# Evolution 85 — Commercialization Plan

**FIRME AI CORPORATION · CAIN-42 · Evolution 85** — branch `evolution-85`, baseline Evolution 84 (`2c419a042`). Development result. Not deployed. Not released. Built by the three-person team.

## Positioning

**CAIN-42 — the governor of vertical AGI.** Whatever intelligence a company deploys in a vertical, CAIN-42 is the
layer that decides what it may do, enforces it at the boundary, can stop it, and proves what happened.

| Product | Role |
|---|---|
| CAIN-42 | governance infrastructure for vertical AI workforces: identity, delegation, policy, kill switch, receipts |
| CAIN Studio | hosted governance and orchestration |
| MCPGate | self-hosted enforcement boundary |
| ClawX | workforce evidence and verification |
| Vertical Workforces | narrow, governed delivery teams on the common core |

Wording rules for public pages, tied to gates:

| Statement | Allowed when |
|---|---|
| "The governor of vertical AGI" (a role) | now, with the measured status beside it |
| "The governed path to vertical AGI" | now |
| "Measured on frontier models" | a remote-provider result exists (credential + spend approval) |
| "Independently witnessed" | an independently administered witness cosigns (G22, G37) |
| "Closest available to vertical AGI", "first mover" | not supported by evidence; "first mover" is retired by the 2026-11-10 owner-approved rule |

## First customer segment

Engineering teams of 5–50 people that ship AI agents or MCP servers and have a buyer, auditor or security reviewer
asking how the agent is controlled. Buyer roles: head of engineering, security lead, founder.

## First three sellable workflows

| Workflow | Pain | Deliverable | Price hypothesis |
|---|---|---|---|
| Agent and MCP security review | "is this server safe to connect?" | reviewed tool list, findings, signed evidence | $500 assessment (up to 20 tools); $2,500–$7,500 full |
| Fabrication audit | "does this code compute what it claims?" | graded modules, signed evidence | $1,500 / $3,500 / $6,000 |
| Signed evidence pack | "our buyer wants proof of control" | receipts, offline verifier, audit export | $2,000–$10,000 |

Prices are hypotheses, not validated rates. The first target is one paid $500 assessment, then one paid implementation.

## Competing approaches

Agent gateways and MCP firewalls (policy at the proxy), observability platforms (traces after the fact), LLM-judge
evaluators (non-deterministic verdicts). The difference offered here: the decision, the enforcement, the stop and the
record are bound into one receipt that a third party can verify offline without CAIN code.

## Deployment, integration, support, security

Deployment: hosted (CAIN Studio) or self-hosted (MCPGate). Integration: the customer routes tool calls through the
boundary and supplies an identity per agent. Support: email, fixed-scope engagements. Security requirements a buyer
will ask for and the present answer: no third-party audit, no SOC 2, keys are files on the host, no hardware root.

## Pilot acceptance criteria (proposal)

1. Every tool call in the pilot scope passes the boundary; a call outside policy is refused with a signed denial.
2. The customer's own engineer verifies ten receipts offline.
3. The kill switch stops the pilot agent within an agreed time, demonstrated live.
4. Findings are reproduced by the customer on their own machine.

## Materials prepared

Technical demonstration (`scripts/evolution85/run_verticals.py`, `run_governed.py`), product explanation (this
document), security architecture (`THREAT_MODEL.md`, `KILL_SWITCH_ARCHITECTURE.md`), evidence demonstration (the
signed bundle and `verify_e85_offline.py`), pilot proposal and pricing (above), implementation checklist and
onboarding (below).

**Onboarding checklist:** agent inventory → one identity per agent → delegation limits → policy with approval gates →
route calls through the boundary → kill-switch drill → first receipts verified by the customer → weekly evidence export.

## Status

No customer has seen this demonstration. No payment resulted from this evolution. A technical milestone is not market
validation, and no revenue is reported here.
