Unclassified

21 — Platform Story

Status: CANONICAL · Commercial OS
Owner: Product Marketing
Last verified: 2026-07-21
Audience note: "platform" is internal vocabulary — customer-facing assets say "the infrastructure beneath the workspace." Never use "platform" as our category (18 §6).

Purpose

How to tell the story of everything beneath the workbench — for technical evaluators, analysts, and investors who need to know the product isn't a thin UI over a chatbot.

Target audience (of the story): CIOs, architects, security teams, technical champions, analysts, technical investors.

1. The Story

The workspace is deliberately calm; the infrastructure beneath it is deliberately serious. Six responsibilities, cleanly separated (platform doctrine):

                    ┌──────────────────────────────┐
                    │   Shared Workbenches (product)│  what the user sees
                    └──────────────┬───────────────┘
                          ContinuityBundle ↑ (scoped context, evidence, authority)
┌───────────────────────────────────────────────────────────────┐
│  PLATFORM — continuity infrastructure                          │
│  · Adaptive Spine: canonical state, identity, relationships    │
│  · Evidence & memory: provenance, freshness, TruthLayer        │
│  · Policy & governance: authority, approvals, audit            │
│  · Signals & proposals: observation, grounded triggers         │
│  · Projection Engine: state → role-composed workbenches        │
│  · Governed execution: approved actions across connected tools │
└──────────────┬────────────────────────────────────────────────┘
               │ IW-Continuity-Bridge (bounded access for approved entry points)
        70+ connected tools (CRM, support, billing, comms, …)

The design argument (tell it as a choice, not a feature list): most AI products put a model between the user and their tools and hope. We put governed state between them — and let models visit. Deterministic systems (pipelines, normalization, sync, validation) do what deterministic systems do best; adaptive systems (the Twin's reasoning) do what they do best; and neither is allowed to impersonate the other. A model output is never truth; UI state is never truth; canonical work state is truth, and everything that touches it is identity-bound, tenant-scoped, policy-checked, and audited.

2. The Six Laws (quote to technical audiences — from platform doctrine)

Canonical work state is authoritative; UI state and model output are not.
Every cross-boundary request is identity-bound, tenant-scoped, policy-checked, time-bounded, and auditable.
Context is projected for a stated purpose — no participant receives unrestricted tenant state.
Evidence, decisions, proposals, and actions preserve provenance and authority separately.
Execution is governed: preparation, recommendation, and approval are distinct stages.
The platform fails closed when it cannot establish authority, boundary, or policy.

3. Why This Matters Commercially

Platform property	Buyer translation
Fail-closed boundaries	"A prompt can't talk its way past policy" — the CISO's sentence
Provider-agnostic models	"Swap AI vendors by configuration; nothing resets" — no AI lock-in
Provenance on every field	"Board decks with sources, not vibes"
Deterministic pipelines + adaptive reasoning separated	"The sync doesn't hallucinate; the reasoning doesn't corrupt the data"
Bridge with bounded projections	"Your other tools can use the truth — under the same governance" (private alpha; don't oversell — 42 Security FAQ)

4. Telling Rules

Technical audiences get this story after the workbench story, never instead of it — the platform exists to make the calm surface trustworthy.
Never expose internal topology as product ("15 domains, 31 services" is repo talk, not sales talk).
The one diagram above is the only platform diagram in external use — one picture, told consistently, beats five accurate ones.
Proof beats: live provenance panel · approval + audit trail · model-swap by config · a failed-closed request shown honestly.

KPIs: technical-evaluation pass rate; security-review cycle time (42); % of technical stakeholders who can restate "AI never owns truth" unprompted after the session.