Context Studio overview

Context Studio brings everything that shapes Orion's understanding into one place: your domains, nodes, metrics, and dimensions; the natural-language context (topics, domain skills, Orion memory); the knowledge graph; and the quality signals (accuracy evals, corrections, certifications). The headline number is the Context Health score, a single 0–100 reading of how ready Orion is to answer accurately.

This page covers what Context Studio is, how it is laid out, and how to read the Overview tab. The individual tabs are documented on their own pages.

Overview

Context Studio is a three-pane workspace:

• Left sidebar — navigation grouped into six labelled sections, plus a live Context Health score pinned to the bottom.
• Center — the active tab's content (KPI tiles, tables, editors, and so on).
• Right info panel — contextual help for whichever tab you are on (collapsible).

The sidebar header reads "Context Studio — Intelligence layer for your semantic model." Tabs that carry an outstanding count (for example Corrections or Cross-Domain) show a number badge, and an alert badge turns red when there is something to act on.

When to use it

Open Context Studio when you need to:

• Check Orion's readiness before relying on it for important questions — read the Context Health score and its three component bars.
• Manage the semantic context that Orion uses — review metrics, domains, topics, domain skills, and Orion memory.
• Investigate accuracy — run accuracy evals, review corrections users have submitted, and resolve cross-domain metric conflicts.
• Govern the model — inspect lineage, run impact analysis, review usage, and manage certifications.

If you are still defining the semantic model itself (writing CML, adding reveals), that work happens in Forge. Context Studio is where you observe and govern the intelligence layer that sits over it.

Concepts

How Context Health is calculated

The score is the sum of three components, each capped at its own maximum:

In the sidebar the score shows as NN / 100 labelled "Orion confidence score," with a coloured bar for each component (green = certified reveals, blue = eval pass rate, purple = corrections). The number is colour-coded: green at 80 and above, amber from 60 to 79, red below 60.

Getting started

• A role of Weaver or Luminary. Tracer and Lens cannot open Context Studio.
• A semantic model with at least one domain and some reveals, so the health components have data to measure.

From the main navigation, select Context Studio (the brain icon). It opens on the Overview tab. Use the left sidebar to move between tabs; the score at the bottom of the sidebar updates as your context changes.

Step-by-step

Examples

The Overview tab is data-driven from two backend endpoints. The first returns the Context Health score and its breakdown:

{
  "score": 84,
  "components": {
    "certifiedReveals": { "score": 32, "certified": 12, "total": 15 },
    "evalPassRate":     { "score": 36, "passRate": 90, "runsExist": true },
    "corrections":      { "score": 16, "resolved": 8, "total": 10 }
  }
}

Here the score of 84 (green) breaks down as 32/40 certified reveals + 36/40 eval pass rate + 16/20 corrections. To push it higher you would certify the remaining 3 reveals and resolve the 2 outstanding corrections.

The second endpoint feeds the AI Hub stat row on Overview:

{
  "accuracy": 90,
  "questions_answered": 142,
  "questions_total": 156,
  "questions_failed": 7,
  "corrections_pending": 2,
  "corrections_applied": 8,
  "topics_total": 24,
  "topics_certified": 19,
  "certifications": 12,
  "eval_pass_rate": 90
}

The six navigation groups

The sidebar organises 23 tabs into six groups (the AI Operations tab appears only for Luminaries, so Weavers see 23):

Best practices

• Treat 80+ as the target. Below 60 the score is red and Orion is more likely to answer inaccurately; aim to keep it green.
• Certify your most-used reveals first. Certified reveals are worth up to 40 points and Orion prefers them, so they move the score and accuracy together.
• Keep the corrections queue at zero. An empty queue scores the full 20 points; let it grow and the score drops while real accuracy problems pile up.
• Re-run evals after meaningful changes. The score reflects the latest eval run, so a stale run can mask the impact of recent model changes.
• Watch the badge counts. The red badges on Corrections and Cross-Domain are your shortlist of things actively hurting Orion.

Tips

Common mistakes

Troubleshooting

Related pages

• Forge — where you build and certify the semantic model the score measures
• Orion — the AI analyst that Context Studio governs
• Workspace Intelligence Score — the separate, broader workspace-maturity measure
• Semantic Health — the Quality-group tab behind the eval and certification signals
• AI Hub — the activity view that feeds the Overview stat tiles
• Corrections — review and resolve the feedback that drives the corrections component