Objective
Objective: Answer staff questions with verifiable internal evidence Scope: Pilot with one team and approved sample data. Explicit exclusions: No autonomous policy decisions; no unreviewed external actions.
Trigger, inputs, outputs & ordered steps
Trigger: Employee submits a question Source data / evidence links: Approved handbook and access-controlled project notes Desired output: Answer with source references or a clear refusal People involved: Requester, domain reviewer, process owner 1. [trigger] Employee submits a question (Trigger) Input: Employee submits a question → Desired output: Identified request Responsible role: Process owner; Intended system: Inbox / scheduler Success → [source] 2. [source] Retrieve only sources this employee may access (Source retrieval) Input: Approved handbook and access-controlled project notes → Desired output: Accessible sources with references Responsible role: Process owner; Intended system: SharePoint Success → [task]; Failure → [fallback] 3. [task] Draft an answer with citations; abstain if evidence is insufficient (Transformation / AI task) Input: Checked sources → Desired output: Answer with source references or a clear refusal Responsible role: Process owner; Intended system: Proposed AI component Success → [validate]; Insufficient evidence → [fallback] 4. [validate] Validate evidence, quality and output format (Validation) Input: Answer with source references or a clear refusal → Desired output: Validated draft Responsible role: Process owner; Intended system: Rules validator Success → [review]; Invalid data → [fallback] 5. [review] Approve draft and external action (Human review) Input: Validated draft → Desired output: Approval or rejection Responsible role: Domain reviewer; Intended system: Review queue Approved → [action]; Rejected → [fallback] 6. [action] Deliver approved output (External action) Input: Approved draft → Desired output: Answer with source references or a clear refusal Responsible role: Process owner; Intended system: Teams Success → [done]; Failure → [fallback] 7. [fallback] Take no action; request evidence or escalate (Fallback) Input: Failure or rejection reason → Desired output: Manual resolution; no automatic retry cycle Responsible role: Domain reviewer; Intended system: Review queue Success → [done] 8. [done] Record outcome or exception (Completion) Input: Outcome status → Desired output: Auditable completion Responsible role: Process owner; Intended system: Audit log
Integrations & permissions
Start with a fictional example. Every step and branch is editable. Software names describe intended integrations; nothing is connected or executed. Intended software: SharePoint → Teams [trigger] Intended system: Inbox / scheduler; Proposed permissions: Minimum read access; role-based source rights [source] Intended system: SharePoint; Proposed permissions: Minimum read access; role-based source rights [task] Intended system: Proposed AI component; Proposed permissions: Minimum read access; role-based source rights [validate] Intended system: Rules validator; Proposed permissions: Minimum read access; role-based source rights [review] Intended system: Review queue; Proposed permissions: Minimum read access; role-based source rights [action] Intended system: Teams; Proposed permissions: Approved recipients / records only; minimum write access [fallback] Intended system: Review queue; Proposed permissions: Minimum read access; role-based source rights [done] Intended system: Audit log; Proposed permissions: Minimum read access; role-based source rights Retrieval must enforce the requesting person's access rights before answering; inaccessible sources must not enter prompts or answers.
Approval & exceptions
A person must approve each external send or record change. Rejection goes to the exception path. Authorization conditions / prerequisites: Unresolved: define allowed recipients, records, access roles and approval authority. [review] Approve draft and external action — Responsible role: Domain reviewer; Failure behavior: Stop, log context and hand over to the domain reviewer. [action] Deliver approved output — Responsible role: Process owner; Failure behavior: Stop, log context and hand over to the domain reviewer. [fallback] Take no action; request evidence or escalate — Responsible role: Domain reviewer; Failure behavior: Stop, log context and hand over to the domain reviewer.
Acceptance criteria & test cases
Success criteria: Illustrative target: 20 test cases; zero unauthorized accesses; all rejected actions blocked. Agree the quality threshold with the team. [trigger] Employee submits a question: Test each listed condition with matching input and verify its target step. Verify the documented output and owner handoff; force an error wherever a failure path is listed. Success → [source] [source] Retrieve only sources this employee may access: Deny an inaccessible source and return no restricted content; cite accessible evidence, or take the insufficient-evidence branch. Success → [task]; Failure → [fallback] [task] Draft an answer with citations; abstain if evidence is insufficient: Test each listed condition with matching input and verify its target step. Verify the documented output and owner handoff; force an error wherever a failure path is listed. Success → [validate]; Insufficient evidence → [fallback] [validate] Validate evidence, quality and output format: Test each listed condition with matching input and verify its target step. Verify the documented output and owner handoff; force an error wherever a failure path is listed. Success → [review]; Invalid data → [fallback] [review] Approve draft and external action: Test each listed condition with matching input and verify its target step. Verify the documented output and owner handoff; force an error wherever a failure path is listed. Approved → [action]; Rejected → [fallback] [action] Deliver approved output: Reject authorization and verify zero external writes; then approve one allowed action and verify exactly one write. Success → [done]; Failure → [fallback] [fallback] Take no action; request evidence or escalate: Test each listed condition with matching input and verify its target step. Verify the documented output and owner handoff; force an error wherever a failure path is listed. Success → [done] [done] Record outcome or exception: Confirm outcome status and auditable log.
Monitoring, owner & rollout
Accountable owner: Unresolved: name the process owner Monitoring: Log failures, rejected approvals and source coverage per run; review daily during the pilot. [ ] Name owner and backup [ ] Confirm data rights, permissions and retention [ ] Verify interfaces and minimum access in a test environment [ ] Test success, rejection and failure paths [ ] Prevent duplicate processing and rehearse shutdown [ ] Start a limited pilot with daily review [ ] Record acceptance and authorize rollout
Unresolved assumptions
Unresolved assumptions: Unresolved: data rights, retention, quality threshold, budget and available interfaces. Authorization conditions / prerequisites: Unresolved: define allowed recipients, records, access roles and approval authority. Illustrative template. Verify source rights, authorization, test targets and ownership before rollout. Methodology v1 · Schema v1. This deterministic planning tool uses authored templates and an acyclic directed graph (up to 20 steps). Arrows define execution order; list order breaks ties. No AI model, performance estimate or live integration is involved. Examples and acceptance targets are illustrative and must be agreed with your team. A connected graph does not prove production readiness.

