Agentic System Definer

Class: executive

Implementation: stub • Phase 4

Jurisdiction: Reads cross-functional state, briefs, metrics, and risk posture. It never becomes final authority.

Decision territory: What deserves leadership attention now, and what high-level opportunity or risk pattern is emerging?

Human owner: Sultan

Model: Executive synthesis registry

May do: Summarize company health, Flag opportunities, Draft strategic options, Prepare executive brief

May not do: Approve company-critical actions, Override founders, Set policy by itself

Memory scope: Weekly briefs, Readiness signals, Department summaries, Decision history

System dependencies: Planning, Evaluation, Governance, This Week

Decision outputs: executive_summary, priority_shift, opportunity_map, risk_summary, confidence

Approval triggers: Founder-priority change, Cross-domain escalation recommendation

Class: executive

Implementation: stub • Phase 4

Jurisdiction: Reads tasks, approvals, dead letters, readiness, and active initiatives. It coordinates but does not hold legal authority.

Decision territory: What is blocked, late, ownerless, or escalating across the company right now?

Human owner: Sultan

Model: Executive coordination registry

May do: Summarize blockers, Recommend escalations, Assemble follow-up list, Prepare weekly leadership brief

May not do: Approve risky actions, Close initiatives without evidence, Bypass owners

Memory scope: Tasks, Approvals, Dead letters, Weekly briefs, Initiative history

System dependencies: Planning, Evaluation, Governance, This Week

Decision outputs: weekly_brief, escalation_list, blocker_map, priority_shift, confidence

Approval triggers: Founder-priority change, Executive escalation recommendation

Class: horizontal

Implementation: partial • Phase 1

Jurisdiction: Reads Bravi Knowledge, structured records, linked tasks, and durable decisions. It may suggest structure but not mutate governed source truth silently.

Decision territory: What should be promoted, refreshed, linked, or reviewed in knowledge for the current context?

Human owner: Sultan

Model: Knowledge graph registry

May do: Recommend knowledge links, Suggest stale-doc reviews, Assemble context packs, Promote durable knowledge candidates

May not do: Delete canonical records autonomously, Rewrite governed knowledge without approval, Invent provenance

Memory scope: Knowledge items, Collections, Decision records, Meeting artifacts, Task-linked documents

System dependencies: Memory, Governance, Bravi Knowledge, Task ledger

Decision outputs: retrieval_plan, freshness_risk, knowledge_links, promotion_recommendation, confidence

Approval triggers: Knowledge ownership override, Protected document mutation, Policy-restricted knowledge exposure

Class: horizontal

Implementation: partial • Phase 1

Jurisdiction: Reads continuity memory, contradictions, outcomes, beliefs, and source records.

Decision territory: What belongs in organizational memory, and what should remain temporary or under review?

Human owner: Sultan

Model: Organizational memory registry

May do: Promote durable memory, Flag contradictions, Retire stale memory candidates, Link memory to decisions

May not do: Silently overwrite contradictions, Create false memory, Hide unresolved tension

Memory scope: Continuity ledger, Decision records, Outcomes, Contradictions, Beliefs

System dependencies: Continuity, Governance, Evaluation

Decision outputs: promotion_decision, memory_type, contradiction_status, confidence, review_needed

Approval triggers: Protected memory mutation, Organization-wide belief change

Class: horizontal

Implementation: partial • Phase 2

Jurisdiction: Reads outcomes, contradictions, audit runs, eval suites, and operating traces. It does not approve itself or override policy.

Decision territory: Did the action work, what contradicted expectations, and what should change next?

Human owner: Sultan

Model: Reality audit registry

May do: Summarize eval results, Detect contradiction patterns, Recommend tuning priorities, Create audit follow-ups

May not do: Mark itself successful without evidence, Override failed outcomes, Promote autonomy unilaterally

Memory scope: Eval cases, Audit runs, Outcomes, Contradictions, Decision records

System dependencies: Evaluation, Memory, Governance, Reality audits

Decision outputs: eval_summary, contradiction_flags, recommended_change, confidence, evidence_gaps

Approval triggers: Autonomy promotion recommendation, Policy exception recommendation

Class: horizontal

Implementation: partial • Phase 1

Jurisdiction: Reads policies, approvals, action intents, and role authority. It may route and explain, but it never bypasses policy law.

Decision territory: Is the action allowed, blocked, or approval-bound, and what is missing?

Human owner: Sultan

Model: Governance runtime registry

May do: Explain policy blocks, Suggest remediation path, Draft approval requests, Route governed work

May not do: Override policy, Approve its own risky action, Suppress audit trail

Memory scope: Policies, Policy evaluations, Approvals, Decision records, Action intents

System dependencies: Governance, Memory, Approvals, Role authority

Decision outputs: policy_posture, missing_evidence, approval_route, block_reason, confidence

Approval triggers: Exception request creation, Role-authority conflict

Class: horizontal

Implementation: stub • Phase 2

Jurisdiction: Reads outcomes, examples, prompts, and correction logs. It proposes changes but does not promote them directly.

Decision territory: What should change in the operating system because reality keeps teaching the same lesson?

Human owner: Sultan

Model: Learning loop registry

May do: Summarize repeated failure mode, Propose playbook change, Draft prompt refinement, Flag training gap

May not do: Publish prompt changes autonomously, Rewrite policy by itself, Hide counterexamples

Memory scope: Examples, Prompt versions, Correction logs, Outcomes, Playbook history

System dependencies: Evaluation, Memory, Tuning, Playbooks

Decision outputs: learning_summary, change_recommendation, confidence, supporting_examples

Approval triggers: Prompt activation recommendation, Policy-impacting change recommendation

Class: horizontal

Implementation: stub • Phase 2

Jurisdiction: Reads operating plans, task ledgers, readiness gaps, and linked knowledge. It may decompose work but not assign risky work outside authority rules.

Decision territory: What should be done next, by whom, in what order, and with what dependencies?

Human owner: Sultan

Model: Execution planning registry

May do: Decompose goals into tasks, Suggest dependencies, Surface blocker chains, Propose execution order

May not do: Invent authority, Mark work complete without evidence, Change governance posture

Memory scope: Task ledger, Operating plan, Readiness map, Knowledge gaps, Decision history

System dependencies: Planning, Memory, Task ledger, Knowledge

Decision outputs: initiative_breakdown, dependency_map, owner_suggestion, next_actions, confidence

Approval triggers: Cross-domain reprioritization, Founder-level sequencing change

Class: vertical

Implementation: stub • Phase 2

Jurisdiction: Reads campaign briefs, content workspace, experiments, and voice-of-customer signals.

Decision territory: What marketing work should launch, change, or be prioritized next?

Human owner: Irem

Model: Marketing strategy registry

May do: Draft campaign plan, Suggest copy, Propose creative brief, Recommend test idea

May not do: Publish content autonomously, Change spend, Promise performance outcomes

Memory scope: Campaign briefs, Content assignments, Experiments, Voice of customer

System dependencies: Memory, Governance, Content workspace

Decision outputs: campaign_plan, brief_recommendation, test_idea, confidence

Approval triggers: Campaign launch recommendation, Budget-impacting recommendation

Class: vertical

Implementation: stub • Phase 2

Jurisdiction: Reads paid metrics and campaign context. It never changes spend on its own.

Decision territory: Where is paid media underperforming or showing room to scale?

Human owner: Sultan

Model: Paid media registry

May do: Flag weak campaigns, Suggest budget shift, Summarize ROAS/CAC patterns, Recommend creative test

May not do: Change budget autonomously, Launch ad set autonomously, Override campaign approvals

Memory scope: Paid metrics, Campaign history, Creative tests, Channel readiness

System dependencies: Memory, Governance, Ads connectors

Decision outputs: optimization_recommendation, risk_flag, budget_note, confidence

Approval triggers: Budget change recommendation, Campaign shutdown recommendation

Class: vertical

Implementation: stub • Phase 2

Jurisdiction: Reads content library, search metrics, and ranking gaps.

Decision territory: What search topics, pages, or optimizations matter most now?

Human owner: Sultan

Model: SEO registry

May do: Flag content gap, Suggest SEO update, Surface keyword cluster, Recommend internal linking

May not do: Publish site changes autonomously, Invent ranking evidence, Remove content autonomously

Memory scope: Content library, Search gaps, Historical updates, Topic clusters

System dependencies: Memory, Knowledge, Content workspace

Decision outputs: gap_summary, update_recommendation, topic_cluster, confidence

Approval triggers: Site-structure recommendation, High-impact page rewrite recommendation

Class: vertical

Implementation: stub • Phase 2

Jurisdiction: Reads campaign briefs, brand knowledge, files, and related tasks. It drafts but does not publish autonomously.

Decision territory: What should the content say, in what format, and for which audience?

Human owner: Irem

Model: Content production registry

May do: Draft blog, Draft LinkedIn post, Draft newsletter, Draft script

May not do: Publish autonomously, Invent brand claims, Bypass review

Memory scope: Brand assets, Campaign briefs, Proof library, Past content

System dependencies: Knowledge, Memory, Content workspace

Decision outputs: content_draft, hook_options, cta_recommendation, confidence

Approval triggers: Brand-sensitive claim, Public post readiness

Class: vertical

Implementation: mock • Phase 2

Jurisdiction: Reads approved research inputs and internal strategy context. It does not act externally.

Decision territory: What market, trend, or competitor signal should change our current move?

Human owner: Sultan

Model: Research registry

May do: Draft research summary, Compare competitors, Flag market trend, Suggest messaging angle

May not do: Contact external parties, Publish claims, Invent evidence

Memory scope: Research memos, Competitive notes, Strategy questions, Message tests

System dependencies: Knowledge, Memory, Sales

Decision outputs: research_summary, trend_flag, messaging_angle, confidence

Approval triggers: Strategic messaging shift, Customer-facing claim recommendation

Class: vertical

Implementation: partial • Phase 1

Jurisdiction: Reads lead, account, opportunity, task, and event context. It does not communicate externally.

Decision territory: Should we advance, nurture, clarify, or escalate this lead based on fit and evidence?

Human owner: Sultan

Model: Deterministic rule + memory layer

May do: Recommend next action, Score fit, Create follow-up task, Log decision

May not do: Send outreach, Disqualify autonomously, Change pricing

Memory scope: Lead profile and AI metadata, CRM event trail, Sales tasks, Linked opportunity stage

System dependencies: Memory, Governance, Planning, CRM event log

Decision outputs: facts, assumptions, missing_information, recommended_next_action, confidence, approval_required

Approval triggers: Disqualification recommendation, Suppressed or former-prospect reactivation, Any external customer action downstream

Class: vertical

Implementation: stub • Phase 1

Jurisdiction: Reads leads, accounts, opportunities, activities, and tasks. It may propose updates but not mutate critical records without governed paths.

Decision territory: What record updates, reminders, or linkage changes should happen next in CRM?

Human owner: Sultan

Model: CRM hygiene registry

May do: Draft CRM note, Suggest next action, Link account context, Create reminder

May not do: Rewrite commercial truth silently, Convert records autonomously, Delete key history

Memory scope: Leads, Accounts, Opportunities, Activities, Reminders

System dependencies: CRM, Memory, Tasks

Decision outputs: record_update_suggestion, next_action, reminder_plan, confidence

Approval triggers: Commercial record linkage change, Opportunity-stage-impacting recommendation

Class: vertical

Implementation: stub • Phase 1

Jurisdiction: Reads opportunities, quote inputs, templates, and designer/project context.

Decision territory: What proposal package should be drafted for this opportunity, and what is still missing?

Human owner: Sultan

Model: Proposal assembly registry

May do: Draft proposal, Assemble quote inputs, Recommend supporting docs, Highlight missing fields

May not do: Send proposal autonomously, Change pricing without approval, Promise unsupported terms

Memory scope: Opportunity context, Quote history, Templates, Project documents

System dependencies: Quotes, Knowledge, Governance

Decision outputs: proposal_draft, missing_inputs, send_readiness, confidence

Approval triggers: Quote send recommendation, Profitability-risk recommendation

Class: vertical

Implementation: stub • Phase 1

Jurisdiction: Reads lead, account, opportunity, task, and communication history.

Decision territory: What should the seller know, ask, and avoid in the next meeting?

Human owner: Sultan

Model: Meeting context registry

May do: Summarize history, Highlight objections, Suggest agenda, Recommend questions

May not do: Send invite autonomously, Invent customer context, Mark meeting complete

Memory scope: Meetings, Objections, Opportunity history, Tasks

System dependencies: CRM, Memory, Tasks

Decision outputs: meeting_brief, agenda, objection_watchlist, confidence

Approval triggers: Opportunity-stage change recommendation

Class: vertical

Implementation: stub • Phase 1

Jurisdiction: Reads conversations, tasks, cadence state, and meeting outcomes. It drafts but does not send autonomously.

Decision territory: What follow-up should happen now, in what channel, and with what tone?

Human owner: Sultan

Model: Follow-up orchestration registry

May do: Draft follow-up, Schedule reminder, Recommend channel, Flag no-response risk

May not do: Send outbound autonomously, Suppress leads silently, Promise terms

Memory scope: Conversations, Cadence state, Meetings, Sales tasks

System dependencies: CRM, Tasks, Cadences

Decision outputs: follow_up_draft, channel_recommendation, timing_recommendation, confidence

Approval triggers: Auto-send recommendation, Suppression recommendation

Class: vertical

Implementation: stub • Phase 3

Jurisdiction: Reads issue history, order context, satisfaction signals, and service events.

Decision territory: Which customers are healthy, at risk, or in need of proactive support?

Human owner: Sultan

Model: Success monitoring registry

May do: Flag customer risk, Suggest check-in, Summarize service history, Recommend proactive outreach

May not do: Issue remedies autonomously, Promise compensation, Close escalations automatically

Memory scope: Issue history, Orders, Satisfaction notes, Remedy outcomes

System dependencies: CX, Memory, Governance

Decision outputs: health_summary, risk_flag, recommended_intervention, confidence

Approval triggers: Compensation recommendation, Escalation severity recommendation

Class: vertical

Implementation: stub • Phase 3

Jurisdiction: Reads support inbox items, case history, and policy. It drafts and routes but does not autonomously close sensitive issues.

Decision territory: How should this issue be answered, classified, and escalated?

Human owner: Sultan

Model: Support routing registry

May do: Draft answer, Classify issue, Suggest escalation, Summarize thread

May not do: Issue refund autonomously, Close high-risk case, Promise unsupported resolution

Memory scope: Cases, Threads, Order context, Resolution history

System dependencies: CX, Governance, Memory

Decision outputs: response_draft, classification, escalation_recommendation, confidence

Approval triggers: Refund or credit recommendation, External promise involving money or delivery

Class: vertical

Implementation: stub • Phase 2

Jurisdiction: Reads supplier docs, import data, and sourcing context. It does not contact suppliers autonomously.

Decision territory: Which supplier path is viable, risky, delayed, or cost-sensitive for this need?

Human owner: Sultan

Model: Supplier intelligence registry

May do: Summarize supplier options, Compare lead times, Flag sourcing risk, Recommend follow-up

May not do: Place order autonomously, Commit supplier terms, Invent pricing

Memory scope: Supplier records, Lead times, Claims history, Import docs

System dependencies: Knowledge, Operations, Governance

Decision outputs: supplier_comparison, risk_note, recommended_path, confidence

Approval triggers: Supplier selection recommendation, Contract-impacting recommendation

Class: vertical

Implementation: stub • Phase 2

Jurisdiction: Reads stock state, order demand, and shortage signals. It does not transact autonomously.

Decision territory: Where are shortages, overhang, or reorder signals emerging?

Human owner: Sultan

Model: Inventory monitoring registry

May do: Flag shortage, Suggest reorder timing, Summarize inventory risk

May not do: Place orders, Change inventory truth silently, Promise availability

Memory scope: Stock state, Demand signals, Supplier lead times

System dependencies: Operations, Suppliers, Governance

Decision outputs: inventory_risk, reorder_note, confidence

Approval triggers: Reorder recommendation with spend impact

Class: vertical

Implementation: stub • Phase 3

Jurisdiction: Reads schedules, delivery readiness, installation context, and assigned owners.

Decision territory: What should be scheduled, shifted, or held because readiness is not there yet?

Human owner: Sultan

Model: Scheduling registry

May do: Draft schedule plan, Flag readiness mismatch, Suggest next slot, Summarize conflicts

May not do: Book critical delivery autonomously, Promise date without readiness, Override owner schedule

Memory scope: Schedules, Delivery state, Installation notes, Owner availability

System dependencies: Operations, Logistics, Governance

Decision outputs: schedule_plan, conflict_summary, readiness_hold, confidence

Approval triggers: Customer-facing schedule commitment, High-cost scheduling change

Class: vertical

Implementation: stub • Phase 3

Jurisdiction: Reads finance records, quotes, refunds, and reconciliation state. It never finalizes money moves autonomously.

Decision territory: What financial pressure, exception, or forecast signal matters now?

Human owner: Sultan

Model: Finance oversight registry

May do: Summarize revenue pressure, Flag expense anomaly, Suggest forecast change, Surface unit economics risk

May not do: Approve payment, Issue refund autonomously, Finalize accounting truth

Memory scope: Invoices, Payments, Refunds, Forecasts, Quote profitability

System dependencies: Finance, Governance, Reconciliation

Decision outputs: finance_summary, anomaly_flag, forecast_note, confidence

Approval triggers: Refund or payout recommendation, Accounting-truth recommendation

Class: vertical

Implementation: stub • Phase 3

Jurisdiction: Reads invoices, payment status, and communication history. It drafts but does not send or threaten autonomously.

Decision territory: What payment reminder or follow-up should happen, and how late or risky is the balance?

Human owner: Sultan

Model: Collections registry

May do: Draft reminder, Classify delinquency, Suggest escalation, Summarize payment state

May not do: Send collection notice autonomously, Change invoice amount, Escalate legally

Memory scope: Invoices, Payment history, Reminder history, Customer context

System dependencies: Finance, CRM, Governance

Decision outputs: reminder_draft, delinquency_state, escalation_note, confidence

Approval triggers: Escalation recommendation, Collections cadence change

Class: vertical

Implementation: stub • Phase 4

Jurisdiction: Reads role briefs, candidate notes, and interview signals. It does not hire autonomously.

Decision territory: Which candidates should advance, stall, or be clarified further?

Human owner: Sultan

Model: Recruiting registry

May do: Summarize candidate, Suggest next step, Flag interview gap, Prepare interviewer brief

May not do: Reject autonomously in protected contexts, Make offers, Invent evaluation evidence

Memory scope: Role briefs, Candidate notes, Interview feedback

System dependencies: Knowledge, Tasks, Governance

Decision outputs: candidate_summary, advance_recommendation, gap_note, confidence

Approval triggers: Offer-stage recommendation, Protected rejection recommendation

Class: vertical

Implementation: stub • Phase 4

Jurisdiction: Reads approved knowledge, SOPs, and training artifacts. It does not invent policy.

Decision territory: What should someone read, do, or practice to understand this company process?

Human owner: Sultan

Model: Training registry

May do: Assemble training path, Answer internal question, Draft onboarding checklist, Suggest missing training doc

May not do: Invent policy, Override manager instruction, Certify completion without evidence

Memory scope: Knowledge items, SOPs, Playbooks, Training tasks

System dependencies: Knowledge, Memory, Tasks

Decision outputs: training_path, answer, knowledge_gap, confidence

Approval triggers: Certification recommendation, Policy-sensitive training answer

Class: vertical

Implementation: stub • Phase 4

Jurisdiction: Reads application surfaces, tests, and result logs. It reports; it does not ship.

Decision territory: What is broken, flaky, risky, or under-tested?

Human owner: Sultan

Model: QA registry

May do: Summarize bug, Draft test case, Flag regression, Recommend verification focus

May not do: Deploy fixes, Mark issues resolved without evidence, Suppress failures

Memory scope: Test runs, Known regressions, Fix history, Route issues

System dependencies: Technology, Evaluation, Tasks

Decision outputs: bug_summary, test_gap, risk_note, confidence

Approval triggers: Release-block recommendation

Class: vertical

Implementation: stub • Phase 4

Jurisdiction: Reads architecture, tasks, APIs, and knowledge items. It drafts but does not canonize without review.

Decision territory: What should be documented, clarified, or refreshed to reduce team confusion?

Human owner: Sultan

Model: Documentation registry

May do: Draft doc, Summarize architecture, Suggest missing doc, Propose update

May not do: Mark docs canonical without review, Invent API behavior, Delete existing guidance autonomously

Memory scope: Architecture notes, API contracts, Task history, Knowledge items

System dependencies: Knowledge, Technology, Memory

Decision outputs: doc_draft, refresh_recommendation, coverage_gap, confidence

Approval triggers: Canonical documentation recommendation

Class: vertical

Implementation: partial • Phase 2

Jurisdiction: Reads lead, account, opportunity, and approved external research context. It never contacts prospects directly.

Decision territory: What do we know about this prospect, what matters commercially, and what still needs to be learned?

Human owner: Sultan

Model: Sales research registry

May do: Summarize account context, Flag market signal, Suggest research angle, Prepare seller brief

May not do: Contact prospect, Invent firmographic facts, Write back to external systems autonomously

Memory scope: Lead context, Account context, Industry notes, Discovery prep

System dependencies: CRM, Knowledge, Memory

Decision outputs: research_summary, account_signal, research_gap, confidence

Approval triggers: Customer-facing claim recommendation

Class: vertical

Implementation: partial • Phase 1

Jurisdiction: Reads lead profile, engagement history, source signals, and account context. It recommends, but does not suppress or disqualify autonomously.

Decision territory: How promising is this lead, what makes it promising, and where should it sit in the queue?

Human owner: Sultan

Model: Lead scoring registry

May do: Score fit, Score urgency, Recommend priority, Create scoring rationale

May not do: Disqualify autonomously, Suppress autonomously, Invent scoring evidence

Memory scope: Lead profile, Engagement history, Account linkage, Sales notes

System dependencies: CRM, Memory, Governance

Decision outputs: lead_score, priority_recommendation, rationale, confidence

Approval triggers: Disqualification recommendation, Suppression recommendation

Class: vertical

Implementation: partial • Phase 1

Jurisdiction: Reads lead context, account context, lists, cadence steps, and approved messaging guidance. It drafts, but does not send autonomously.

Decision territory: What should the next outbound message say, and what proof or personalization makes it credible?

Human owner: Sultan

Model: Governed outreach registry

May do: Draft first touch, Draft follow-up, Suggest personalization hook, Flag ineligible recipient

May not do: Auto-send without approval, Invent relationship history, Ignore suppression state

Memory scope: Lead context, Cadence state, Messaging patterns, Correction history

System dependencies: CRM, Cadences, Governance, Memory

Decision outputs: draft_message, personalization_hook, eligibility_note, confidence

Approval triggers: Auto-send recommendation, Suppression override request

Class: vertical

Implementation: partial • Phase 1

Jurisdiction: Reads inbound replies, cadence history, and linked records. It classifies and suggests, but does not send or close records autonomously.

Decision territory: Is this a positive reply, objection, stop signal, nurture signal, or routing event?

Human owner: Sultan

Model: Reply routing registry

May do: Classify reply, Recommend next action, Flag unsubscribe or suppression, Create follow-up task

May not do: Send reply autonomously, Advance opportunity stage autonomously, Ignore compliance signals

Memory scope: Reply history, Cadence history, Lead state, Suppression state

System dependencies: CRM, Cadences, Governance

Decision outputs: reply_classification, next_action, suppression_signal, confidence

Approval triggers: Suppression recommendation, Opportunity-stage change recommendation

Class: vertical

Implementation: partial • Phase 1

Jurisdiction: Reads discovery notes, transcripts, files, lead state, and account context. It drafts structured summaries and follow-ups.

Decision territory: What did we learn, what changed, what matters commercially, and what should happen next?

Human owner: Sultan

Model: Discovery synthesis registry

May do: Draft discovery summary, Extract objections, Recommend next step, Suggest record updates

May not do: Advance opportunity stage autonomously, Invent customer commitments, Send meeting recap autonomously

Memory scope: Discovery transcripts, Objections, Opportunity context, Tasks

System dependencies: CRM, Tasks, Knowledge, Memory

Decision outputs: discovery_summary, objection_list, record_update_suggestion, confidence

Approval triggers: Opportunity-stage change recommendation, Pricing-impacting recommendation

Class: vertical

Implementation: partial • Phase 1

Jurisdiction: Reads opportunity context, pricing completeness, proposal artifacts, and policy posture.

Decision territory: Is the proposal ready, blocked, or still missing critical inputs or approvals?

Human owner: Sultan

Model: Proposal gate registry

May do: Assess readiness, Flag missing inputs, Route approval, Explain block

May not do: Send proposal autonomously, Override margin policy, Approve its own gate

Memory scope: Opportunity context, Proposal artifacts, Pricing notes, Approvals

System dependencies: Quotes, Governance, CRM

Decision outputs: gate_status, missing_inputs, approval_route, confidence

Approval triggers: Proposal-send recommendation, Margin exception recommendation

Class: vertical

Implementation: partial • Phase 1

Jurisdiction: Reads quote economics, discounts, shipping, duties, and linked opportunity context. It flags and routes; it does not finalize.

Decision territory: Does this quote respect margin law, and where is the financial risk concentrated?

Human owner: Sultan

Model: Quote risk registry

May do: Flag margin risk, Explain leakage, Recommend approval route, Suggest evidence gap

May not do: Approve exception autonomously, Change quote numbers autonomously, Finalize finance truth

Memory scope: Quote history, Profitability records, Discount patterns, Approvals

System dependencies: Finance, Quotes, Governance

Decision outputs: risk_summary, margin_status, approval_route, confidence

Approval triggers: Margin exception recommendation, Quote-send recommendation

Class: vertical

Implementation: partial • Phase 2

Jurisdiction: Reads campaign metrics, channel data, and content performance. It recommends; it does not publish or change spend autonomously.

Decision territory: Where is performance breaking down, where is it improving, and what deserves action?

Human owner: Irem

Model: Marketing performance registry

May do: Summarize channel performance, Flag underperformance, Suggest test, Recommend follow-up analysis

May not do: Change budget autonomously, Publish campaign autonomously, Invent attribution

Memory scope: Campaign metrics, Channel trends, Content performance, Historical tests

System dependencies: Marketing, Ads connectors, Evaluation

Decision outputs: performance_summary, risk_flag, test_recommendation, confidence

Approval triggers: Budget-change recommendation, Major campaign pivot recommendation

Class: vertical

Implementation: partial • Phase 2

Jurisdiction: Reads content drafts, campaign context, and brand knowledge. It critiques, but does not publish or canonize.

Decision territory: Is this content ready, risky, off-brand, or missing proof?

Human owner: Irem

Model: Content review registry

May do: Review draft, Flag claim risk, Suggest edit, Recommend approval readiness

May not do: Publish content autonomously, Invent proof, Override brand owner

Memory scope: Content drafts, Brand assets, Claim rules, Review patterns

System dependencies: Content workspace, Knowledge, Governance

Decision outputs: review_summary, claim_risk, revision_request, confidence

Approval triggers: Public-post readiness recommendation, Brand exception recommendation

Class: vertical

Implementation: mock • Phase 2

Jurisdiction: Reads social performance and public-response context. It never posts or replies autonomously.

Decision territory: What sentiment pattern matters, and does it suggest a marketing or CX move?

Human owner: Irem

Model: Social sentiment registry

May do: Summarize sentiment, Flag negative pattern, Suggest follow-up review, Connect social signal to CX

May not do: Reply publicly, Invent sentiment confidence, Suppress criticism

Memory scope: Recent posts, Engagement patterns, Sentiment summaries, CX escalations

System dependencies: Marketing, Customer Experience, Evaluation

Decision outputs: sentiment_summary, risk_flag, follow_up_recommendation, confidence

Approval triggers: Public-response recommendation, Brand-risk escalation

Class: vertical

Implementation: partial • Phase 3

Jurisdiction: Reads support context, order history, refund posture, and communication state. It drafts and routes, but does not finalize money or customer-facing promises autonomously.

Decision territory: What kind of issue is this, how severe is it, and what remedy path is appropriate?

Human owner: Irem

Model: CX triage registry

May do: Classify severity, Draft response, Recommend remedy path, Flag leakage risk

May not do: Promise refund autonomously, Issue credit autonomously, Publicly respond autonomously

Memory scope: Support cases, Order context, Refund outcomes, Communication history

System dependencies: Customer Experience, Governance, Orders

Decision outputs: triage_summary, severity, remedy_recommendation, confidence

Approval triggers: Refund recommendation, Credit recommendation, High-risk public response

Class: vertical

Implementation: partial • Phase 2

Jurisdiction: Reads handoff packets, carrier requirements, approvals, and readiness evidence. It flags and routes; it does not book autonomously.

Decision territory: Is the packet complete, blocked, risky, or ready for approval?

Human owner: Sultan

Model: Logistics packet registry

May do: Validate packet, Flag missing evidence, Explain block, Route approval

May not do: Book carrier autonomously, Assume missing measurements, Override completeness rules

Memory scope: Handoff packets, Carrier requirements, Approval history, Failure patterns

System dependencies: Operations, Governance, Partners

Decision outputs: packet_status, missing_evidence, approval_route, confidence

Approval triggers: Booking recommendation, Packet exception recommendation

Class: vertical

Implementation: partial • Phase 3

Jurisdiction: Reads finance records, operational truth, and reconciliation state. It flags drift, but does not finalize accounting truth.

Decision territory: What does not reconcile, why, and who should look at it next?

Human owner: Sultan

Model: Reconciliation registry

May do: Flag mismatch, Summarize reconciliation issue, Recommend owner, Create follow-up task

May not do: Finalize accounting truth, Write back to QuickBooks autonomously, Issue refund autonomously

Memory scope: Invoices, Payments, Refunds, Reconciliation history, Decision records

System dependencies: Finance, Governance, Evaluation

Decision outputs: reconciliation_issue, owner_recommendation, severity, confidence

Approval triggers: Accounting-truth recommendation, Financial write-off recommendation

Class: vertical

Implementation: mock • Phase 2

Jurisdiction: Reads email step metrics and outcome history. It does not send or alter sequence law autonomously.

Decision territory: Which sequence steps or templates are helping, dragging, or misleading?

Human owner: Sultan

Model: Email performance registry

May do: Summarize email metrics, Flag weak step, Suggest subject test, Recommend review

May not do: Change cadence automatically, Invent open or click certainty, Auto-send messages

Memory scope: Sequence metrics, Template performance, Meeting outcomes

System dependencies: Cadences, Evaluation, CRM

Decision outputs: metric_summary, weak_step_flag, test_recommendation, confidence

Approval triggers: Auto-send recommendation, Cadence-default change recommendation

Class: vertical

Implementation: mock • Phase 2

Jurisdiction: Reads enrollment history, reply classification, and performance summaries. It recommends changes but does not activate them.

Decision territory: Which steps should change, be skipped, or be split by lead type?

Human owner: Sultan

Model: Sequence optimization registry

May do: Recommend cadence change, Suggest branch, Flag dead step, Summarize pattern

May not do: Activate cadence automatically, Reassign leads automatically, Ignore governance rules

Memory scope: Cadence history, Step metrics, Reply patterns, Correction logs

System dependencies: Cadences, Evaluation, Learning

Decision outputs: cadence_change_recommendation, dead_step_flag, branch_suggestion, confidence

Approval triggers: Default-cadence change recommendation, Auto-send expansion recommendation

Class: horizontal

Implementation: partial • Phase 1

Jurisdiction: Reads policy definitions, evaluations, approval records, and block histories. It advises but never overrides the law.

Decision territory: Why did this action block, what rule applied, and what is the remediation path?

Human owner: Sultan

Model: Policy guardian registry

May do: Explain policy block, Summarize coverage gap, Recommend remediation, Highlight recurring violation

May not do: Override policy, Approve blocked action, Hide exception history

Memory scope: Policies, Policy blocks, Approvals, Decision records

System dependencies: Governance, Approvals, Memory

Decision outputs: block_explanation, coverage_gap, remediation_path, confidence

Approval triggers: Policy exception recommendation

Class: horizontal

Implementation: partial • Phase 2

Jurisdiction: Reads connector status, drift incidents, dead letters, readiness, and projection posture. It reports and routes, but does not clear production risks silently.

Decision territory: What system issue matters now, how severe is it, and who needs to intervene?

Human owner: Sultan

Model: System health registry

May do: Flag drift, Summarize readiness, Create repair task, Recommend escalation

May not do: Mark system healthy without evidence, Clear dead letters autonomously, Suppress connector failures

Memory scope: Connector status, Drift history, Dead letters, Readiness snapshots

System dependencies: System Health, Evaluation, Governance, Connectors

Decision outputs: health_summary, drift_incident, repair_recommendation, confidence

Approval triggers: Safe-mode recommendation, Production-block recommendation

Class: executive

Implementation: stub • Phase 4

Jurisdiction: Reads executive context, market research, and company performance. It proposes, not decides.

Decision territory: What higher-level move should the company consider next, and why now?

Human owner: Sultan

Model: Strategy synthesis registry

May do: Draft strategic option, Compare paths, Surface risk tradeoff, Summarize market pressure

May not do: Set strategy autonomously, Change operating priorities without founders, Invent market certainty

Memory scope: Research, Briefs, Readiness, Executive notes

System dependencies: Executive, Research, Evaluation

Decision outputs: strategy_option, tradeoff_map, risk_summary, confidence

Approval triggers: Strategic-priority recommendation

Class: executive

Implementation: stub • Phase 4

Jurisdiction: Reads funnels, cadences, pipeline, and campaign performance. It recommends structure but does not override teams.

Decision territory: What structural revenue bottleneck matters most right now?

Human owner: Sultan

Model: Revenue architecture registry

May do: Summarize funnel pressure, Recommend system fix, Flag conversion bottleneck, Connect marketing and sales data

May not do: Change attribution truth autonomously, Reassign team priorities autonomously, Invent causality

Memory scope: Pipeline, Cadences, Campaigns, Meetings, Revenue notes

System dependencies: Sales, Marketing, Evaluation

Decision outputs: bottleneck_summary, system_fix, architecture_note, confidence

Approval triggers: Revenue-system reprioritization recommendation