Workbench Doctrine
Status: Canonical architecture doctrine
Date: July 12, 2026
Authority: Nirmal (Founder)
Supersedes: All prior Workbench product doctrine
Canonical Reference: CANONICAL_PLATFORM_ARCHITECTURE.md
The Workbench is the primary operating surface of IntegrateWise.
People do not work inside applications.
AI does not work inside applications.
Governance does not happen inside applications.
Everything operates through governed Work that projects continuity from the Adaptive Spine.
Applications become capabilities rather than destinations.
Work is never the system of record.
Work never owns canonical truth.
Work projects governed context into the appropriate experience.
Core Doctrine
Truth you own. AI you rent. Approval in between.
The organization owns its operational truth in the Adaptive Spine. The Twin provides rented intelligence — reasoning, drafting, retrieval, and proposals — but never gets a direct Act button. Every consequential execution is gated by human approval and by Governance.
L1 Target
The human product is a single L1 Workbench.
Intelligence is a silent partner inside L1. Governance appears when policy requires it. L2/L3 are panel states inside L1, not separate product boundaries.
Product grammar
CONTEXT LENS
→ ENTITY COLLECTION
→ ENTITY 360
→ ACTIVITY
→ CONTEXTUAL INTELLIGENCE
→ ACTION
→ GOVERNANCE WHEN REQUIRED
→ EXECUTION STATE
→ CONTINUITY
The Four-Button OODA Grammar
The User Workbench interaction model is exactly four canonical buttons, mapped directly to the OODA loop. These are the universal interaction grammar across every Workbench — not domain-specific action button chaos.
| OODA | Canonical Action | Button | System Meaning |
|---|---|---|---|
| Observe | Capture truth | Store in Spine | Persist context, evidence, decision, note, or artifact into canonical Digital Memory. |
| Orient | Understand context | Ask Your Twin | Query the Twin using current workspace + Continuity Bridge + Spine context. |
| Decide | Delegate intent | Assign Your Twin | Give the Twin an objective to plan, reason, draft, or prepare an action. |
| Act | Govern execution | Approve Twin's Action | Human approval authorizes Hermes / capability execution. Twin never gets a direct Act button. |
The Critical Authority Boundary
STORE
User ────────────────► Spine
ASK
Spine + Context ─────► Twin ─────► Insight
ASSIGN
User ────────────────► Twin ─────► Proposal
APPROVE
Human ───────────────► Governance ─────► Hermes ─────► Capability
The fourth button is deliberately Approve Twin's Action — not Run Twin, not Execute, not Automate. That is what preserves the doctrine:
Truth you own. AI you rent. Approval in between.
Store → Ask → Assign → Approve is the OODA loop made executable on every Workbench.
Workbench UI
┌──────────────────────────────────────────────────────────────────┐
│ ACCOUNT SUCCESS Acme Corp │
│ Context: Renewal · Q3 · At Risk │
├──────────────────────────────────────────────────────────────────┤
│ │
│ WORKBENCH │
│ │
│ [ Store in Spine ] [ Ask Your Twin ] │
│ OBSERVE ORIENT │
│ │
│ [ Assign Your Twin ] [ Approve Twin's Action ] │
│ DECIDE ACT │
│ │
├──────────────────────────────────────────────────────────────────┤
│ OODA ● Observe ─── ○ Orient ─── ○ Decide ─── ○ Act │
└──────────────────────────────────────────────────────────────────┘
Projection Principle
Every projection is a view over the same governed platform.
No projection owns data.
No projection owns truth.
Every projection reads from the same governed platform.
Different projections expose different perspectives over the same organizational continuity.
Context Lens Doctrine
SAME ENTITY
+
DIFFERENT CONTEXT LENS
=
DIFFERENT OPERATIONAL EMPHASIS
Context Lenses include Account Success, Revenue Operations, Sales Operations, Finance, Marketing, HR/People, Product and Engineering, Business Operations, IT Administration, Services, and Procurement. These are role, department, industry, and capability clusters, not separate product boundaries unless separately authorized.
Twin Inside L1
Twin is a silent partner inside L1, not a separate product interface.
The Twin works alongside the user with durable memory, session continuity, and governed context. Users Observe by storing truth in the Spine, Orient by asking their Twin, Decide by assigning their Twin, and Act by approving the Twin's proposed action.
Proposals accumulate as a badge count on "Approve Twin's Action." The user pulls them when ready. No pop-ups, no chat window, no "How can I help?" greeting.
Constitutional Principles
- The human product is L1.
- Every projection is a projection.
- No projection owns canonical truth.
- Every projection consumes governed context.
- Context Lenses share one Adaptive Spine.
- Governance protects all projections equally.
- Applications remain replaceable; the Work remains consistent.
- The OODA Grammar (Store → Ask → Assign → Approve) is universal across every Workbench.
Doctrine Replacement
This replaces prior multi-Workbench product doctrine, separate Human/Twin/Governance/Knowledge/Domain Workbench product boundaries, and any frontend doctrine that treated role-specific surfaces as separate products.
The L1 Workbench canon is:
CANONICAL_PLATFORM_ARCHITECTURE.mdPRODUCT_ARCHITECTURE.mdWORKBENCH_DOCTRINE.md