> ## Documentation Index > Fetch the complete documentation index at: https://docs.flowx.ai/llms.txt > Use this file to discover all available pages before exploring further. # Agentic Readiness Index > Score each AI workflow's readiness for autonomy across controls, audit-trace completeness, reversal rate and ROI, with blocking gates and evidence packs. > [!NOTE] > > **Observatory availability.** In FlowX.AI 6.0, Observatory runs on environments operated by FlowX. It isn't included in client deployments in this release. The Agentic Readiness Index (ARI) answers one question: is this workflow ready for the level of autonomy it runs at? Open it from **Governance › Readiness Index**. Each workflow gets a deterministic, audit-defensible score from 0 to 100 across four dimensions (Autonomy, Audit-Trace, Reversal Rate and ROI), a Green, Amber or Red band, and the evidence behind the number. *** ## What does the page show? * **Org readiness**: the composite readiness for the selected organization. It's also the **Agentic readiness** card on the [Governance Hub](https://docs.flowx.ai/6.x/observatory/governance/overview.md#governance-hub), weighted by exposure across scored workflows. * **Workflow readiness scores**: one row per workflow with its composite score, band and any **Blocking gate** that fired. Use **Recompute** on a row or **Recompute all** to rescore. * **Scoring models**: the published scoring models available to the org. Open a workflow's drill-down to see the **Dimension breakdown**, its **Declared autonomy**, the state of each control, the **Score history** and the publishing actions. A score marked **Provisional (insufficient data)** left out at least one dimension because there wasn't enough data for it in the scoring window. *** ## How is a workflow scored? ARI computes four dimension scores from 0 to 100, then combines them with the weights of the selected scoring model. Scoring is deterministic and uses no LLM: the same inputs and the same scoring model version always produce the same score. | Dimension | What it measures | | - | - | | **Autonomy** | How well the workflow's controls fit its declared autonomy level. Each required control counts as present (full credit), partial (half credit) or absent (no credit). | | **Audit-Trace** | Audit-trace completeness: the share of executions in the window that have a complete audit trace. | | **Reversal Rate** | The share of autonomous actions that were reversed, mapped to a score: 0.5% or less scores 100, up to 2% scores 85, up to 5% scores 65, up to 10% scores 40, and anything higher scores 15. | | **ROI** | Realized value against target, taken from the [ROI Hub](https://docs.flowx.ai/6.x/observatory/roi/overview.md#roi-hub): 125% or more of target scores 100, at least 100% scores 80, at least 75% scores 60, at least 50% scores 40, and less scores 15. | > [!NOTE] > > The reversal rate is an estimate. It's derived from discarded and overridden actions, not from a dedicated reversal event. A dimension with no data in the window (no executions, no autonomous actions, or not enough ROI data) is left out of the composite, and the score is marked provisional. ### Scoring models | Model | Profile | Autonomy | Audit-Trace | Reversal Rate | ROI | | - | - | - | - | - | - | | `regulated-v1` (default) | Control and auditability weighted heaviest | 30% | 30% | 25% | 15% | | `value-v1` | ROI weighted heaviest | 20% | 20% | 20% | 40% | A published scoring model never changes. Any change to weights or thresholds ships as a new model version, so every score stays reproducible from its inputs and its model version. ### Bands | Band | Composite score | | - | - | | **Green** | 85–100 | | **Amber** | 65–84 | | **Red** | 0–64 | *** ## What are blocking gates? A blocking gate caps the band at **Red** whatever the composite score, and the workflow row shows which gate fired. Each gate is checked on its own: | Gate | Fires when | | - | - | | High-risk, low audit trace | A high-risk workflow has audit-trace completeness below 90%. | | Autonomous without a kill switch | The declared autonomy is L3 or higher and no kill switch is present. | | High-risk L4/L5 without reversibility | A high-risk workflow at L4 or L5 has no reversibility mechanism. | | High reversal on irreversible actions | More than 10% of autonomous irreversible actions were reversed. | A workflow counts as high risk when its risk classification is high, critical or unacceptable. *** ## How do autonomy levels and controls work? Each workflow carries a **Declared autonomy** level. A workflow nobody has classified shows **Default (never declared)**. To change it, select **Edit** in the drill-down, pick the **Autonomy tier** and enter a **Note**, which is required. | Level | Meaning | | - | - | | **L0 Advisory** | Agent suggests; human executes every action. | | **L1 Assisted** | Agent drafts or prepares; human approves each action before it fires. | | **L2 Supervised** | Agent executes low-risk actions autonomously; risk-tiered actions are human-gated. | | **L3 Conditional** | Agent executes most actions; human intervenes only on flagged exceptions. | | **L4 High** | Agent executes end-to-end; human review is post-hoc or sampled. | | **L5 Full** | No human in the loop. | The Autonomy dimension checks five control families: **Approval gate**, **Reversibility**, **Kill switch**, **Scoped authorization** and **Escalation path**. Higher autonomy requires more of them: | Level | Required controls | | - | - | | L0 | Scoped authorization, Escalation path | | L1 | Approval gate, Scoped authorization, Escalation path | | L2 | Approval gate, Reversibility, Scoped authorization, Escalation path | | L3 | Approval gate, Reversibility, Kill switch, Scoped authorization, Escalation path | | L4, L5 | Reversibility, Kill switch, Scoped authorization, Escalation path | High-risk workflows also require an approval gate, a kill switch and reversibility at every level. Set each control's state to **Present**, **Partial** or **Absent** in the drill-down, with an **Evidence URL** and a note. *** ## How do I share a score? From a workflow's drill-down: * **Evidence pack (PDF)** exports the score with its evidence for an auditor or risk committee. * **JSON** exports the same score as structured data. * **Publish score** sends the latest score and band to the org's webhooks subscribed to the `ari.score.published` event, with an optional note. Nothing is written to FlowX core. The drill-down then shows when the score was last published and by whom. *** ## Related resources > [card] [Governance](https://docs.flowx.ai/6.x/observatory/governance/overview.md) > > The Governance Hub and the rest of the controls layer. > [card] [Risk Hub](https://docs.flowx.ai/6.x/observatory/governance/risk-dashboard.md) > > Six-dimensional risk score per app. > [card] [ROI & value](https://docs.flowx.ai/6.x/observatory/roi/overview.md) > > The ROI Hub numbers that feed the ROI dimension. > [card] [EU AI Act](https://docs.flowx.ai/6.x/observatory/compliance/eu-ai-act.md) > > How Observatory maps to the EU AI Act. ## Related topics - [Governance](https://docs.flowx.ai/6.x/observatory/governance/overview.md) - [Build and change apps with the Developer Agentic Harness](https://docs.flowx.ai/6.x/ai-platform/developer-agentic-harness.md) - [Configure Elasticsearch indexing for process instances](https://docs.flowx.ai/6.x/setup-guides/flowx-engine-setup-guide/configuring-elasticsearch-indexing/elasticsearch-indexing.md) - [Indexing config guidelines](https://docs.flowx.ai/6.x/setup-guides/flowx-engine-setup-guide/configuring-elasticsearch-indexing/process-instance-indexing-config-guidelines.md) - [Builds](https://docs.flowx.ai/6.x/docs/projects/runtime/builds.md) This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.