> ## Documentation Index > Fetch the complete documentation index at: https://docs.inviolet.ai/llms.txt > Use this file to discover all available pages before exploring further. # The maturity model > Seven tenets, five levels — how teams climb from observation to structural enforcement, one principle per level, scored against NIST SP 800-207. Inviolet scores your program against the **seven tenets of [NIST SP 800-207](https://csrc.nist.gov/pubs/sp/800/207/final)** (Zero Trust Architecture), each measured 0–100. A tenet counts as **in force** once its coverage crosses **θ = 60**. Your position on the **five-level ladder** is then the highest rung where *every tenet that rung requires* is in force: **L1 Observe → L2 Test → L3 Enforce → L4 Isolate → L5 Surround**. The point of the ladder is not the number — it's *which of the [five principles](/concepts/principles) is in force at each rung*. The levels and principles are 1:1: you climb by putting the next principle into practice, which lights up the tenets that level needs. ## The seven tenets Each tenet re-expresses one of NIST 800-207's zero-trust tenets in terms of what Inviolet can actually measure. They're listed in the order they come into force as you climb. | Tenet | What it measures | | --------------------------- | ------------------------------------------------------------------------------------------------------------------------------ | | **T7 — Intent capture** | Agent prompts flow through Inviolet as typed, human-owned intentions — the signal identity can't see | | **T5 — Monitoring** | The health of your approval loop and the breadth of shadow evaluation and SIEM coverage | | **T4 — Dynamic policy** | How much access is decided by dynamic policy on the declared intention and its conditions, not static grants | | **T3 — Per-session access** | Whether access is granted per session via short-lived, brokered leases rather than standing connections | | **T6 — Strict enforcement** | How much authorization is dynamically and strictly enforced *before* access, live rather than in shadow | | **T2 — Secure comms** | Whether credentials are brokered from a vault and traffic is secured regardless of network location | | **T1 — Resource coverage** | Whether *every* path — every data source, every LLM host, direct egress — is treated as a protected resource and proven closed | Below its eligibility line a tenet reads **N/A** rather than zero — it isn't scored or counted toward your level until the setup that makes it measurable is in place, so the onboarding view can tell you exactly what to do to start scoring it. ## The five levels Each level puts exactly one principle into force. They come in order — each rung assumes the ones below it are already in place. | Level | Principle | What goes live | Tenets required | | ----------------- | ------------------------------------ | -------------------------------------------------------------------------------------- | ---------------------- | | **L1 — Observe** | Learn the intent behind every prompt | Gateway installed; intentions declared and captured; query patterns learned | T5, T7 | | **L2 — Test** | Test before enforcing | Policies authored, evaluated in shadow mode against live traffic | T4, T5, T7 | | **L3 — Enforce** | Least access as the system learns | Live column-grain denials; gray zone routed to humans; approvals adopted as policy | T3, T4, T5, T6, T7 | | **L4 — Isolate** | Agents never touch credentials | Vault bound, OAuth broker active; every dispense a short-lived, intention-scoped lease | T2, T3, T4, T5, T6, T7 | | **L5 — Surround** | Impossible to bypass | All hosts covered, SWG egress, fingerprinting, bypass tests, SIEM, attestation | all seven | ### Level 1 — Observe Every prompt declares an intention and Inviolet captures it — you can see what your agents claim, and the system learns from every request. Every action still runs; nothing is blocked yet. This puts **Principle 1 — learn the intent behind every prompt** into force (tenets T7 *intent capture* and T5 *monitoring*). ### Level 2 — Test You author intent cards and policies, and they evaluate against live traffic **in shadow mode**. The dashboard shows what *would* have been denied, routed to approval, or allowed — before anything can break. This is **Principle 2 — test before enforcing**, which brings **T4 — dynamic policy** into force. ### Level 3 — Enforce You promote policies from shadow to live. Decisions now block real queries at the column grain, the gray zone routes to a human, and approved decisions are adopted as standing policy so the queue shrinks over time. This is **Principle 3 — least access as the system learns**, lighting up **T3 — per-session access** and **T6 — strict enforcement**. ### Level 4 — Isolate Credentials move off the agent: the vault is bound, the OAuth broker holds provider tokens, and every dispense becomes a short-lived, intention-scoped lease. The real key never reaches the agent. This is **Principle 4 — agents never touch credentials**, which brings **T2 — secure comms** into force. ### Level 5 — Surround Every remaining path is closed and proven: SWG egress catches direct calls, all LLM hosts are covered, traffic is fingerprinted, a bypass-test suite runs in CI, and SIEM detections watch for any attempt to route around the gateway — all attestation-grade. This is **Principle 5 — impossible to bypass**, which lights up the last tenet, **T1 — resource coverage**. Per the architecture, **only Ultraviolet-tier customers can reach Level 5.** ## How your level is scored Scoring is **cumulative**, not a weighted average you can game. Each tenet earns a 0–100 coverage measure; a tenet is **in force** once its coverage crosses **θ = 60**. Your level is the highest rung `L` for which *every* tenet in that level's required set is in force — so a single lagging tenet holds you at the rung below until you lift it, and you can't paper over it by maxing out another. Advancing a level therefore takes real coverage on the right tenets, backed by that level's **adoption checklist** — concrete, checkable signals like "1+ active policy denying something," "credential vault bound," or "zero standing credentials" — that make each tenet measurable in the first place. So the fastest way up is to lift the specific tenet that's holding your rung — not to over-invest in one you've already carried past θ. ## Reading the dashboard's level badge The maturity band on your dashboard shows your current level, an **N-of-7 tenets in force** headline, the per-tenet coverage radar, and the next level's outstanding tenets and checklist items. Click any item to jump to the intent card or integration that would unlock it. ## Where to start Most teams enter below **L1** (gateway installed, no policy, no vault) and climb: 1. Instrument a runtime and declare your first intention — makes **T7** and **T4** measurable and puts Principle 1 to work 2. Assign approvers or run a policy in shadow — starts scoring **T5** and crosses into **L2 / Test** 3. Promote a policy out of shadow — lifts **T6** and **T3** and crosses into **L3 / Enforce** 4. Bind the vault and broker your first provider — brings **T2** into force, starting **L4 / Isolate** 5. Deploy egress + bypass coverage — the last tenet, **T1**, and the door to **L5 / Surround** ## Read next * **[The five principles](/concepts/principles)** — the doctrine each level puts into force * **[Generalize from observation](/guides/generalize-from-observation)** — graduating an intention from L1 to L3 * **[Tier comparison](/tiers/overview)** — which Inviolet tier unlocks which principle