Skip to content

Site under construction, formerly SigmaX has begun the process of transitioning to AIgentXS.

Governed AI Transactions
Review a workflow

AI Compute + InferenceSigmaΛ

Know what can run. Account for what did.

Carry a workload's data boundary into admission review. Then compare what a provider reported with a characterized meter, before a separate commercial review.

Review a workflow Discuss your workflow

Explore a fictional example here. Connecting your organization starts with a scoped conversation.

AI Compute + InferenceArchitecture study
From workload to witnessed effectAdmission / metering / reconciliation
Illustrative sample — not client data

A passport for one workload.

A permitted request, a delivery claim, and authority to pay are different things. Inspect each boundary without dispatching a workload.

BURST DEMO-LAMBDA-3310 · inference queue relief · CP-07 v9

Two ledgers. One permission gap.

Illustrative sample — not client data

Admission · before dispatch

Fictional CP-07 v9 boundary recorded at 01:58.

Delivery · provider meter

Fictional report retrieved at 09:04. No provider queried.

GPU-hours · admitted400
=
GPU-hours · reported400
Region · admittedus-east-2
=
Region · reportedus-east-2
Hardware class · admittedA100-80GB
=
Hardware class · reportedA100-80GB
Window · fictional UTC · admitted02:00–08:00
=
Window · fictional UTC · reported02:04–07:58
Cost · admitted$6,800 ceiling
=
Cost · reported$6,742
Data residency · admittedUS only
=
Data residency · reportedUS only

= means the reported value fits the admitted boundary; a ceiling or time window need not be numerically identical. ≠ marks a failed boundary. ? means evidence is unavailable or not comparable.

The gap

GPU-hour variance
0
Against the cost ceiling
−$58
Hours above admitted quantity
0
Residency exposure
None in fixture

What a spend dashboard would show

−$58 against the admitted ceiling. Cost alone does not describe the other boundaries.

This burst comparison uses the fictional CP-07 v9 capacity policy. The deeper lifecycle example below uses its own W-5022 planning policy. Neither provisions capacity, verifies real execution, or authorizes payment.

Keep the explanation with the decision

A record you can inspect.

Follow the evidence behind this example, compare its revisions, or take the context into your next conversation.

DEMO-LAMBDA-3310-v1Input version 1

Proposed action

Compare admission and provider-meter ledgers for fictional burst DEMO-LAMBDA-3310

Evidence
2 sources2 declared present
Policy
CP-07 v96 pass
Review owner
Platform operations reviewerneeded

Next step

Inspect the two fictional sources. Agreement is not independent verification of execution or authority to pay.

Fictional record · no external action

Find a decision in the sample recordsTrace an action, a reviewer, or a reason back to its record.

Only the fictional records supplied with this example are searchable. No customer records are queried.

3 sample records found

Prepare a workflow briefBring this example into a conversation about your team's workflow.

Sketch the workflow before a conversation. Use general descriptions; leave out personal data, credentials, and customer documents. This draft stays in memory until you leave or reload.

Keep this example as context

Compare admission and provider-meter ledgers for fictional burst DEMO-LAMBDA-3310

Fictional policy CP-07 v9 · Platform operations reviewer

The review includes the current inputs, findings, evidence and next step, separately labeled from your proposed workflow.

Illustrative sample — not client data

The decision record

DEMO-LAMBDA-3310-v1 · input version 1

Proposed action

Compare admission and provider-meter ledgers for fictional burst DEMO-LAMBDA-3310

Inputs checked
Fictional burst
DEMO-LAMBDA-3310
Provider meter source
present
Admission · GPU-hours
400
Provider meter · GPU-hours
400
Admission · Region
us-east-2
Provider meter · Region
us-east-2
Admission · Hardware class
A100-80GB
Provider meter · Hardware class
A100-80GB
Admission · Window · fictional UTC
02:00–08:00
Provider meter · Window · fictional UTC
02:04–07:58
Admission · Cost
$6,800 ceiling
Provider meter · Cost
$6,742
Admission · Data residency
US only
Provider meter · Data residency
US only
Review owner
Platform operations reviewer
Review state
needed
Policy version
CP-07 v9 · fictional
Next step
Inspect the two fictional sources. Agreement is not independent verification of execution or authority to pay.

What the rules found

  • pass

    GPU-hours: admitted 400; provider reports 400. Inside the admitted boundary.

    Rule burst-hours · Sources: burst-admission, burst-meter
  • pass

    Region: admitted us-east-2; provider reports us-east-2. Inside the admitted boundary.

    Rule burst-region · Sources: burst-admission, burst-meter
  • pass

    Hardware class: admitted A100-80GB; provider reports A100-80GB. Inside the admitted boundary.

    Rule burst-hardware · Sources: burst-admission, burst-meter
  • pass

    Window · fictional UTC: admitted 02:00–08:00; provider reports 02:04–07:58. Inside the admitted boundary.

    Rule burst-window · Sources: burst-admission, burst-meter
  • pass

    Cost: admitted $6,800 ceiling; provider reports $6,742. Inside the admitted boundary.

    Rule burst-cost · Sources: burst-admission, burst-meter
  • pass

    Data residency: admitted US only; provider reports US only. Inside the admitted boundary.

    Rule burst-residency · Sources: burst-admission, burst-meter

Evidence and unresolved items

  • Fictional CP-07 v9 admission at 01:58 before the proposed dispatch; US-only committed databurst-admission · declared present · fictional source
  • Fictional provider-meter export for DEMO-LAMBDA-3310, retrieved at 09:04. Provider reporting is not independent proof.burst-meter · declared present · fictional source

How this example reached its current state

  1. Opened fictional burst DEMO-LAMBDA-3310. No provider was contacted.
Authorization
None
Execution
Not attempted
Observed effect
Not observed
Integrity verification
Not performed

This page holds a local sample, not an original customer audit record. Downloading it does not sign it, persist it to your organization, or prove an external action.

Technical record
{
  "schemaVersion": "design-example-v1",
  "example": true,
  "sector": "ai-compute-inference",
  "scenarioId": "burst-ledgers",
  "recordId": "DEMO-LAMBDA-3310-v1",
  "action": {
    "summary": "Compare admission and provider-meter ledgers for fictional burst DEMO-LAMBDA-3310",
    "target": "Inference queue relief · CP-07 v9",
    "inputVersion": 1,
    "inputs": [
      {
        "id": "burst-case",
        "label": "Fictional burst",
        "value": "DEMO-LAMBDA-3310"
      },
      {
        "id": "meter-source",
        "label": "Provider meter source",
        "value": "present"
      },
      {
        "id": "admitted-hours",
        "label": "Admission · GPU-hours",
        "value": "400"
      },
      {
        "id": "delivered-hours",
        "label": "Provider meter · GPU-hours",
        "value": "400"
      },
      {
        "id": "admitted-region",
        "label": "Admission · Region",
        "value": "us-east-2"
      },
      {
        "id": "delivered-region",
        "label": "Provider meter · Region",
        "value": "us-east-2"
      },
      {
        "id": "admitted-hardware",
        "label": "Admission · Hardware class",
        "value": "A100-80GB"
      },
      {
        "id": "delivered-hardware",
        "label": "Provider meter · Hardware class",
        "value": "A100-80GB"
      },
      {
        "id": "admitted-window",
        "label": "Admission · Window · fictional UTC",
        "value": "02:00–08:00"
      },
      {
        "id": "delivered-window",
        "label": "Provider meter · Window · fictional UTC",
        "value": "02:04–07:58"
      },
      {
        "id": "admitted-cost",
        "label": "Admission · Cost",
        "value": "$6,800 ceiling"
      },
      {
        "id": "delivered-cost",
        "label": "Provider meter · Cost",
        "value": "$6,742"
      },
      {
        "id": "admitted-residency",
        "label": "Admission · Data residency",
        "value": "US only"
      },
      {
        "id": "delivered-residency",
        "label": "Provider meter · Data residency",
        "value": "US only"
      }
    ]
  },
  "policy": {
    "version": "CP-07 v9",
    "origin": "fictional_organization_policy",
    "findings": [
      {
        "ruleId": "burst-hours",
        "status": "pass",
        "explanation": "GPU-hours: admitted 400; provider reports 400. Inside the admitted boundary.",
        "sourceRefs": [
          "burst-admission",
          "burst-meter"
        ]
      },
      {
        "ruleId": "burst-region",
        "status": "pass",
        "explanation": "Region: admitted us-east-2; provider reports us-east-2. Inside the admitted boundary.",
        "sourceRefs": [
          "burst-admission",
          "burst-meter"
        ]
      },
      {
        "ruleId": "burst-hardware",
        "status": "pass",
        "explanation": "Hardware class: admitted A100-80GB; provider reports A100-80GB. Inside the admitted boundary.",
        "sourceRefs": [
          "burst-admission",
          "burst-meter"
        ]
      },
      {
        "ruleId": "burst-window",
        "status": "pass",
        "explanation": "Window · fictional UTC: admitted 02:00–08:00; provider reports 02:04–07:58. Inside the admitted boundary.",
        "sourceRefs": [
          "burst-admission",
          "burst-meter"
        ]
      },
      {
        "ruleId": "burst-cost",
        "status": "pass",
        "explanation": "Cost: admitted $6,800 ceiling; provider reports $6,742. Inside the admitted boundary.",
        "sourceRefs": [
          "burst-admission",
          "burst-meter"
        ]
      },
      {
        "ruleId": "burst-residency",
        "status": "pass",
        "explanation": "Data residency: admitted US only; provider reports US only. Inside the admitted boundary.",
        "sourceRefs": [
          "burst-admission",
          "burst-meter"
        ]
      }
    ]
  },
  "evidence": [
    {
      "id": "burst-admission",
      "state": "declared_present",
      "sourceLabel": "Fictional CP-07 v9 admission at 01:58 before the proposed dispatch; US-only committed data",
      "sample": true
    },
    {
      "id": "burst-meter",
      "state": "declared_present",
      "sourceLabel": "Fictional provider-meter export for DEMO-LAMBDA-3310, retrieved at 09:04. Provider reporting is not independent proof.",
      "sample": true
    }
  ],
  "review": {
    "ownerRole": "Platform operations reviewer",
    "status": "needed"
  },
  "authorization": "none",
  "externalAction": false,
  "effect": "not_observed",
  "integrityVerification": "not_performed",
  "nextStep": "Inspect the two fictional sources. Agreement is not independent verification of execution or authority to pay.",
  "history": [
    {
      "sequence": 1,
      "summary": "Opened fictional burst DEMO-LAMBDA-3310. No provider was contacted.",
      "sample": true
    }
  ]
}
Explore admission, delivery, and commercial lifecycle checks

DEMO-LAMBDA-5022

This separate W-5022 example retains its own planning mandate, characterized meter comparison, pricing terms, and historical review snapshots.

Request
W-5022 Model fixture M1 · confidential inference
Proposed destination
US East Fallback: US West
Planning estimate
$450 Fictional USD · not a purchase
Data boundary
Metadata only No prompts · proposed 3-day retention

The cheapest fallback can still be excluded.

Keep the confidential workload inside its permitted region. A fictional EU West fallback quote of $320 does not change that boundary.

  1. US East

    Within the selected region scope

    Proposed primary destination

  2. US West

    Within the selected region scope

    Proposed fallback destination

  3. EU West

    Excluded by the organization’s US-only boundary

    Alternative destination

The route diagram is a local policy comparison. No provider availability, network route, or compute inventory is queried.

Admission inputs

Record a sample review of the proposed route. No dispatch occurs.

Keep the explanation with the decision

A record you can inspect.

Follow the evidence behind this example, compare its revisions, or take the context into your next conversation.

DEMO-LAMBDA-5022-admission-v1Input version 1

Proposed action

Check a confidential workload before dispatch

Evidence
6 sources6 declared present
Policy
LAMBDA-M110 pass · 2 fail
Review owner
Tenant workload authorityneeded

Next step

Record a sample review of the proposed route. No dispatch occurs.

Fictional record · no external action

Find a decision in the sample recordsTrace an action, a reviewer, or a reason back to its record.

Only the fictional records supplied with this example are searchable. No customer records are queried.

3 sample records found

Prepare a workflow briefBring this example into a conversation about your team's workflow.

Sketch the workflow before a conversation. Use general descriptions; leave out personal data, credentials, and customer documents. This draft stays in memory until you leave or reload.

Keep this example as context

Check a confidential workload before dispatch

Fictional policy LAMBDA-M1 · Tenant workload authority

The review includes the current inputs, findings, evidence and next step, separately labeled from your proposed workflow.

Customize the sample mandateCompare a rule change before applying it to this example.

Define the boundary in your own terms.

Edit this fictional mandate, compare the changes, then apply a local version. Existing organizational caps still apply.

Fictional policy · LAMBDA-M1
Why does this matter?

The organization always excludes EU West for this confidential workload, including fallback.

Source: fictional organization policy for this example.

Atlas guide · scripted example

This is a written guide, not a model response. Start with the action you want to permit, name the evidence required, and assign the person who can resolve a gap. Keep data permissions narrow. A draft that passes these fictional checks is not an approved organizational policy.

Changes stay in this page session. No customer rule is saved, approved, or sent to an external model.

Illustrative sample — not client data

The decision record

DEMO-LAMBDA-5022-admission-v1 · input version 1

Proposed action

Check a confidential workload before dispatch

Inputs checked
Proposed workload
W-5022 · model fixture M1
Requested region
US East
Fallback destination
US West
Planning estimate in USD
450
Proposed metadata retention in days
3
Permitted destinations
US East and US West
Selected admission estimate ceiling in USD
500
Selected maximum metadata retention in days
7
Admission review ownership
Tenant workload authority
Prompt content included
Off
Required review owner
Tenant workload authority
Review owner
Tenant workload authority
Review state
needed
Policy version
LAMBDA-M1 · fictional
Next step
Record a sample review of the proposed route. No dispatch occurs.

What the rules found

  • pass

    Admission values are valid.

    Rule admission-input · No supporting source attached
  • pass

    The primary and optional fallback destinations match the US-only boundary and selected mandate.

    Rule admission-region · Sources: LAMBDA-ORG, LAMBDA-REQUEST
  • pass

    The estimate fits the selected limit and $500 fictional organization cap.

    Rule admission-budget · Sources: LAMBDA-ORG, LAMBDA-REQUEST
  • pass

    Metadata-only proposal: 3 days of retention, within both limits.

    Rule admission-data · Sources: LAMBDA-ORG, LAMBDA-REQUEST
  • pass

    Meter quantities have valid finite values.

    Rule delivery-input · Sources: LAMBDA-PROVIDER, LAMBDA-TENANT
  • pass

    The fictional tenant-meter source has full window coverage and current provenance.

    Rule delivery-source · Sources: LAMBDA-TENANT
  • pass

    Both fixtures refer to workload W-5022, GPU-hours on GPU class A, September 8 10:00–11:00 UTC, with full coverage.

    Rule delivery-comparability · Sources: LAMBDA-PROVIDER, LAMBDA-TENANT
  • fail

    12 GPU-hours unresolved; provider reporting exceeds tenant observation.

    Rule delivery-quantity · Sources: LAMBDA-PROVIDER, LAMBDA-TENANT
  • pass

    Claim quantities and pricing have valid values.

    Rule commercial-input · Sources: LAMBDA-CLAIM
  • pass

    Claim pricing matches fictional accepted terms v1: $2 per GPU-hour.

    Rule commercial-terms · Sources: LAMBDA-TERMS, LAMBDA-CLAIM
  • pass

    Claim quantity matches the provider report; delivery agreement is still a separate prerequisite.

    Rule commercial-quantity · Sources: LAMBDA-CLAIM, LAMBDA-PROVIDER
  • fail

    Resolve the admission and delivery findings before the commercial claim can be ready for review.

    Rule commercial-prerequisites · Sources: LAMBDA-ORG, LAMBDA-PROVIDER, LAMBDA-TENANT

Evidence and unresolved items

  • Fictional organization: US regions only; $500 estimate cap; 7-day metadata cap; no prompt content.LAMBDA-ORG · declared present · fictional source
  • W-5022 / model fixture M1 / us-east / fallback us-west / estimate $450 / metadata 3 days.LAMBDA-REQUEST · declared present · fictional source
  • Provider fixture: 412 GPU-hours, class A, W-5022, September 8 10:00–11:00 UTC. Not independent attestation.LAMBDA-PROVIDER · declared present · fictional source
  • Tenant fixture: 400; comparison same; source current; 12 hours old; coverage 100% if available.LAMBDA-TENANT · declared present · fictional source
  • Accepted-terms fixture v1: $2/GPU-hour. Selected source: v1. Proposed v2 ($2.50) has no accepted amendment.LAMBDA-TERMS · declared present · fictional source
  • Claim fixture: 412 gpu-hours at $2 per unit. No invoice sent or paid.LAMBDA-CLAIM · declared present · fictional source

How this example reached its current state

  1. Opened a fictional confidential workload. No prompt content or provider connection.
Authorization
None
Execution
Not attempted
Observed effect
Not observed
Integrity verification
Not performed

This page holds a local sample, not an original customer audit record. Downloading it does not sign it, persist it to your organization, or prove an external action.

Technical record
{
  "schemaVersion": "design-example-v1",
  "example": true,
  "sector": "ai-compute-inference",
  "scenarioId": "admission",
  "recordId": "DEMO-LAMBDA-5022-admission-v1",
  "action": {
    "summary": "Check a confidential workload before dispatch",
    "target": "Fictional workload W-5022",
    "inputVersion": 1,
    "inputs": [
      {
        "id": "workload",
        "label": "Proposed workload",
        "value": "W-5022 · model fixture M1"
      },
      {
        "id": "requested-region",
        "label": "Requested region",
        "value": "US East"
      },
      {
        "id": "fallback-region",
        "label": "Fallback destination",
        "value": "US West"
      },
      {
        "id": "planning-estimate-usd",
        "label": "Planning estimate in USD",
        "value": "450"
      },
      {
        "id": "proposed-retention-days",
        "label": "Proposed metadata retention in days",
        "value": 3
      },
      {
        "id": "permitted-destinations",
        "label": "Permitted destinations",
        "value": "US East and US West"
      },
      {
        "id": "estimate-ceiling-usd",
        "label": "Selected admission estimate ceiling in USD",
        "value": 500
      },
      {
        "id": "retention-limit-days",
        "label": "Selected maximum metadata retention in days",
        "value": 7
      },
      {
        "id": "admission-review-owner",
        "label": "Admission review ownership",
        "value": "Tenant workload authority"
      },
      {
        "id": "prompt-content",
        "label": "Prompt content included",
        "value": false
      },
      {
        "id": "review-owner",
        "label": "Required review owner",
        "value": "Tenant workload authority"
      }
    ]
  },
  "policy": {
    "version": "LAMBDA-M1",
    "origin": "fictional_organization_policy",
    "findings": [
      {
        "ruleId": "admission-input",
        "status": "pass",
        "explanation": "Admission values are valid.",
        "sourceRefs": []
      },
      {
        "ruleId": "admission-region",
        "status": "pass",
        "explanation": "The primary and optional fallback destinations match the US-only boundary and selected mandate.",
        "sourceRefs": [
          "LAMBDA-ORG",
          "LAMBDA-REQUEST"
        ]
      },
      {
        "ruleId": "admission-budget",
        "status": "pass",
        "explanation": "The estimate fits the selected limit and $500 fictional organization cap.",
        "sourceRefs": [
          "LAMBDA-ORG",
          "LAMBDA-REQUEST"
        ]
      },
      {
        "ruleId": "admission-data",
        "status": "pass",
        "explanation": "Metadata-only proposal: 3 days of retention, within both limits.",
        "sourceRefs": [
          "LAMBDA-ORG",
          "LAMBDA-REQUEST"
        ]
      },
      {
        "ruleId": "delivery-input",
        "status": "pass",
        "explanation": "Meter quantities have valid finite values.",
        "sourceRefs": [
          "LAMBDA-PROVIDER",
          "LAMBDA-TENANT"
        ]
      },
      {
        "ruleId": "delivery-source",
        "status": "pass",
        "explanation": "The fictional tenant-meter source has full window coverage and current provenance.",
        "sourceRefs": [
          "LAMBDA-TENANT"
        ]
      },
      {
        "ruleId": "delivery-comparability",
        "status": "pass",
        "explanation": "Both fixtures refer to workload W-5022, GPU-hours on GPU class A, September 8 10:00–11:00 UTC, with full coverage.",
        "sourceRefs": [
          "LAMBDA-PROVIDER",
          "LAMBDA-TENANT"
        ]
      },
      {
        "ruleId": "delivery-quantity",
        "status": "fail",
        "explanation": "12 GPU-hours unresolved; provider reporting exceeds tenant observation.",
        "sourceRefs": [
          "LAMBDA-PROVIDER",
          "LAMBDA-TENANT"
        ]
      },
      {
        "ruleId": "commercial-input",
        "status": "pass",
        "explanation": "Claim quantities and pricing have valid values.",
        "sourceRefs": [
          "LAMBDA-CLAIM"
        ]
      },
      {
        "ruleId": "commercial-terms",
        "status": "pass",
        "explanation": "Claim pricing matches fictional accepted terms v1: $2 per GPU-hour.",
        "sourceRefs": [
          "LAMBDA-TERMS",
          "LAMBDA-CLAIM"
        ]
      },
      {
        "ruleId": "commercial-quantity",
        "status": "pass",
        "explanation": "Claim quantity matches the provider report; delivery agreement is still a separate prerequisite.",
        "sourceRefs": [
          "LAMBDA-CLAIM",
          "LAMBDA-PROVIDER"
        ]
      },
      {
        "ruleId": "commercial-prerequisites",
        "status": "fail",
        "explanation": "Resolve the admission and delivery findings before the commercial claim can be ready for review.",
        "sourceRefs": [
          "LAMBDA-ORG",
          "LAMBDA-PROVIDER",
          "LAMBDA-TENANT"
        ]
      }
    ]
  },
  "evidence": [
    {
      "id": "LAMBDA-ORG",
      "state": "declared_present",
      "sourceLabel": "Fictional organization: US regions only; $500 estimate cap; 7-day metadata cap; no prompt content.",
      "sample": true
    },
    {
      "id": "LAMBDA-REQUEST",
      "state": "declared_present",
      "sourceLabel": "W-5022 / model fixture M1 / us-east / fallback us-west / estimate $450 / metadata 3 days.",
      "sample": true
    },
    {
      "id": "LAMBDA-PROVIDER",
      "state": "declared_present",
      "sourceLabel": "Provider fixture: 412 GPU-hours, class A, W-5022, September 8 10:00–11:00 UTC. Not independent attestation.",
      "sample": true
    },
    {
      "id": "LAMBDA-TENANT",
      "state": "declared_present",
      "sourceLabel": "Tenant fixture: 400; comparison same; source current; 12 hours old; coverage 100% if available.",
      "sample": true
    },
    {
      "id": "LAMBDA-TERMS",
      "state": "declared_present",
      "sourceLabel": "Accepted-terms fixture v1: $2/GPU-hour. Selected source: v1. Proposed v2 ($2.50) has no accepted amendment.",
      "sample": true
    },
    {
      "id": "LAMBDA-CLAIM",
      "state": "declared_present",
      "sourceLabel": "Claim fixture: 412 gpu-hours at $2 per unit. No invoice sent or paid.",
      "sample": true
    }
  ],
  "review": {
    "ownerRole": "Tenant workload authority",
    "status": "needed"
  },
  "authorization": "none",
  "externalAction": false,
  "effect": "not_observed",
  "integrityVerification": "not_performed",
  "nextStep": "Record a sample review of the proposed route. No dispatch occurs.",
  "history": [
    {
      "sequence": 1,
      "summary": "Opened a fictional confidential workload. No prompt content or provider connection.",
      "sample": true
    }
  ]
}
Telemetry has a source and a boundary

OpenTelemetry's May 2026 GenAI observability article distinguishes model/usage metadata from optional sensitive content capture. That informs the metadata-only design here; it does not attest to delivery, certify this page, or establish a live integration.

From your workflow to a scoped implementation

Start with the decision your team already owns.

Bring one proposed action, the rule it must satisfy, and the person responsible. Together, we can map the evidence and define what a useful first test would need to demonstrate.

Plan a ai compute & inference workflow

A conversation about scope and fit. This page does not provision a connector or start a customer trial.