Part III is for the reader who has to decide something: whether to engage, when, and on what evidence. It opens with the roadmap, because the honest answer to where the platform is going is also the honest answer to how much of it exists. The roadmap is stated in three bands rather than on a calendar. The active-complete band is what is built and live; refinement continues, but a reader can open it today. The in-process band is what is specified and under construction, with parts already live where the text says so. The future band is direction: themes we intend to pursue, with no schedule, no price and no promise attached, and each marked as such. Every entry is a theme, not a ticket. The record we plan from carries far more detail than this chapter, and that detail stays internal. A public roadmap that names dates will be held to them. A roadmap that names internal work items will be misread as a feature list.
#Active-complete.
#The agent front door and discovery. An agent can find OntoRamp, connect to it and consult governance before it acts. It does so through one discoverable front door on an open protocol, with a free plugin for the decision tools and standalone registry listings for the capabilities an agent most often needs on its own. The pieces that make that a round trip are in place: discovery descriptors an agent or an answer engine can read, including how to authenticate; intake of an assessment request from an agent; and readback of the result in a form the agent can act on. The commercial step, from the free tier to a paid plugin, is taken on the pricing page.
#The assessment and certification ladder. The Governance Ladder is in place end to end: the self-serve evaluation, the Decision Integrity Scan that leads into a full assessment, Enterprise Governance Assessment across all seven governance domains, the standalone Agentic Readiness Certification at L4 Resonant, and the Readiness Certificate that anyone can verify. One engine issues every reading.
#The Architecture Pack and the open exchange format. An organization's architecture is modelled as a standards-based model and can leave the platform the same way: the pack is delivered with a machine-readable export in the open ArchiMate exchange format. The exemplar that calibrates the engine is published in the same format, so the shape of the model can be inspected before any engagement and without anyone's architecture on display.
#Governance gate evaluation and the physics engine. The engine that computes a rung from evidence is live behind every door. The evaluation of a proposed decision, the pass, warn or block verdict of chapter seven, is live through the free plugin. Changes to the engine and its gates go through the discipline chapter nine describes: specified, evidence-linked and reviewed before they ship.
#Answer-engine readiness and the public-claims discipline. Every figure on the OntoRamp site comes from one public-claims source, and the site is checked against it before it ships. The same source answers the machine-readable questions an answer engine or an agent asks about what OntoRamp is and how to connect. This document is a product of the same discipline: its figures are named constants filled from that source at the date in its header.
#Knowledge-base integrity and customer isolation. The posture stated in chapter six is live. Customer evidence is scoped to the customer. Anonymous access to scoped data is denied, and that denial is tested by the platform itself. Every external door applies a classification ceiling, so that classification governs what leaves the platform regardless of the door.
#In process.
#Web-native presentation and governance reports. The readings the platform produces are moving from documents into surfaces. The Governance Assessment Report and the Transformation Roadmap become navigable, linked presentations in which each finding opens onto its evidence, rather than files that are read once and filed.
#Governed multi-agent execution. Chapter eight described it. A bounded group of agents is put to work on a pre-defined governance scenario within a cost limit the customer sets, and returns a decision packet and a governance attestation for a person to decide on. Parts of it run inside our own operation; the customer-facing form is specified.
#The governance agent and the continuous mode. The agent that keeps a governance picture current from the systems an organization already runs is built and operating in a pilot, in observation mode. Enforcement is built and switched off, and the agent's limits are applied by people today rather than by the platform. The continuous mode of the assessment, described in chapter seven, is what this agent becomes once observation, evaluation and enforcement are turned on in that order. The agent is designed as a dedicated instance for each customer, set up for that customer alone rather than as one service shared across customers, and it is built that way today in a pilot. The approved direction is for that instance to run inside the customer's own environment. Governed multi-agent execution follows the same model: it is designed to start hosted by us, and the approved direction is for it to move inside the customer's environment.
#Knowledge-graph quality, entity resolution and retrieval. The corpus is being brought to a defined mature state, and the work is measured against it. In that state each concept is held once with all of its evidence together, every answer is complete and traceable, the same question gets the same answer today and tomorrow, and every door answers with the same quality for everything it is permitted to answer. Duplicate concepts are being resolved with provenance kept through every repair; retrieval quality is measured; and every answer that leaves the platform is being made to carry the evidence it rests on.
#Executable-opportunity derivation. The capability described in chapter eight, by which the operating system tells an organization what it could do next with it and has not yet done, is specified and not yet built. It will be proven in our own operation before it is offered.
#Autonomy in bands. The discipline of chapter nine grants autonomy in bands, by a person's explicit act, and widens it only behind a recorded human decision. It is in force in part of our own operation; bringing all of our own automation under its terms is in process.
#Operational self-governance. The tooling that governs our own work is under continuous construction. None of it is an offering. All of it is the discipline of chapter nine made mechanical, and it is listed here because a company that sells governance should say what it is doing about its own.
#Hardening customer isolation. The posture of chapter six holds today, and strengthening it further is part of the current work.
#Governance in the deal. The first of the acquirer's three questions of chapter eight, the target's posture computed from its evidence, is answered today: the due-diligence use case on the OntoRamp site describes how the assessment reads a target from its governance material and returns a diligence summary, and the Decision Integrity Scan is the directional read that leads into a full assessment. The valuation computation that connects posture to what an enterprise is worth exists within our own operation and is not yet offered as a surface.
#Future.
#Enterprise deployment of the front door. An enterprise's own front door, isolated from the shared one; self-serve provisioning of an organization's own governed graph; a recorded sign-off step behind the highest certifications; and stronger isolation of one customer's knowledge from another's. The customer-environment deployment of the governance agent and of governed multi-agent execution, stated with the agent theme above, is the same theme seen from the agents' side.
#Reaching the agent where it already works. Connection guides for more of the agent platforms teams already use; channel integrations, so that a governance reading arrives in the tools where decisions are made rather than on a page someone has to visit; and a third mode of the live agent demo that shows the operating system in a scenario shaped like a customer's.
#Deliverables for the board. Board-ready packaging of the evaluation and the decision record; an export of the decision record as an audit artifact; an evidence map that links a reading to its sources; and a counter that shows what unmade decisions cost over time.
#The combined entity and the price. A due-diligence pack for acquirers and portfolio reviews; a projection of the combined entity's posture before close; and the translation of a change in posture into a change in what the enterprise is worth.
#Policy made public. The policy that governs what we publish, the rules that govern how we name what we publish, and the policy that governs isolation between customers, each published in its own right rather than applied silently.
#Where autonomy sits in each band. The autonomy discipline runs through the whole roadmap, so it is worth stating once, at band level, how much of our own work agents do and under whose authority.
In the active-complete band, people decide, and standing automation that a person has approved runs on its own. In the in-process band, agents work on tasks a person opens and scopes, which is how most of the platform is now built, and running several agents at once is proven in part. Bringing every piece of our own automation under the discipline's terms is itself in process. In the future band sit the wider forms of autonomy the discipline names but does not enable; each is held shut until a person approves the specific widening.
Autonomy is also bounded by a safety-pattern discipline that runs across every band. Our own controls are designed against a versioned set of named safety patterns, each tied to the harm it exists to prevent. A change to production is held until a recorded human authorization is attached to it. The patterns themselves are internal; what a customer can rely on is that the discipline exists, is versioned, and shapes the controls that are built.