Start with your assurance question. Find the first source that matters, check what it can establish, and see exactly where the proof stops.
Pick a real assurance question. This guide recommends an evidence path and shows what it still cannot resolve. No live account or system is inspected here.
See which source to start with, what it can establish, and what should remain unassessed.
Choose a source for the question it can help answer, not simply because it is connected. A relevant source may inform a claim without independently proving it. MCP is a workflow interface and the target system is often its own evidence owner.
Source availability alone is not qualified proof. Each claim requires a matching identity, scope, evidence producer and evaluation basis.
| Source | What it may establish | What it cannot establish alone |
|---|---|---|
| GitHub Qualified source when preflight and source execution succeed | Supported agent, tool, handler and consequential API paths in a verified, authorized repository snapshot. | Source reachability cannot establish policy approval, effective runtime credentials, execution or target effect. |
| Local source Executed local analysis may qualify; declarations are draft context | An executed local source check may inform a bounded code-path finding. Submitted declarations and design documents provide context only until independently qualified. | A declaration or file upload is not independently verified runtime truth. Local tooling and intake availability must be checked for the user environment. |
| AWS IAM Partial authority observation only | Partial AWS identity-policy, principal and candidate permission observations for the examined account. Other authorization layers are not fully modeled. | Present AWS observation is partial. Resource, session, endpoint and other authorization layers can affect real grants; an identity-policy allow does not alone establish canonical effectively granted access. |
| AWS logs Read-only cloud activity and source context | Account/region/time-window bounded read-only AWS resource and CloudWatch activity records, subject to source authorization and evidence qualification. | Adapter availability does not mean a customer has connected it. CloudWatch records are raw observations, not automatic control evaluations or target effects. |
| ServiceNow Instance-specific read-only source where ACL permits | Authorized read-only workflow, AICT asset, agent execution and table records, each with its own ACL, type, provenance and time basis. | Login does not prove every table permission. ACL denials do not mean empty evidence. No general approval writeback or complete ServiceNow estate discovery is implied. |
| OTLP Received telemetry is not yet evaluated assurance | Scoped telemetry records received and accepted through configured OTLP ingress, with producer attribution and available action correlation. | Collector configuration is not a live collector. HTTP 200 or accepted telemetry does not prove evaluation, action application, prevention, or a target-side business change. |
| HAIEC MCP Authorized HAIEC workflow interface, not an evidence source | Authorized agent-guided preflight, source discovery, evaluation initiation, report navigation and scoped reads when the registered HAIEC tools allow them. | MCP is a transport and interaction interface, not proof an external tool ran, that the target was protected, or that an unavailable source exists. Tool and credential scopes matter. |
| Policy and approval Approved policy/decision artifact when qualified | Approved capability scope, named accountable reviewer, policy version, expiry and action-bound authorization where qualified and connected. | Policy intent is separate from effective IAM, source capability and actual interception; a human acknowledgment is not necessarily a deployment approval. |
| Target records Potential authoritative target witness, customer-dependent | A target-owned action or transaction witness may establish applied, rejected or completed effects for an exact resource and time. | Not a universal, ready-made HAIEC integration. Requires authorized customer-native APIs/logs and identity join. A guard decision or dispatch acknowledgment alone does not prove target effect. |
| Provider inventory Provider admin inventory context only | Authorized provider project, model, usage and cost inventory. This is administrative context rather than an agent action request. | Administrative inventory does not prove the agent action path, MCP caller identity, protected resource authorization or observed external business effect. |
Explore each integration path with its conditions, onboarding steps and evidence boundary. These are source-specific capabilities, not evidence that your organization is connected. Check the exact preflight before treating a record as eligible.
Supported agent, tool, handler and consequential API paths in a verified, authorized repository snapshot.
Boundary: Source reachability cannot establish policy approval, effective runtime credentials, execution or target effect.
An executed local source check may inform a bounded code-path finding. Submitted declarations and design documents provide context only until independently qualified.
Boundary: A declaration or file upload is not independently verified runtime truth. Local tooling and intake availability must be checked for the user environment.
Partial AWS identity-policy, principal and candidate permission observations for the examined account. Other authorization layers are not fully modeled.
Boundary: Present AWS observation is partial. Resource, session, endpoint and other authorization layers can affect real grants; an identity-policy allow does not alone establish canonical effectively granted access.
Account/region/time-window bounded read-only AWS resource and CloudWatch activity records, subject to source authorization and evidence qualification.
Boundary: Adapter availability does not mean a customer has connected it. CloudWatch records are raw observations, not automatic control evaluations or target effects.
Authorized read-only workflow, AICT asset, agent execution and table records, each with its own ACL, type, provenance and time basis.
Boundary: Login does not prove every table permission. ACL denials do not mean empty evidence. No general approval writeback or complete ServiceNow estate discovery is implied.
Scoped telemetry records received and accepted through configured OTLP ingress, with producer attribution and available action correlation.
Boundary: Collector configuration is not a live collector. HTTP 200 or accepted telemetry does not prove evaluation, action application, prevention, or a target-side business change.
Authorized agent-guided preflight, source discovery, evaluation initiation, report navigation and scoped reads when the registered HAIEC tools allow them.
Boundary: MCP is a transport and interaction interface, not proof an external tool ran, that the target was protected, or that an unavailable source exists. Tool and credential scopes matter.
Approved capability scope, named accountable reviewer, policy version, expiry and action-bound authorization where qualified and connected.
Boundary: Policy intent is separate from effective IAM, source capability and actual interception; a human acknowledgment is not necessarily a deployment approval.
A target-owned action or transaction witness may establish applied, rejected or completed effects for an exact resource and time.
Boundary: Not a universal, ready-made HAIEC integration. Requires authorized customer-native APIs/logs and identity join. A guard decision or dispatch acknowledgment alone does not prove target effect.
Authorized provider project, model, usage and cost inventory. This is administrative context rather than an agent action request.
Boundary: Administrative inventory does not prove the agent action path, MCP caller identity, protected resource authorization or observed external business effect.
Choose any source card to inspect its setup, qualification and missing-proof boundary.
Click each plane to see the kinds of sources that may help answer it. None is a universal prerequisite for finding a source-backed action path.
Delegated Action Integrity asks whether a material choice was entrusted to the correct decision-maker. It cuts across relevant authority facts and is not a sixth evidence plane.
Source setup, authorized access, observed records, qualified evidence and persisted assurance are different stages. Select a stage to learn precisely what it does not prove.
Bind tenant, AI System, agent or principal, tool/action and exact resource. Similar names are not a trusted match.
Carry source commit, deployment build, policy version, producer time and evaluation snapshot separately.
Match request/action correlation IDs to a customer-owned target outcome before asserting a business effect.
A purchasing assistant could update supplier payment details.
Evidence question: What can this agent reach, who allowed the choice, and did the ledger change?
A vendor support agent is proposed for access to customer CRM data.
Evidence question: What may this vendor agent see or write, and what source is accessible?
A network operations agent is authorized to recommend or request service changes.
Evidence question: Could the agent invoke the controller, and was the specific change applied?
An AI planning agent can trigger purchase orders or work-order changes.
Evidence question: Was the operation permitted and did ERP/MES apply it to the right record?
Choose a consequential AI scenario, inspect its source requirements and select which sources could realistically be supplied. The guide will show remaining proof gaps without inventing an evaluation.
Select potentially available sources below. Availability is your fictional workshop input, not a test of a real tenant.
Twelve source-specific decisions test whether you can separate a tool, an observation, a qualified fact and a verified consequence.
Bind identity, consent, freshness and evidence quality.
Source capability can be reported immediately, within its own scope.
Cloud policy layers and runtime principals can alter the answer.
Correlate actor, time, resource and independent outcome.
Read the exact frozen report and admitted evidence frontier.
Turn the exercise into a practical, owner-ready plan. Enter only information you are comfortable keeping in your browser. The download stays local.
Exports are locally generated educational worksheets. They are not HAIEC Assurance Runs, U5 dispositions, U6 receipts or actual tenant integration records.
No. HAIEC has separate authorized onboarding paths for connected source, local source, runtime-only inputs and declared artifacts. Each supports different claims, and unverified declarations remain declarations.
Not in the current general case. The AWS observation is partial because resource, session, endpoint and other authorization layers may be missing. Keep effective-grant claims bounded until those layers are qualified.
No. A trace, metric or log can provide observation evidence. An external business target effect needs independently qualified target-side or equivalent evidence.
They may provide context or eligible records when authorized, but the specific approved decision, scope, person and version still need verification. A ticket acknowledgment is not automatically authorization.
MCP is an assistant integration and action interface. A qualified tool response may lead to source evidence, but tool discovery or an agent-generated explanation does not establish a customer-side permission, action or target effect.
Yes, when the source analysis and its scope qualify, code-capable action paths can be meaningful initial findings. Runtime behavior, deployed permission and target outcome must remain unestablished if not independently supported.
No. Connection, data intake, assurance result, human deployment decision and native enforcement are different actions and evidence classes.
Check the current authenticated HAIEC integration interface and scoped preflight. Availability depends on your tenant, credentials, configured sources, permissions, and the tool registry. This offline guide makes no live checks.
Explore an actual source-backed report, connect HAIEC through authorized MCP tools, or scope one system and one business decision for enterprise assurance.
