.

Conceptual architecture: four MCOS process layers lead to retained evidence. A claim from the record is examined through separate Sentinel controls for integrity, meaning, and current-use authority.

Choose a stage or claim check to explore.Blue marks the selected view.

MCOS What the agent’s work rested on

Sentinel Whether a particular claim may be used

MCOS / L1 / INPUT

Keep the original source and its context.

Preserve supplied material, speaker turns and source spans before agent interpretation, so later agents can trace the work to its original inputs.

FOLLOW THE RETAINED RECORDSee one record in practice

Conceptual view. Human review, approval and intervention depend on the workflow, implemented controls and permissions. MCOS and Sentinel have separate checks and evidence; integration is workflow-specific. Retained records do not make arbitrary agents or models interchangeable.

Explore the layer contractsMCOS · the structure behind the retained record

Typed contracts structure what crosses L1–L4. The supported workflow preserves source references and applies selected checks, including financial checks within their specified scope. A later review inspects the retained record. These checks do not establish the truth of every claim.

TYPED CONTRACTS ACROSS L1–L4MCOS
L1InputSpeaker · contextL2ExtractionTyped facts · operandsL3AnalysisModel-assisted judgmentL4DeliverableProtocol-specific outputLOCATEPROJECTASSEMBLEPOST-DELIVERABLEReview the recordTrace sourcesRecompute calculationsReview qualificationsL1InputSpeaker · contextL2ExtractionTyped facts · operandsL3AnalysisModel-assisted judgmentL4DeliverableProtocol-specific outputPOST-DELIVERABLEReview the recordTrace sources · recompute calculationsReview qualifications
Typed contracts apply across L1–L4. The post-deliverable review inspects the retained record; supported checks and review scope vary by protocol.

Sentinel implements Claimlatch’s reusable profile. Integration and enforcement depend on the implemented workflow.

THE RESEARCH FOUNDATION

The architecture.
The evidence behind it.

Two illustrated technical briefs. Practical examples, mechanisms and their limits.

01 / MCOS

TECHNICAL BRIEF / NOVEMBER 2026

MCOS

Keep the Evidence Behind the Answer

Follow the four layers, a worked staffing calculation and the record a later agent can inspect.

Read the MCOS briefPDF
02 / SENTINEL

TECHNICAL BRIEF / NOVEMBER 2026

Sentinel

Before the Next Agent Acts

Explore integrity, meaning and current-use authority, with an example of holds, rechecks and release.

Read the Sentinel briefPDF

Informative briefs with scoped synthetic examples and assistant-operated checks. Universal accuracy and certification are not claimed.

Explore the briefs and operational sovereignty